A three-dimensional dynamic simulation method and system based on multi-modal weather data

Abstract

The present application relates to the technical field of three-dimensional weather simulation, and particularly relates to a three-dimensional dynamic simulation method and system based on multi-modal weather data. The method comprises the following steps: obtaining multi-modal meteorological observation data including ground observation station information, radar observation information and satellite observation information, and calculating the time difference value between the observation times of each data source; taking the ratio of the time difference value and the spatial distance between key observation points as the disturbance propagation speed; establishing a space-time correlation window based on the disturbance propagation speed, calculating the bias correlation coefficient between each observation source in the space-time correlation window, and judging the atmospheric state type through the bias correlation coefficient; and mapping the atmospheric state type into multi-source coordination parameters. Through the space-time correlation window, the present application can set a suitable range according to the propagation characteristics of the weather system, realize the observation data set approximation with physical consistency, and effectively avoid the misalignment problem of different observation source data in the time stamp or coverage range.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional weather simulation technology, and in particular to a three-dimensional dynamic simulation method and system based on multimodal weather data. Background Technology

[0002] Modern meteorological monitoring systems have formed a diversified observation network, including automatic weather stations, weather radars, and meteorological satellites. These different types of observation equipment each have their advantages in terms of spatial coverage, temporal resolution, observation elements, and measurement accuracy: ground-based observation stations provide high-precision point-like observation data, weather radars can detect precipitation and wind field structures in three-dimensional space, and meteorological satellites have the capability for large-scale continuous observation. Three-dimensional dynamic simulation based on multimodal weather data displays the evolution of weather phenomena in real time in three-dimensional space, including not only horizontal distribution but also vertical structure and temporal evolution.

[0003] Different observation sources exhibit inconsistent observation times due to factors such as equipment operating cycles, data transmission delays, and geographical differences. Current technologies typically treat this as a source of error that needs to be eliminated, employing methods such as time interpolation or forced synchronization. However, these methods often result in the loss of crucial meteorological dynamic information. Summary of the Invention

[0004] Therefore, it is necessary for the present invention to provide a three-dimensional dynamic simulation method and system based on multimodal weather data to solve at least one of the above-mentioned technical problems.

[0005] To achieve the above objectives, a three-dimensional dynamic simulation method based on multimodal weather data includes the following steps:

[0006] Step S1: Acquire multimodal meteorological observation data including ground observation station information, radar observation information and satellite observation information, and calculate the time difference between the observation times of each data source;

[0007] Step S2: Use the ratio of the time difference to the spatial distance between key observation points as the disturbance propagation speed;

[0008] Step S3: Establish a spatiotemporal correlation window based on the disturbance propagation speed, calculate the deviation correlation coefficient between each observation source within the spatiotemporal correlation window, determine the atmospheric state type through the deviation correlation coefficient, and map the atmospheric state type to multi-source coordination parameters;

[0009] Step S4: Calculate the spatial gradient consistency index of different observation sources within the same region. When there is a divergence in gradient direction or amplitude, it is marked as an observational anomaly region.

[0010] Step S5: Optimize the multimodal meteorological observation data with geometric complementary weights using multi-source coordination parameters and observational anomaly areas, generate a three-dimensional meteorological field, and achieve visualization output.

[0011] The present invention also provides a three-dimensional dynamic simulation system based on multimodal weather data, used to execute the above-described three-dimensional dynamic simulation method based on multimodal weather data, wherein the three-dimensional dynamic simulation system based on multimodal weather data includes:

[0012] The multi-source data acquisition module is used to acquire multimodal meteorological observation data, including information from ground observation stations, radar observations, and satellite observations, and to calculate the time difference between the observation times of each data source.

[0013] The propagation speed calculation module is used to take the ratio of the time difference to the spatial distance between key observation points as the perturbation propagation speed;

[0014] The state recognition weight module is used to establish a spatiotemporal correlation window based on the disturbance propagation speed, calculate the deviation correlation coefficient between each observation source within the spatiotemporal correlation window, determine the atmospheric state type through the deviation correlation coefficient, and map the atmospheric state type to multi-source coordination parameters.

[0015] The anomaly detection module is used to calculate the spatial gradient consistency index of different observation sources within the same region. When there is a divergence in gradient direction or amplitude, it is marked as an observation anomaly region.

[0016] The 3D dynamic reconstruction module is used to perform geometric complementary weight optimization on multimodal meteorological observation data using multi-source coordinated parameters and observational anomaly areas, and to generate a 3D meteorological field and achieve visualization output.

[0017] This invention significantly improves the accuracy and reliability of multimodal meteorological data fusion by adjusting the fusion weight coefficients based on observed anomaly regions. In actual observation, due to differences in sensing equipment, observation time, and spatial resolution among different observation sources, data inconsistencies or anomalies often occur in some areas. By reducing the weight of the corresponding observation source in anomaly regions, the system effectively weakens the negative impact of abnormal observation data on the fusion results, avoiding erroneous judgments and information distortion caused by anomalies in a single data source, thus ensuring the scientific validity and stability of the fused data. Simultaneously, maintaining the original weights in normal regions fully leverages the advantages of each observation source, achieving complementary strengths. This dynamic weight adjustment mechanism enhances the adaptability and robustness of multi-source data fusion, enabling the fusion model to adapt to complex and changing meteorological environments. Using the adjusted fusion weights to perform weighted fusion of multimodal meteorological observation data, the fused data comprehensively reflects the spatial and temporal information from multiple sources such as ground stations, radar, and satellites, covering multi-level characteristics of point, area, and volumetric observations, effectively compensating for the deficiencies of a single data source. Through weighted fusion, the spatial coverage and accuracy characteristics of data from different observation sources are reasonably balanced, improving the integrity and spatial continuity of meteorological field data. The fused data more accurately reflects the detailed features of atmospheric physical processes, contributing to improved accuracy in subsequent meteorological analysis and forecasting. A three-dimensional grid structure is constructed based on the spatial range of the spatiotemporal correlation window, and the fused meteorological data is interpolated onto the grid nodes, greatly improving the regularization and systematic nature of the data. Regularized three-dimensional meteorological field data not only facilitates computer processing and model input but also enhances the data's operability and application scope. Interpolation algorithms transform discrete observation data into a continuous three-dimensional field, achieving smooth spatial transitions and detailed representation, making the meteorological field data more scientific and observationally representative. This three-dimensional grid structure can cover complex terrain and multi-scale meteorological phenomena, laying a solid foundation for refined meteorological simulation and analysis. Overall, this step constructs a complete system for multimodal meteorological data fusion and 3D dynamic visualization. By dynamically adjusting the fusion weights, it effectively resists interference from abnormal data and ensures data quality and fusion accuracy. Through 3D grid construction and interpolation, it achieves the continuity and regularization of observation data, enhancing the scientific expression of the data. Through advanced volume rendering technology, it improves the display effect and analysis convenience of meteorological information. Attached Figure Description

[0018] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0019] Figure 1 This is a flowchart illustrating the steps of a three-dimensional dynamic simulation method based on multimodal weather data according to the present invention.

[0020] Figure 2 This is a block diagram of a three-dimensional dynamic simulation system based on multimodal weather data according to the present invention;

[0021] Figure 3 This is a schematic diagram of establishing a spatiotemporal correlation window based on the perturbation propagation speed in this invention;

[0022] Figure 4 This is a decision-making flowchart for determining atmospheric state type using the deviation correlation coefficient. Detailed Implementation

[0023] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0024] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0025] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0026] To achieve the above objectives, please refer to Figures 1 to 4 This invention provides a three-dimensional dynamic simulation method based on multimodal weather data, the method comprising the following steps:

[0027] Step S1: Acquire multimodal meteorological observation data including ground observation station information, radar observation information and satellite observation information, and calculate the time difference between the observation times of each data source;

[0028] Further, step S1 includes the following steps:

[0029] Simultaneously, it receives point meteorological data from ground-based automatic weather stations, spatial scan data from weather radar, and spatial scan data from meteorological satellites as the raw observation dataset;

[0030] The original observation dataset was converted into a standardized format to obtain multimodal meteorological observation data;

[0031] Calculate the time difference between different data sources for multimodal meteorological observation data within the same observation period. The time difference includes the time difference between ground station and satellite observations, the time difference between ground station and radar observations, and the time difference between radar and satellite observations.

[0032] In some embodiments, point meteorological data collected by ground-based automatic weather stations, multi-elevation-angle volume scan spatial scanning data acquired by weather radar, and high-resolution visible light / infrared remote sensing image data transmitted downlink from weather satellites are used. These three types of data differ significantly in acquisition methods, spatial resolution, and temporal resolution; therefore, they need to undergo format unification and standardization before entering the fusion calculation process. Specifically, the hourly or minute-by-minute records of temperature, humidity, air pressure, wind speed, and other elements from ground-based weather stations are constructed as observation point data with latitude and longitude markers; the radar reflectivity factor (Z), radial velocity (V), and other data from weather radar are resampled to a unified spatial grid structure and labeled with scan time and corresponding elevation angle information; the pixel brightness temperature values ​​or reflectivity data from weather satellite data are projected into a georeferenced format according to orbital parameters and observation time, constructing a two-dimensional layer composed of pixel location and timestamp. After the above preprocessing, a unified multimodal meteorological observation dataset is obtained, possessing the ability to describe spatiotemporal labels and physical elements. The temporal synchronization between the three types of observation sources within the same observation period is quantified. For this purpose, the following calculations are required:

[0033] (1) The time difference between ground station and satellite observation;

[0034] (2) The time difference between ground station and radar observation;

[0035] (3) The time difference between radar and satellite observation.

[0036] The specific implementation method is as follows: extract the radar scanning point closest to each ground automatic weather station from the meteorological radar scanning data and record the scanning timestamp of the scanning point; extract the satellite pixels covering the location of each ground station from the satellite remote sensing image and record its image acquisition timestamp; compare the above two timestamps with the observation timestamp of the ground station respectively to obtain the ground-radar time difference and the ground-satellite time difference.

[0037] Furthermore, a target area containing several observation points is selected, and the average observation time of all coverage points of the weather radar and the average observation time of all pixels of the satellite are calculated within this area. The difference between the two is the radar-satellite time difference for this area.

[0038] It should be noted that since radar scans are typically performed on a volume scan cycle of several minutes, and the frequency of satellite transits is also affected by orbit control, there is an inherent asynchrony in the observation times. To improve the accuracy of subsequent perturbation propagation velocity calculations, this step prioritizes selecting the combination of observation points with the smallest time difference for propagation velocity estimation, and ensures that all types of observation data are unified under a high-precision timestamp system (such as UTC) to avoid value distortion caused by time zone conversion or time format errors.

[0039] Furthermore, the calculation of the time difference between multimodal meteorological observation data from different data sources within the same observation period includes:

[0040] Extract the effective scan point closest to the spatial location of the target ground station from the spatial scan data of the weather radar;

[0041] Compare the actual observation timestamp of the target ground station with the scanning timestamp of the effective scanning point, and calculate the absolute difference, which is recorded as the ground station-radar time difference of the target ground station.

[0042] Based on the pixel unit information corresponding to the geographical location of the target ground station in the spatial scanning data of meteorological satellites, the precise observation timestamp of the pixel unit is read to obtain the pixel timestamp;

[0043] The pixel timestamp is compared with the actual observation timestamp of the target ground station, and the absolute time interval between the two timestamps is taken as the ground station-satellite time difference of the target ground station.

[0044] The radar-satellite time difference of the target area is obtained by calculating the time difference between meteorological radar and meteorological satellite observation data based on the target area range.

[0045] The time difference between ground station and radar, the time difference between ground station and satellite, and the time difference between radar and satellite are combined into a single time difference value.

[0046] In some embodiments, using a target automatic weather station as a reference, the nearest valid radar scan point to the weather radar's spatial scan data is retrieved from its known latitude and longitude coordinates. A valid scan point typically refers to a location with reflectivity or velocity observation data within a vertical height range, with data from the lowest elevation angle section being preferred to reduce altitude interference. After locating the nearest point, the timestamp information of that scan point (i.e., the specific time of the radar scan at that location) is extracted and compared with the observation timestamp recorded by the target ground station. The absolute time difference between the two is calculated to obtain the ground station-radar time difference for that station. This time difference reflects the degree of time offset between the two types of sensors when observing the same spatial area.

[0047] From the spatial scan data of meteorological satellites, satellite pixel units corresponding to the geographic coordinates of the target ground station are extracted. These pixel units can be back-calculated using geographic projection to obtain their row and column numbers or actual coverage locations. In most polar-orbiting or geostationary meteorological satellites, each pixel typically carries an independent observation timestamp, representing the precise moment that location was scanned. After extracting this pixel timestamp, it is compared with the observation timestamp of the target ground station to calculate the absolute time interval, thus obtaining the ground station-satellite time difference.

[0048] To obtain a more regionally representative assessment of the time difference between radar and satellite data, it is necessary to construct a target area. This target area can be defined as a spatial grid cell containing several observation points, or the area covered by a known weather system. Within this area, the observation times of all relevant scan points are extracted from meteorological radar data, and their average value is calculated as the "regional radar average time." Simultaneously, the observation timestamps of all satellite pixels covering this area are extracted from meteorological satellite imagery, and their average value is calculated as the "regional satellite average time." The absolute time difference between these two values ​​constitutes the radar-satellite time difference for the target area.

[0049] After calculating the three time differences separately, they are combined to form a complete set of time difference values, which correspond to:

[0050] (1) Ground station-radar time difference;

[0051] (2) Ground station-satellite time difference;

[0052] (3) Radar-satellite time difference.

[0053] This time difference set will serve as an important input for subsequent calculations of disturbance propagation speed and construction of spatiotemporal correlation windows.

[0054] It should be noted that in practical applications, the data update intervals of different observation sources vary. Radar scanning frequencies may be 5-10 minutes per cycle, satellite data acquisition intervals may be as high as 15-30 minutes, while ground stations generally operate at a frequency of 1 minute or higher. To reduce the propagation of time errors, the observation pairs with the shortest time intervals are prioritized for paired calculations, and all timestamps are uniformly converted to Coordinated Universal Time (UTC) or GPS time system to avoid errors caused by inconsistencies in time zones or formats. Furthermore, in extreme weather conditions or areas with missing data, some observation points may not have corresponding radar or satellite observations. In such cases, these points can be skipped, or interpolation estimations can be performed using surrounding observations.

[0055] Furthermore, the calculation of the time difference between meteorological radar and meteorological satellite observation data based on the target area range includes:

[0056] Define a target area, which is specifically a spatial grid cell or a weather feature area;

[0057] The average time of all scan points constituting the observation of the target area is extracted from the spatial scan data of the weather radar and recorded as the regional radar average time.

[0058] The average time of all satellite pixels covering the target area is extracted from the spatial scanning data of meteorological satellites and recorded as the regional satellite average time.

[0059] The radar-satellite time difference of the target area is obtained by calculating the absolute difference between the regional radar average time and the regional satellite average time.

[0060] In some embodiments, to further enhance the registration accuracy between different observation sources in the time dimension, it is necessary to calculate the time difference between meteorological radar and meteorological satellite observation data based on the target area range. This step mainly targets meteorological radar and meteorological satellite, two data sources with spatial scanning characteristics, and extracts their observation time characteristics in a common area, thereby quantifying the degree of time asynchrony between the two when observing the same spatial target.

[0061] The spatial extent of the target area needs to be clearly defined, and this area can be flexibly determined based on specific operational needs or weather process characteristics. Common methods include: selecting one or more grids from a spatially divided grid of fixed resolution (e.g., a 1km × 1km grid) as the target area; or extracting the corresponding continuous pixel area based on the spatial boundary of a weather system (e.g., a convective cell, a frontal region, or a typhoon cloud system) as the target weather feature area. After determining the area boundary, all observation points corresponding to that area can be retrieved from radar and satellite data respectively.

[0062] Based on the geographic projection parameters of the radar volume scan data, determine which radar echo cells (e.g., reflectivity pixels) fall within the target area and collect the timestamps of these scan points. It's important to note that since radar generates 3D volume scan data using elevation scanning, the scan time at different elevation angles can vary by several seconds to tens of seconds. Therefore, this step typically selects the scan time at the same elevation angle (e.g., the lowest elevation angle) or uses scan points on a fixed elevation plane to maintain time consistency. The average timestamps of all valid scan points are then calculated to obtain the average radar time for the area.

[0063] Simultaneously, all satellite pixel units covering the target area are extracted from the spatial scan data of meteorological satellites. Satellite observations have a scan strip or scan line structure, and the observation time of different pixels changes line by line along the scan trajectory. Through geographic projection or orbital registration, the target area is mapped to pixel areas in the satellite image, and the observation timestamps of these pixels are extracted. Similarly, the timestamps of all covered pixels are averaged to obtain the satellite average time.

[0064] The difference between the regional radar average time and the regional satellite average time is calculated, and the absolute value is taken to obtain the radar-satellite time difference within the target area. This value reflects the time asynchrony between radar and satellite when observing the area, and is one of the key input parameters for subsequent disturbance propagation velocity estimation and spatiotemporal correlation window establishment.

[0065] It should be noted that due to differences in observation frequencies and scanning mechanisms between radar and satellites, the selected region size will affect the accuracy of time difference calculation. A larger region may include points with long scanning time spans, leading to an offset in the average value; while a region that is too small may result in unstable results due to insufficient observation points. Therefore, in practical applications, an appropriate region size should be selected based on the resolution and temporal distribution characteristics of the observation data, and the observation timestamps should be appropriately weighted or filtered to improve the representativeness and robustness of the regional time difference estimation.

[0066] Step S2: Use the ratio of the time difference to the spatial distance between key observation points as the disturbance propagation speed;

[0067] Further, step S2 includes:

[0068] Select the two different types of observation points with the smallest observation time difference from the multimodal meteorological observation data as key observation point pairs, including combinations of ground stations and radar stations, combinations of ground stations and satellite pixels, or combinations of radar stations and satellite pixels;

[0069] Calculate the actual spatial distance between key observation point pairs;

[0070] The propagation time is determined based on the time difference between key observation points. When there are multiple time differences, the time difference with the smallest value is selected as the propagation time.

[0071] The propagation speed is obtained by dividing the actual spatial distance between key observation point pairs by the propagation time.

[0072] In some embodiments, from standardized, timestamped multimodal meteorological observation data, two observation points of different types with the smallest observation time difference are selected to form a "key observation point pair". This point pair can come from one of the following three combinations: scanning points of a ground automatic weather station and a weather radar; pixels of a ground automatic weather station and a weather satellite; or scanning points of a weather radar and pixels of a weather satellite. In actual implementation, the system calculates the timestamp difference for all possible multi-source observation point pairs and prioritizes selecting the pair of heterogeneous observation points with the smallest absolute time difference as the key observation point pair to improve the proximity of the disturbance occurrence time, thereby enhancing the reliability of the propagation velocity estimation. The spatial distance between the key observation point pair is calculated. Here, the spatial distance refers to the actual distance between the two observation points in three-dimensional geographic space, which can be calculated as follows: for ground stations and satellite pixels, radar scanning points, etc., the spherical distance (great circle distance) between their latitude and longitude is calculated using the WGS84 ellipsoid model, and the three-dimensional spatial distance is estimated by combining the vertical height difference. For example, if a radar scanning point is at an altitude of 2 km and the ground station is on the ground, the horizontal distance and the vertical distance need to be combined to obtain a more accurate spatial path length. Combining the timestamp information of this observation point pair, the time span of disturbance propagation, i.e., the propagation time, is determined. When two observation points record the same or related meteorological disturbances (such as changes in the same wind field or cloud movement) at close intervals, the difference in their timestamps is the propagation time. In some cases, the system may detect multiple time differences (e.g., time alignment combinations of multiple radar elevation scan points and satellite pixels). In such cases, the time difference with the smallest value should be selected as the propagation time because a smaller time difference ensures that the observed event is the spatial transfer of the same disturbance event, rather than an error caused by the superposition of multiple events.

[0073] Dividing the calculated spatial distance by the propagation time yields the propagation speed of the disturbance. This speed value represents the average propagation rate of meteorological disturbances (such as sudden wind field changes, echo movement, cloud drift, etc.) under specific spatiotemporal conditions, and is expressed in meters per second (m / s) or kilometers per hour (km / h). It will serve as a key reference factor when constructing spatiotemporal correlation windows.

[0074] It should be noted that, to avoid velocity anomalies caused by observation errors or differences in resolution between different types of data, a physically reasonable threshold range (e.g., 5–300 m / s) can be set for propagation velocity in the actual system. Point pairs exceeding this range can be discarded or recalculated using suboptimal key observation point pairs. Furthermore, if the meteorological elements recorded by two observation points are fundamentally different (e.g., one is wind speed, the other brightness temperature), their correlation in disturbance response must be confirmed through feature matching algorithms before they can be used as valid key observation point pairs.

[0075] Of particular importance is that the actual spatial distance between the calculated key observation point pairs is specifically as follows:

[0076] For a combination of ground station and radar station, the spherical distance between the geographic coordinates of the ground station and the geographic coordinates of the radar station is calculated as the actual spatial distance.

[0077] For the combination of ground station and satellite pixels, the actual spatial distance is calculated as the distance between the geographical coordinates of the ground station and the position of the ground station at the satellite's nadir point at the time of satellite transit.

[0078] For the combination of radar station and satellite pixels, the distance between the geographic coordinates of the radar station and the coordinates of the geometric center point of the target area is calculated as the actual spatial distance, where the geometric center point of the target area is the center position of the overlapping part of the radar scanning range and the satellite observation area.

[0079] In some embodiments, a crucial step in accurately estimating the propagation speed of meteorological disturbances between different observation platforms is precisely calculating the actual spatial distance between key observation point pairs. Since multimodal observation data originates from different observation platforms—including automatic weather stations, weather radars, and weather satellites—and these platforms differ significantly in spatial distribution, observation angle, and observation area, appropriate distance calculation methods must be employed based on different types of key observation point combinations to ensure that the distance estimation has physical meaning and geographical accuracy. Specifically, this includes the following three scenarios:

[0080] For a combination of a ground station and a radar station, since both are located on the ground, or the radar station is at a relatively low tower height, their spatial location can be simplified to two points on a sphere. In this case, using the Great-circle Distance model, the spherical distance between the two stations is calculated using the Haversine or Vincenty formulas based on their latitude and longitude coordinates; this is the actual spatial distance of the combination. For example, if the coordinates of ground station A are (30.00°N, 114.00°E) and the coordinates of radar station B are (30.05°N, 114.10°E), then the spherical distance between them is approximately 11.2 km, which is the spatial distance of this key observation point pair.

[0081] When combining ground stations and satellite pixels, the unique characteristics of satellite observation must be considered. Since satellites are orbital platforms, their observations are conducted vertically or at an angle towards the ground, while the ground station is actually located at the corresponding geographical location of a pixel unit in the observed image. Therefore, to standardize calculations, this step uses the "nadir position of the ground station at the moment of satellite transit" as the geographical location of the corresponding satellite point. The nadir point refers to the position of the satellite directly facing the Earth's surface at a given moment; it is the projection of the satellite's trajectory onto the Earth's surface. By consulting satellite orbital data (such as TLE data) and imaging parameters, the latitude and longitude of the nadir point can be accurately obtained, and the spherical distance between it and the ground station's latitude and longitude can be calculated as the actual spatial distance between the ground station and the satellite pixel. For example, if the ground station is located at (30.00°N, 114.00°E), and the corresponding nadir point at that moment is (30.01°N, 113.98°E), then the distance is approximately 2.6 km.

[0082] For the combination of radar station and satellite pixels, since both observe a relatively wide area and have different observation angles, this step uses a regional center point as a spatial reference point to avoid error amplification. The specific method is as follows: First, determine the spatial overlap area between the current radar scanning range and the satellite observation image, i.e., the area where their observation coverage intersects; then, extract the latitude and longitude set of all valid radar echo pixels and satellite image pixels from this overlap area, and calculate its geometric center point (i.e., the center latitude and longitude coordinates of the overlap area); finally, use the spherical distance between this center point and the fixed geographical location of the radar station as the actual spatial distance. For example, if the radar station coordinates are (30.00°N, 114.00°E) and the center point of the overlap area is (30.10°N, 114.05°E), then the calculated spherical distance is 12.6 km.

[0083] It should be noted that there may be resolution differences or geometric distortions between satellite pixels and radar echo cells. Therefore, when determining the geometric center point, registered geographic coordinate data should be used, and invalid observations (such as cloud top obscuration or strongly attenuated radar points) should be removed to improve the physical representativeness of the center point. Furthermore, for geostationary orbit satellites, their nadir is basically fixed above the equator, and their spatial projection method differs from that of low Earth orbit polar orbit satellites. Therefore, distance calculations should be performed separately for each satellite type to avoid estimation errors caused by differences in altitude and scanning mode.

[0084] Step S3: Establish a spatiotemporal correlation window based on the disturbance propagation speed, calculate the deviation correlation coefficient between each observation source within the spatiotemporal correlation window, determine the atmospheric state type through the deviation correlation coefficient, and map the atmospheric state type to multi-source coordination parameters;

[0085] Furthermore, the step S3 of establishing a spatiotemporal correlation window based on the perturbation propagation speed includes:

[0086] The geometric midpoint of the key observation point pair is used as the spatial center coordinate, and the average observation time of the two observation points is used as the time center coordinate.

[0087] Divide the preset reference time constant by the disturbance propagation speed to obtain the time radius value, and multiply the disturbance propagation speed by the time radius value to obtain the spatial radius value;

[0088] Calculate the azimuth angle of the line connecting key observation point pairs as the main propagation direction angle;

[0089] When the disturbance propagation speed is greater than the preset speed threshold, it is determined to be a directional propagation phenomenon and an elliptical shape parameter is set. When the disturbance propagation speed is less than or equal to the preset speed threshold, it is determined to be a diffusion propagation phenomenon and a circular shape parameter is set, thus obtaining the geometric shape parameter.

[0090] Using the spatial center coordinates and the time center coordinates as the positioning center of the spatiotemporal correlation window, the spatial boundary range and time extension range of the window are determined by the spatial radius value and the time radius value, respectively. The circular or elliptical structure of the window is set according to the geometric shape parameters. When it is elliptical, the direction of the major axis of the ellipse is determined by the angle of the main propagation direction, thus obtaining the spatiotemporal correlation window.

[0091] In some embodiments, an atmospheric disturbance region, i.e., a spatiotemporal correlation window, is constructed to identify the spatiotemporal consistency among multi-source observation data. This window is then used to identify atmospheric state types, which are further mapped to weighting coefficients for multi-source data fusion. This step requires not only considering the spatiotemporal locations of key observation pairs but also integrating factors such as disturbance propagation speed and propagation patterns to form a physically meaningful dynamic correlation window. Specific implementation methods include the following:

[0092] The geometric midpoint of a key observation point pair is selected as the spatial center coordinate to ensure the symmetry of the window area with respect to the disturbance propagation path. This geometric midpoint is obtained by calculating the average latitude and longitude of the two points, representing the position of the propagation midline of the disturbance between the two current observation points. Simultaneously, the average observation time of the two observation points is used as the time center coordinate to determine the central reference point of the disturbance on the time axis. This combination constitutes the "anchor point" of the spatiotemporal window. Based on the disturbance propagation speed and a preset time reference constant (e.g., 300 seconds), the temporal and spatial scales of the window are calculated. Specifically, the time reference constant is divided by the disturbance propagation speed to obtain the time radius (i.e., the time coverage of the window); then, the disturbance propagation speed is multiplied by the time radius to obtain the spatial radius (i.e., the maximum distance the disturbance can propagate within this time period). For example, if the disturbance propagation speed is 50 m / s and the time reference constant is 300 seconds, then the time radius is 6 minutes and the spatial radius is 15 km. Calculate the azimuth angle of the line connecting key observation points, which is the geographic direction angle from one point to another (usually starting at 0° due north and increasing clockwise to 360°). This angle represents the main propagation direction of the disturbance. Spherical trigonometry or simplified planar trigonometry can be used, depending on the map projection method used.

[0093] The propagation pattern of the disturbance is determined based on its velocity, and the geometry of the window is accordingly. When the disturbance propagation velocity is greater than a certain set threshold (e.g., 30 m / s), the disturbance is considered to have a clear directionality and is classified as a "directional propagation phenomenon." In this case, an elliptical window with the main propagation direction as its major axis is generated. Conversely, when the disturbance velocity is less than or equal to the threshold, the disturbance is considered to exhibit non-directional expansion characteristics and is classified as a "diffusion propagation phenomenon." In this case, a circular window with the spatial radius as its radius is generated. The length of the major axis of the ellipse is set to the spatial radius, and the length of the minor axis can be set to a certain proportion of the major axis (e.g., 0.6–0.8 times) to reflect the asymmetry of propagation.

[0094] Finally, using the spatial center coordinates, temporal center coordinates, temporal radius, spatial radius, main propagation direction angle, and geometric parameters (ellipse or circle) obtained above, a three-dimensional spatiotemporal correlation window is generated. This window describes the range of influence of the disturbance within a specific time and space. This window will serve as the analysis area in subsequent steps, used to filter various observation data points falling within this range and calculate their relative deviation correlations, thereby identifying the atmospheric state of the disturbance (such as fronts, convection development, stratiform advection, etc.).

[0095] See Figure 3This figure details the technical solution for establishing a spatiotemporal correlation window based on the perturbation propagation speed. The figure uses a two-dimensional coordinate system, where the X-axis represents spatial distance (km) and the Y-axis represents time (minutes), with the origin located in the lower left corner. A key observation point pair is marked at the center of the coordinate system (coordinates 230, 300). This center point is determined by calculating the geometric midpoint of the key observation point pair as the spatial center coordinates, and the average observation time of the two observation points as the time center coordinates. The figure shows two different spatiotemporal correlation window formats:

[0096] The first type is the diffusion propagation window, represented by a solid circle. This circular window has a radius of 2 km and is suitable for disturbance propagation speeds less than or equal to 30 m / s. When a diffusion propagation phenomenon is determined, the system sets the circular shape parameter to indicate that the disturbance exhibits non-directional expansion characteristics.

[0097] The second type is the directional propagation window, represented by a dashed ellipse. This elliptical window has a major axis of 4 km and a minor axis of 2 km, with the major axis pointing northeast (displayed at a 45-degree angle in the figure). It is suitable for disturbance propagation speeds greater than 30 m / s. When a directional propagation phenomenon is determined, the system sets the ellipse shape parameters so that the direction of the ellipse's major axis is aligned with the main propagation direction.

[0098] The calculation relationship between the spatial radius and the time radius is indicated by dashed lines in the figure. The spatial radius is obtained by multiplying the disturbance propagation speed by the time radius, while the time radius is calculated by dividing a preset reference time constant (e.g., 300 seconds) by the disturbance propagation speed. In this embodiment, the spatial radius is 2 km, and the corresponding time radius is 2 minutes.

[0099] The main propagation direction is determined by calculating the azimuth angle of the line connecting the key observation point pairs, which is represented by a straight line with an arrow in the figure, pointing northeast. This azimuth angle is used to determine the direction of the major axis of the elliptical window.

[0100] By establishing the aforementioned spatiotemporal correlation window, the system can extract various types of observation data within this window and perform matching analysis with numerical weather prediction background field data, providing spatial and temporal constraints for subsequent deviation correlation coefficient calculations and atmospheric state type determination. This window design fully considers the physical propagation characteristics of meteorological disturbances, ensuring the scientific rigor and accuracy of spatiotemporal consistency analysis of multi-source observation data.

[0101] It should be noted that when the disturbance propagation speed is low, even if the calculated spatial radius is small, the system can set a minimum spatial radius threshold (such as 1km) to avoid the window range being insufficient to cover the effective observation area due to the speed being too low. In addition, for nonlinear trajectory disturbances (such as squall lines and vortices), multiple overlapping windows can be constructed through multiple sets of key point pairs to improve the stability and coverage of state recognition.

[0102] Furthermore, step S3, calculating the deviation correlation coefficient between each observation source within the spatiotemporal correlation window, includes:

[0103] Extract various types of observation data within the spatiotemporal correlation window and obtain the numerical forecast background field data corresponding to the window region;

[0104] Various observation data are matched with background field grid points in numerical weather prediction background field data, and the nearest neighbor method is used to determine the background field value corresponding to each observation point, forming an observation-background field matching dataset.

[0105] The deviation sequence between each observation source and the background field is calculated based on the observation-background field matching dataset. The deviation sequence includes the ground deviation sequence, radar deviation sequence and satellite deviation sequence.

[0106] The deviation correlation coefficient between observation sources is calculated by using the deviation value sequence. The deviation correlation coefficients between ground deviation sequence and radar deviation sequence, ground deviation sequence and satellite deviation sequence, and radar deviation sequence and satellite deviation sequence are calculated respectively.

[0107] In some embodiments, during a spring severe convective weather monitoring event, the system collects temperature and humidity observation data from ground automatic weather stations (intervals of 10 minutes), reflectivity products from weather radar (updated every 2 minutes), and cloud top brightness temperature data from geostationary meteorological satellites (updated every 15 minutes) within an established 50km × 50km spatiotemporal correlation window. Simultaneously, it retrieves the corresponding numerical weather prediction background field data (horizontal resolution 5km) from the WRF model. To achieve point-by-point matching between observations and the background field, the system employs the nearest neighbor method: for example, if a weather station is located at 120.25°E, 31.55°N, its nearest grid point in the background field grid is (120.20°, 31.60°), and the system directly uses the predicted temperature of that grid point as the background reference value for that station. Thus, each record includes the observation location, observation time, observed value, and corresponding background value. Subsequently, the system calculates the deviation value: Deviation = Observed value – Background value. For example, if the actual temperature measured at the weather station is 22.8℃ and the background temperature is 21.4℃, then the temperature deviation is +1.4℃; similarly, the deviation is calculated for each element such as radar reflectivity and satellite brightness temperature.

[0108] Finally, the Pearson correlation coefficient was used to calculate the linear correlation between the three types of biased sequences:

[0109] Pearson correlation coefficient between ground bias sequence and radar bias sequence;

[0110] Pearson correlation coefficient between ground-based bias sequences and satellite bias sequences;

[0111] Pearson correlation coefficient between radar bias sequence and satellite bias sequence.

[0112] These deviation correlation coefficients can be used to determine whether the same disturbance is observed among different sources in the region, whether the disturbance responds consistently to the background error, and whether the disturbance has systematic characteristics. Furthermore, the atmospheric state type of the current region (such as fronts, convective activity, convergence zones, background stationary states, etc.) can be classified according to the level of the deviation correlation coefficients, and mapped to different multi-source coordination parameters in subsequent steps accordingly.

[0113] It should be noted that, in order to avoid the impact of insufficient sample size on the stability of the Pearson correlation coefficient results, the system can set a minimum sample size threshold (such as more than 10 valid matching points) as a prerequisite for calculating the bias correlation coefficient; in addition, for "missing measurement points" or "obscured values" in radar or satellite, they need to be removed or interpolated before calculation to prevent the correlation results from being distorted.

[0114] Furthermore, step S3, which involves determining the atmospheric state type using the deviation correlation coefficient, includes:

[0115] Check the positive and negative signs of the ground-radar bias correlation coefficient, the ground-satellite bias correlation coefficient, and the radar-satellite bias correlation coefficient, and record the sign combination patterns of the three correlation coefficients;

[0116] When the symbol combination pattern shows that all three correlation coefficients are positive, it is determined to be a stratified stable atmospheric state. When the symbol combination pattern shows that the ground-radar deviation correlation coefficient is negative while the other two are positive, it is determined to be a convective development atmospheric state. When the symbol combination pattern shows two positive and one negative or one positive and two negative, it is determined to be a frontal transit atmospheric state, thus obtaining the basic state type.

[0117] The deviation correlation coefficient is divided into three levels: strong correlation, moderate correlation and weak correlation according to the preset threshold rule. When the deviation correlation coefficient is strong, it is judged as a stable state. When the intensity level is weak, it is judged as a transitional state. When the intensity level is large, it is judged as a composite state, thus obtaining the stability characteristics.

[0118] The basic state type is combined with the stability characteristics, and the atmospheric state type is output according to the preset corresponding rules.

[0119] In some embodiments, identifying atmospheric states using the deviation correlation coefficient of multimodal observation sources is a key step in achieving precise classification of complex atmospheric processes and reasonable allocation of fusion weights. This process, based on the sign (positive or negative) and numerical strength of the deviation correlation coefficients between three main observation source pairs, distinguishes different basic atmospheric state types and their stability characteristics through preset discrimination rules, ultimately outputting an accurate and physically meaningful atmospheric state type.

[0120] See Figure 4 This diagram details the complete decision-making process for determining atmospheric state type using the deviation correlation coefficient. The diagram employs a top-down tree structure, clearly presenting the entire judgment logic from sign checking to the final state type output.

[0121] Figure 4 The top section is the deviation correlation coefficient sign check node. This node is responsible for checking the sign of three key parameters: the ground-radar deviation correlation coefficient, the ground-satellite deviation correlation coefficient, and the radar-satellite deviation correlation coefficient, and recording the sign combination pattern of the three correlation coefficients. This step is the foundation of the entire atmospheric state type determination. In the second-level judgment branch, the system divides the judgment path into three main branches based on the different sign combination patterns: the left branch corresponds to the sign combination pattern of "all three are positive" (+,+,+), represented by a solid white circle in the figure. When the three deviation correlation coefficients of ground-radar, ground-satellite, and radar-satellite are all positive, it indicates that the deviations of the three types of observation data all show the same direction of change, the disturbance is layered and relatively stable, and the system determines it to be a layered stable atmospheric state. The middle branch corresponds to the sign combination pattern of "ground-radar is negative, others are positive" (-,+,+), represented by a dashed circle in the figure. When the ground-radar correlation coefficient is negative, while both the ground-satellite and radar-satellite correlation coefficients are positive, it indicates a significant reverse bias between ground and radar observations. This phenomenon typically corresponds to strong local convection activity, and the system classifies it as a convective-developing atmospheric state. The right branch corresponds to a mixed sign combination pattern of "two positive and one negative or one positive and two negative," including patterns such as (+,+,-) and (+,-,-), represented by circles filled with diagonal lines in the figure. When the correlation coefficient exhibits a complex mixed positive and negative pattern, it reflects the presence of frontal processes at the disturbance boundary, leading to a complex and interwoven distribution of the observational bias. The system classifies this as a frontal-transiting atmospheric state. The third layer is the basic state type output layer, corresponding to three different atmospheric state types:

[0122] Layered stable atmospheric state: characterized by uniform atmospheric structure, weak and continuous stability of disturbances, and highly consistent deviation distribution between the three types of observation sources and the background field.

[0123] Convective-developing atmospheric state: characterized by significant local convective activity, with ground-based and radar observations showing inverse deviations, while satellite observations maintain a positive correlation with other observation sources.

[0124] Frontal transit atmospheric conditions: characterized by unstable and rapidly changing atmospheric structure, complex and variable deviations between different observation sources, reflecting the impact of frontal systems.

[0125] Figure 4 The bottom section displays the final state type output node, which combines the basic state type with stability characteristics. The system categorizes the deviation correlation coefficients into three levels according to preset threshold rules: strong correlation (|r|≥0.7), moderate correlation (0.4≤|r|<0.7), and weak correlation (|r|<0.4). Based on the distribution of the intensity levels of the three correlation coefficients, stability characteristics are further determined: a stable state is defined as all deviation correlation coefficients exhibiting strong correlation; a transitional state is defined as the presence of weak correlation; and a composite state is defined as the significant difference in intensity levels.

[0126] The final atmospheric state type is output by combining the basic state type with stability characteristics according to preset corresponding rules. For example, when the basic state type is stratified stable and the stability characteristic is stable, the output is a stratified steady state; when the basic state type is convective development and the stability characteristic is transitional, the output is a locally active convective state; when the basic state type is frontal transit and the stability characteristic is complex, the output is an alternating frontal unstable state, etc.

[0127] The basic state type is combined with stability characteristics, and the final atmospheric state type is output according to a preset mapping rule. Wherein:

[0128] When the basic state type is stratified stable and the stability characteristics are stable, the corresponding atmospheric state type is "stratified steady state", which is characterized by a uniform atmospheric structure, weak disturbances and continuous stability.

[0129] When the basic state type is stratified stable and the stability characteristic is transitional, the corresponding atmospheric state type is "stratification transitional state", which is characterized by fluctuations in atmospheric stratification and the presence of slight disturbances.

[0130] When the basic state type is layered stability and the stability characteristics are composite, the corresponding atmospheric state type is "layered composite disturbance state", which is characterized by complex disturbances mixed in the layered structure and diverse changes.

[0131] When the basic state type is convection development type and the stability characteristic is stable state, the corresponding atmospheric state type is "local convection stable state", which is characterized by the formation of convection activity but its relative stability.

[0132] When the basic state type is convection development type and the stability characteristic is transitional state, the corresponding atmospheric state type is "local convection active state", characterized by enhanced convection, rapid development, and enhanced disturbance.

[0133] When the basic state type is convection development type and the stability characteristic is complex state, the corresponding atmospheric state type is "convection complex alternating state", which is characterized by multi-level superposition of convective disturbances and complex structure.

[0134] When the basic state type is frontal transit and the stability characteristic is stable, the corresponding atmospheric state type is "smooth frontal transit state," which is characterized by clear frontal influence but overall relative stability.

[0135] When the basic state type is frontal transit and the stability characteristic is transitional, the corresponding atmospheric state type is "active frontal transit state", characterized by obvious frontal disturbances and frequent meteorological phenomena.

[0136] When the basic state type is frontal transit and the stability characteristic is a composite state, the corresponding atmospheric state type is "frontal alternation unstable state", which is characterized by the superposition of fronts and other disturbances, resulting in extreme instability.

[0137] It should be noted that in practical applications, the above category names and judgment thresholds can be appropriately adjusted or expanded according to regional climate characteristics, observation quality, and operational needs to better adapt to specific meteorological scenarios. Furthermore, when there is insufficient observation data or the correlation coefficient of the deviation cannot be clearly determined, default or fuzzy judgment strategies can be designed to ensure the stable operation of the system.

[0138] It should be noted that the judgment threshold can be appropriately adjusted according to the specific regional climate characteristics and observation quality to improve the judgment sensitivity. In addition, the symbol combination pattern is only used as a basic judgment basis. For special and complex meteorological processes, the system can combine historical data and artificial intelligence-assisted methods for verification to avoid misjudgment.

[0139] Furthermore, step S3, mapping atmospheric state types to multi-source coordinated parameters, includes:

[0140] The atmospheric state types are mapped to the satellite observation-dominated mode, the flow-development-type atmospheric state is mapped to the radar observation-dominated mode, and the frontal transit-type atmospheric state is mapped to the ground observation-dominated mode.

[0141] A baseline weight is assigned to the identified dominant observation sources, and a secondary weight is assigned to the non-dominant observation sources to obtain a basic weight allocation strategy. The baseline weight is greater than the secondary weight, and the sum of the weights of the three observation sources meets the preset total weight requirement.

[0142] When the stability feature is in a stable state, the basic weight allocation remains unchanged. When the stability feature is in a transitional state, the weight of the dominant observation source is adjusted in the first stage according to the preset adjustment coefficient. When the stability feature is in a composite state, the weight of the dominant observation source is adjusted in the second stage according to the preset adjustment coefficient, thus obtaining the multi-source coordination parameters that combine the stability features.

[0143] In some embodiments, the atmospheric state type determined above is mapped to a corresponding dominant observation mode to guide the weight allocation of different meteorological observation data. Specifically, if the atmospheric state type is stratified stable, it is determined to be the satellite observation dominant mode; if it is convective development type, it is determined to be the radar observation dominant mode; and if it is frontal transit type, it is determined to be the ground observation dominant mode. Based on this, the identified dominant observation source is assigned a higher baseline weight, while the other two non-dominant observation sources are assigned lower secondary weights to ensure that the sum of the weights of the three types of observation sources meets the preset total weight requirement, usually 1 or 100%. For example, if the total weight is 1, the weight of the dominant observation source can be set to 0.6, and the weights of the two non-dominant observation sources can each be 0.2, ensuring overall weighted balance. The weights are dynamically adjusted in conjunction with stability characteristics. If the stability characteristics indicate a stable state, the basic weight allocation remains unchanged, reflecting the relatively balanced quality of observational data under the current atmospheric conditions. If it is a transitional state, the weight of the dominant observation source is increased by a preset adjustment coefficient, such as 10%-20%, to strengthen the data influence of the dominant observation source and assist in smoothing the fusion process during the transition phase. If it is a complex state, a higher-level second-level adjustment is made, potentially increasing the weight of the dominant observation source to 70% or higher, while correspondingly reducing the weight of non-dominant observation sources to highlight the main observation source's response capability to complex atmospheric disturbances. For example, in a convective-developing atmospheric state with complex stability characteristics, radar observation, as the dominant observation source, may have its weight increased from the basic 0.6 to 0.75, while ground station and satellite observations may decrease from 0.2 to 0.125, ensuring that the fusion results better reflect the superior information of radar observations.

[0144] It should be noted that the specific magnitude and proportion of weight adjustments need to be pre-set based on the quality statistics of historical observation data and operational needs to avoid excessive adjustments that could lead to information loss or amplified biases. Furthermore, under certain extreme weather conditions, when there are numerous anomalies in data from non-dominant observation sources, the weight adjustment strategy can appropriately incorporate anomaly detection results to locally adjust the weights of observation sources corresponding to anomaly areas, thereby improving the overall accuracy and stability of the fusion.

[0145] Step S4: Calculate the spatial gradient consistency index of different observation sources within the same region. When there is a divergence in gradient direction or amplitude, it is marked as an observational anomaly region.

[0146] In some embodiments, spatial gradient calculations are performed on meteorological elements (such as temperature, humidity, and wind speed) observed by different observation sources within the same region based on multimodal meteorological observation data. Specifically, for each observation source data within the region, a gradient vector is calculated using the numerical differences between neighboring spatial points. This gradient vector contains information about the gradient direction and magnitude. Then, the gradient vectors corresponding to different observation sources are compared within the same spatial grid or target area, focusing on the consistency of gradient direction and the similarity of magnitude. During the comparison, by calculating the angle between gradient directions and the relative difference in magnitude, if the angle exceeds a preset threshold (e.g., greater than 30 degrees) or the magnitude difference exceeds a set proportion (e.g., greater than 20%), the spatial point is identified as a gradient inconsistency point, and the spatial area where this point is located is marked as an observation anomaly area. This anomaly marking reflects the incoordination in the changing trends of meteorological elements in the region between different observation sources, which may originate from observation errors, data delays, or complex local meteorological phenomena.

[0147] It should be noted that the consistency determination of spatial gradients depends not only on the thresholds of direction and amplitude, but also on the temporal synchronicity and spatial resolution differences of the observed data, to avoid misjudgments caused by time delays or differences in spatial scale. Furthermore, the labeling of anomalous regions can be combined with a neighborhood expansion strategy to avoid local misjudgments caused by single-point noise, thereby improving the robustness and accuracy of anomaly detection.

[0148] Step S5: Optimize the multimodal meteorological observation data with geometric complementary weights using multi-source coordination parameters and observational anomaly areas, generate a three-dimensional meteorological field, and achieve visualization output.

[0149] Of particular importance, step S5 includes:

[0150] The multi-source coordination parameters are adjusted based on the observed anomaly regions. The weight of the corresponding observation source is reduced in the anomaly region, while the original weight is maintained in the normal region, resulting in the adjusted fusion weight.

[0151] The multimodal meteorological observation data are weighted and fused using the adjusted fusion weights to generate fused meteorological data.

[0152] A three-dimensional grid structure is established based on the spatial range of the spatiotemporal correlation window. The fused meteorological data is interpolated onto the grid nodes to form regularized three-dimensional meteorological field data.

[0153] The three-dimensional meteorological field data is processed for volume rendering. The color mapping and transparency parameters of meteorological elements are set to generate a three-dimensional visualization image, realizing the dynamic display output of the three-dimensional meteorological field.

[0154] In some embodiments, an observational anomaly range located in a certain area along the East China coast was identified based on previous steps. The radar echo intensity in this area was significantly higher than normal, inconsistent with the trends in ground station wind speed and satellite cloud top temperature. To mitigate the impact of anomalous data on the fusion results, the system adjusted the fusion weight of radar observations in this area from 0.5 to 0.3, while increasing the ground station weight from 0.25 to 0.35, and keeping the satellite observation weight unchanged at 0.25. For other normal areas, the original weight allocation continued. During the fusion phase, the system performed weighted calculations for factors such as temperature, humidity, and wind speed at each 5km × 5km grid point. For example, for the temperature at a certain grid point, the ground station observation value was 18.2℃, the radar inversion value was 19.7℃, and the satellite inversion value was 18.8℃. The weighted result was: T_fuse = 18.2 × 0.35 + 19.7 × 0.3 + 18.8 × 0.25 ≈ 18.8℃. This method ensures that a single anomalous observation source does not excessively influence the results. A 3D grid with a vertical height of 0–12 km and a horizontal resolution of 5 km is established within the entire spatiotemporal correlation window, and the fused meteorological data is inserted into each grid node. For example, when using the inverse distance weighting (IDW) method, the value of each grid node is obtained by weighting and summing the values ​​of the surrounding 8 observation points according to the inverse of the square of the distance, thus forming a continuous and smooth data distribution in space. Finally, the system loads the 3D grid data into the volume rendering engine, sets the grayscale mapping for temperature (lighter colors for higher temperatures), the dashed line density mapping for humidity (denser dashed lines for higher humidity), and sets higher transparency for low humidity areas to make high humidity areas stand out more in the 3D view. Users can use the timeline playback function to view the dynamic changes of the meteorological field gradually moving from the sea to the inland over the past 6 hours, thus intuitively grasping the trend of frontal movement.

[0155] It should be noted that the weight adjustment range and interpolation method should be flexibly determined based on the characteristics of the actual observation data and operational needs, in order to balance computational efficiency and fusion effect. At the same time, the color and transparency mapping parameters of the volume rendering need to be set reasonably to avoid visual misleading and ensure the scientific representation of meteorological elements.

[0156] The present invention also provides a three-dimensional dynamic simulation system 100 based on multimodal weather data, used to execute the above-described three-dimensional dynamic simulation method based on multimodal weather data, wherein the three-dimensional dynamic simulation system based on multimodal weather data includes:

[0157] The multi-source data acquisition module 101 is used to acquire multimodal meteorological observation data, including ground observation station information, radar observation information and satellite observation information, and to calculate the time difference between the observation times of each data source.

[0158] The propagation speed calculation module 102 is used to take the ratio of the time difference to the spatial distance between key observation points as the perturbation propagation speed;

[0159] The state recognition weight module 103 is used to establish a spatiotemporal correlation window based on the disturbance propagation speed, calculate the deviation correlation coefficient between each observation source within the spatiotemporal correlation window, determine the atmospheric state type through the deviation correlation coefficient, and map the atmospheric state type to multi-source coordination parameters.

[0160] The anomaly region detection module 104 is used to calculate the spatial gradient consistency index of different observation sources in the same region. When there is a divergence in gradient direction or amplitude, it is marked as an observation anomaly region.

[0161] The 3D dynamic reconstruction module 105 is used to perform geometric complementary weight optimization on multimodal meteorological observation data using multi-source coordination parameters and observational anomaly areas, and to generate a 3D meteorological field and achieve visualization output.

[0162] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0163] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A three-dimensional dynamic simulation method based on multimodal weather data, characterized in that, Includes the following steps: Step S1: Acquire multimodal meteorological observation data including ground observation station information, radar observation information and satellite observation information, and calculate the time difference between the observation times of each data source; Step S2: Use the ratio of the time difference to the spatial distance between key observation points as the disturbance propagation speed; Step S3: Establish a spatiotemporal correlation window based on the disturbance propagation speed, calculate the deviation correlation coefficient between each observation source within the spatiotemporal correlation window, and determine the atmospheric state type through the deviation correlation coefficient; Map atmospheric state types to multi-source coordinated parameters; The spatiotemporal correlation window established based on the perturbation propagation speed includes: The geometric midpoint of the key observation point pair is used as the spatial center coordinate, and the average observation time of the two observation points is used as the time center coordinate. Divide the preset reference time constant by the disturbance propagation speed to obtain the time radius value, and multiply the disturbance propagation speed by the time radius value to obtain the spatial radius value; Calculate the azimuth angle of the line connecting key observation point pairs as the main propagation direction angle; When the disturbance propagation speed is greater than the preset speed threshold, it is determined to be a directional propagation phenomenon and an elliptical shape parameter is set. When the disturbance propagation speed is less than or equal to the preset speed threshold, it is determined to be a diffusion propagation phenomenon and a circular shape parameter is set, thus obtaining the geometric shape parameter. Using the spatial center coordinates and the time center coordinates as the positioning center of the spatiotemporal correlation window, the spatial boundary range and time extension range of the window are determined by the spatial radius value and the time radius value, respectively. The circular or elliptical structure of the window is set according to the geometric shape parameters. When it is elliptical, the direction of the major axis of the ellipse is determined by the angle of the main propagation direction, thus obtaining the spatiotemporal correlation window. The calculation of the deviation correlation coefficient between each observation source within the spatiotemporal correlation window includes: Extract various types of observation data within the spatiotemporal correlation window and obtain the numerical forecast background field data corresponding to the window region; Various observation data are matched with background field grid points in numerical weather prediction background field data, and the nearest neighbor method is used to determine the background field value corresponding to each observation point, forming an observation-background field matching dataset. The deviation sequence between each observation source and the background field is calculated based on the observation-background field matching dataset. The deviation sequence includes the ground deviation sequence, radar deviation sequence and satellite deviation sequence. The deviation correlation coefficient between observation sources is calculated by using the deviation value sequence. The deviation correlation coefficients between the ground deviation sequence and the radar deviation sequence, the ground deviation sequence and the satellite deviation sequence, and the radar deviation sequence and the satellite deviation sequence are calculated respectively. Step S4: Calculate the spatial gradient consistency index of different observation sources within the same region. When there is a divergence in gradient direction or amplitude, it is marked as an observational anomaly region. Step S5: Optimize the multimodal meteorological observation data with geometric complementary weights using multi-source coordination parameters and observational anomaly areas, generate a three-dimensional meteorological field, and achieve visualization output.

2. The three-dimensional dynamic simulation method based on multimodal weather data according to claim 1, characterized in that, Step S1 includes the following steps: Simultaneously, it receives point meteorological data from ground-based automatic weather stations, spatial scan data from weather radar, and spatial scan data from weather satellites as the raw observation dataset; The original observation dataset was converted into a standardized format to obtain multimodal meteorological observation data; Calculate the time difference between different data sources for multimodal meteorological observation data within the same observation period. The time difference includes the time difference between ground station and satellite observations, the time difference between ground station and radar observations, and the time difference between radar and satellite observations.

3. The three-dimensional dynamic simulation method based on multimodal weather data according to claim 2, characterized in that, The calculation of the time difference between multimodal meteorological observation data from different data sources within the same observation period includes: Extract the effective scan point closest to the spatial location of the target ground station from the spatial scan data of the weather radar; Compare the actual observation timestamp of the target ground station with the scanning timestamp of the effective scanning point, and calculate the absolute difference, which is recorded as the ground station-radar time difference of the target ground station. Based on the pixel unit information corresponding to the geographical location of the target ground station in the spatial scanning data of meteorological satellites, the precise observation timestamp of the pixel unit is read to obtain the pixel timestamp; The pixel timestamp is compared with the actual observation timestamp of the target ground station, and the absolute time interval between the two timestamps is taken as the ground station-satellite time difference of the target ground station. The radar-satellite time difference of the target area is obtained by calculating the time difference between meteorological radar and meteorological satellite observation data based on the target area range. The time difference between ground station and radar, the time difference between ground station and satellite, and the time difference between radar and satellite are combined into a single time difference value.

4. The three-dimensional dynamic simulation method based on multimodal weather data according to claim 3, characterized in that, The calculation of the time difference between meteorological radar and meteorological satellite observation data based on the target area includes: Define a target area, which is specifically a spatial grid cell or a weather feature area; The average time of all scan points constituting the observation of the target area is extracted from the spatial scan data of the weather radar and recorded as the regional radar average time. The average time of all satellite pixels covering the target area is extracted from the spatial scanning data of meteorological satellites and recorded as the regional satellite average time. The radar-satellite time difference of the target area is obtained by calculating the absolute difference between the regional radar average time and the regional satellite average time.

5. The three-dimensional dynamic simulation method based on multimodal weather data according to claim 4, characterized in that, Step S2 includes: Select the two different types of observation points with the smallest observation time difference from the multimodal meteorological observation data as key observation point pairs, including combinations of ground stations and radar stations, combinations of ground stations and satellite pixels, or combinations of radar stations and satellite pixels; Calculate the actual spatial distance between key observation point pairs; The propagation time is determined based on the time difference between key observation points. When there are multiple time differences, the time difference with the smallest value is selected as the propagation time. The propagation speed is obtained by dividing the actual spatial distance between key observation point pairs by the propagation time.

6. The three-dimensional dynamic simulation method based on multimodal weather data according to claim 5, characterized in that, Step S3, which involves determining the atmospheric state type using the deviation correlation coefficient, includes: Check the positive and negative signs of the ground-radar bias correlation coefficient, the ground-satellite bias correlation coefficient, and the radar-satellite bias correlation coefficient, and record the sign combination patterns of the three correlation coefficients; When the symbol combination pattern shows that all three correlation coefficients are positive, it is determined to be a stratified stable atmospheric state. When the symbol combination pattern shows that the ground-radar deviation correlation coefficient is negative while the other two are positive, it is determined to be a convective development atmospheric state. When the symbol combination pattern shows two positive and one negative or one positive and two negative, it is determined to be a frontal transit atmospheric state, thus obtaining the basic state type. The deviation correlation coefficient is divided into three levels: strong correlation, moderate correlation and weak correlation according to the preset threshold rule. When the deviation correlation coefficient is strong, it is judged as a stable state. When the intensity level is weak, it is judged as a transitional state. When the intensity level is large, it is judged as a composite state, thus obtaining the stability characteristics. The basic state type is combined with the stability characteristics, and the atmospheric state type is output according to the preset corresponding rules.

7. The three-dimensional dynamic simulation method based on multimodal weather data according to claim 6, characterized in that, Step S3, mapping atmospheric state types to multi-source coordinated parameters, includes: The atmospheric state types are mapped to the satellite observation-dominated mode, the flow-development-type atmospheric state is mapped to the radar observation-dominated mode, and the frontal transit-type atmospheric state is mapped to the ground observation-dominated mode. A baseline weight is assigned to the identified dominant observation sources, and a secondary weight is assigned to the non-dominant observation sources to obtain a basic weight allocation strategy. The baseline weight is greater than the secondary weight, and the sum of the weights of the three observation sources meets the preset total weight requirement. When the stability feature is in a stable state, the basic weight allocation remains unchanged. When the stability feature is in a transitional state, the weight of the dominant observation source is adjusted in the first stage according to the preset adjustment coefficient. When the stability feature is in a composite state, the weight of the dominant observation source is adjusted in the second stage according to the preset adjustment coefficient, thus obtaining the multi-source coordination parameters that combine the stability features.

8. A three-dimensional dynamic simulation system based on multimodal weather data, characterized in that, For executing the three-dimensional dynamic simulation method based on multimodal weather data as described in claim 1, the three-dimensional dynamic simulation system based on multimodal weather data includes: The multi-source data acquisition module is used to acquire multimodal meteorological observation data, including information from ground observation stations, radar observations, and satellite observations, and to calculate the time difference between the observation times of each data source. The propagation speed calculation module is used to take the ratio of the time difference to the spatial distance between key observation points as the perturbation propagation speed; The state recognition weight module is used to establish a spatiotemporal correlation window based on the disturbance propagation speed, calculate the deviation correlation coefficient between each observation source within the spatiotemporal correlation window, determine the atmospheric state type through the deviation correlation coefficient, and map the atmospheric state type to multi-source coordination parameters. The anomaly detection module is used to calculate the spatial gradient consistency index of different observation sources within the same region. When there is a divergence in gradient direction or amplitude, it is marked as an observation anomaly region. The 3D dynamic reconstruction module is used to perform geometric complementary weight optimization on multimodal meteorological observation data using multi-source coordinated parameters and observational anomaly areas, and to generate a 3D meteorological field and achieve visualization output.

Images