Radar reflectivity inversion method and device and nonvolatile storage medium

By matching historical radar reflectivity and satellite data, screening severe convection samples and training radar reflectivity prediction models, the problem of low versatility of geostationary satellite observation methods was solved, and higher-precision and widely applicable radar reflectivity inversion was achieved.

CN120742318APending Publication Date: 2025-10-03PEKING UNIV CHONGQING RES INST OF BIG DATA
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202510819115.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-06-18
Publication Date
2025-10-03

AI Technical Summary

Technical Problem

The existing radar reflectivity inversion method based on geostationary satellite observations has low versatility and cannot accurately invert radar reflectivity, especially in remote areas and areas with complex terrain, where the observation quality is limited.

Method used

By matching satellite data based on the time and geographic location information of historical radar reflectivity data, screening strong convection samples, dividing the training set and validation set, and using multi-channel imager data and solar zenith angle data to train the radar reflectivity prediction model, the weighted loss function and U-Net model are used to improve the inversion accuracy.

Benefits of technology

It improves the accuracy and versatility of radar reflectivity inversion, enhances adaptability to different geographical environments, and expands the applicable time range of visible light channel information.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120742318A_ABST
    Figure CN120742318A_ABST
Patent Text Reader

Abstract

The invention discloses a radar reflectivity inversion method and device and a nonvolatile storage medium. The method comprises the steps that historical satellite data corresponding to historical radar reflectivity data are matched according to time information and geographic position information of the historical radar reflectivity data, a first data set is obtained, the first data set comprises the historical radar reflectivity data and the historical satellite data which are in one-to-one correspondence, and the first data set comprises the historical radar reflectivity data and the historical satellite data which are in one-to-one correspondence; the historical satellite data comprises solar zenith angle data and multi-channel imager data; screening severe convection samples in the first data set, and taking the screened severe convection samples as a second data set; dividing the second data set into a first training set and a second training set according to the solar zenith angle data; and training the radar reflectivity prediction model according to the first training set and the second training set. The technical problem that the radar reflectivity cannot be accurately inverted due to the fact that an existing radar reflectivity inversion method based on satellite observation is low in universality is solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of monitoring and early warning of severe convective weather, and specifically to a radar reflectivity inversion method, device and non-volatile storage medium. Background Art

[0002] The radar reflectivity factor characterizes the intensity of the scattered echo from a precipitation target in response to electromagnetic waves transmitted by a weather radar. It is related to the size, number, and phase of precipitation particles per unit volume of the precipitation target. The intensity of radar reflectivity can be used to determine the intensity, distribution, movement, and evolution of atmospheric precipitation. Areas with radar echoes greater than 15 dBZ typically indicate significant precipitation, while areas with strong echoes greater than 35 dBZ generally correspond to heavy to torrential rain. Compared to ground-based radar, geostationary satellites have a wider observation range and are not restricted by surface information, effectively compensating for the lack of meteorological observations in areas such as oceans, plateaus, mountains, and deserts. However, existing radar reflectivity inversion methods based on geostationary satellite observations lack versatility and cannot accurately invert radar reflectivity.

[0003] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention

[0004] The embodiments of the present application provide a radar reflectivity inversion method, device, and non-volatile storage medium to at least solve the technical problem of the inability to accurately invert radar reflectivity due to the low versatility of existing radar reflectivity inversion methods based on satellite observations.

[0005] According to one aspect of an embodiment of the present application, a radar reflectivity inversion method is provided, comprising: matching historical satellite data corresponding to the historical radar reflectivity data based on time information and geographic location information of the historical radar reflectivity data to obtain a first data set, wherein the first data set includes one-to-one corresponding historical radar reflectivity data and historical satellite data, and the historical satellite data includes solar zenith angle data and multi-channel imager data, and the multi-channel imager data is used to determine band characteristic data; screening strong convection samples in the first data set, and using the screened strong convection samples as a second data set, wherein the strong convection samples are samples whose radar reflectivity data meets a preset intensity condition; and dividing the second data set into a first training set and a second training set based on the solar zenith angle data. The solar zenith angle of the historical satellite data in the first training set is less than or equal to a first preset threshold, and the solar zenith angle of the historical satellite data in the second training set is greater than the first preset threshold. The first training set includes solar zenith angle data, terrain height data, first band feature data and second band feature data. The second training set includes solar zenith angle data, terrain height data and second band feature data. The bands of the first band feature data include visible light band and shortwave infrared band, and the bands of the second band feature data include water vapor band and medium and long wave infrared band. The radar reflectivity prediction model is trained based on the first training set and the second training set, wherein the radar reflectivity prediction model is used to invert the radar reflectivity under target conditions, and the target conditions include target time point and target area.

[0006] Optionally, screening the severe convection samples in the first data set includes: cropping the radar reflectivity data in the first data set according to a preset window size to obtain a radar image block, wherein the radar image block includes a plurality of grid points; screening valid grid points according to preset screening rules, wherein the valid grid points are grid points whose radar reflectivity data meet the preset screening rules; determining a valid radar image block based on the valid grid points, wherein the valid radar image block only includes valid grid points; determining a first valid radar image block with the largest average reflectivity from the valid radar image blocks, and removing a second valid radar image block from the first data set, wherein the second valid radar image block is the first valid radar image block except the first valid radar image block. a valid radar image block other than the first valid radar image block and including the center grid point of the first valid radar image block; determining a new first valid radar image block again from the third valid radar image block, wherein the third valid radar image block is a valid radar image block other than the first valid radar image block in the first dataset; removing the second valid radar image block corresponding to the new first valid radar image block from the first dataset, updating the third valid radar image block, and determining a new first valid radar image block again from the new third valid radar image block, until no second valid radar image block corresponding to the new first valid radar image block exists in the first dataset.

[0007] Optionally, filtering valid grid points according to preset filtering rules includes: counting the number of valid data for each grid point, wherein the number of valid data is the number of times the radar reflectivity is greater than a second preset threshold in the historical radar reflectivity data corresponding to the grid point; and determining the grid point whose number of valid data is greater than a third preset threshold as a valid grid point.

[0008] Optionally, before training the radar reflectivity prediction model based on the first training set and the second training set, the method also includes: normalizing the first band characteristic data and the second band characteristic data in the first training set; normalizing the second band characteristic data in the second training set; cosine encoding the solar zenith angle data in the first training set or the second training set, and normalizing the cosine-encoded solar zenith angle data; updating the value of the terrain height data in the first training set or the second training set that is lower than the fourth preset threshold to the fourth preset threshold, and normalizing the updated terrain height data.

[0009] Optionally, the radar reflectivity prediction model includes a first radar reflectivity prediction model and a second radar reflectivity prediction model, and training the radar reflectivity prediction model based on the first training set and the second training set includes: inputting the first training set into the first radar reflectivity prediction model to be trained; inputting the second training set into the second radar reflectivity prediction model to be trained; training multiple radar reflectivity prediction models to be trained based on a weighted loss function to obtain multiple radar reflectivity prediction models to be evaluated, wherein the weights of the weighted loss function include prior weights based on probability density distribution and error weights based on mean absolute error in training; determining the scores of the radar reflectivity prediction models to be evaluated based on preset scoring rules; determining the model with the highest score in the first radar reflectivity prediction model to be evaluated as the first radar reflectivity prediction model; and determining the model with the highest score in the second radar reflectivity prediction model to be evaluated as the second radar reflectivity prediction model.

[0010] Optionally, determining the score of the radar reflectivity prediction model to be evaluated according to a preset scoring rule includes: determining the category of the prediction result of the radar reflectivity prediction model to be evaluated according to a preset target interval range, wherein the categories include: the prediction result is in the target interval and the corresponding historical radar reflectivity is in the target interval, the prediction result is in the target interval and the corresponding historical radar reflectivity is not in the target interval, the prediction result is not in the target interval and the corresponding historical radar reflectivity is in the target interval, and the prediction result is not in the target interval and the corresponding historical radar reflectivity is not in the target interval; counting the number of prediction results in each category, and determining the first score of the radar reflectivity prediction model to be evaluated corresponding to the preset target interval according to a preset scoring formula; determining the average of the first scores corresponding to each preset target interval range as the score of the radar reflectivity prediction model to be evaluated.

[0011] Optionally, the weighted loss function is as follows: Among them, n b is the total number of grid points corresponding to the current training batch samples, y′ i and y i The predicted value corresponding to the i-th grid point and the true value corresponding to the i-th grid point, respectively, The current training batch samples are divided into n c +1 interval, is the total number of grid points corresponding to the k-th interval sample in the second data set, is the total number of grid points corresponding to the kth interval sample in the current training batch.

[0012] Optionally, a first score corresponding to the radar reflectivity prediction model to be evaluated and the preset target interval is determined according to the following formula: Among them, HSS is the first score, N A N is the number of samples whose prediction results are in the target interval and whose corresponding historical radar reflectivity is in the target interval. B N is the number of samples whose prediction results are in the target interval and whose corresponding historical radar reflectivity is not in the target interval. C If the prediction result is not in the target interval and the corresponding historical radar reflectivity is in the target interval, N D is the number of samples whose prediction results are not in the target interval and whose corresponding historical radar reflectivity is not in the target interval.

[0013] Optionally, after training the radar reflectivity prediction model based on the first training set and the second training set, the method further includes: determining target satellite data based on the target time point and the target area, wherein the difference between the observation time point of the target satellite data and the target time point is less than a preset interval, and the observation area of ​​the target satellite data is the same as the target area; cropping the target satellite data based on a preset window size to obtain target satellite data blocks and block indexes; classifying the target satellite data blocks based on solar zenith angle data of the target satellite data blocks to obtain a first target satellite data block and a second target satellite data block, wherein the solar zenith angle of the first target satellite data block is less than or equal to a first preset threshold, and the solar zenith angle of the second target satellite data block is greater than the first preset threshold; obtaining prediction results of the first target satellite data block and the second target satellite data block based on the radar reflectivity prediction model, wherein the prediction results include the radar reflectivity corresponding to the first target satellite data block or the second target satellite data block; and superimposing and calculating each prediction result according to a preset calculation rule to determine the radar reflectivity corresponding to the target condition.

[0014] Optionally, performing superposition calculation on each prediction result according to a preset calculation rule to determine the radar reflectivity corresponding to the target condition includes: multiplying each prediction result by a preset weight; initializing a first all-zero array having the same size as the target satellite data; adding the prediction result multiplied by the preset weight to the corresponding position of the first all-zero array according to the block index to obtain a first array; initializing a second all-zero array having the same size as the target satellite data; adding the preset weight to the corresponding position of the second all-zero array according to the block index to obtain a second array; dividing the first array by the second array, and determining the calculation result as the radar reflectivity corresponding to the target condition.

[0015] Optionally, the radar reflectivity prediction model includes a first radar reflectivity prediction model and a second radar reflectivity prediction model, the first radar reflectivity prediction model is obtained by training based on the first training set, and the second radar reflectivity prediction model is obtained by training based on the second training set; obtaining prediction results of the first target satellite data block and the second target satellite data block based on the radar reflectivity prediction model includes: inputting first feature data of the first target satellite data block into the first radar reflectivity prediction model to obtain a prediction result, wherein the first feature data includes solar zenith angle data, terrain height data, first band feature data and second band feature data, and the bands of the first band feature data include visible light band and infrared band; inputting second feature data of the second target satellite data block into the second radar reflectivity prediction model to obtain a prediction result, wherein the second feature data includes solar zenith angle data, terrain height data and second band feature data.

[0016] According to another aspect of the embodiment of the present application, a radar reflectivity inversion device is also provided, including: a matching module, for matching historical satellite data corresponding to the historical radar reflectivity data based on the time information and geographic location information of the historical radar reflectivity data to obtain a first data set, wherein the first data set includes one-to-one corresponding historical radar reflectivity data and historical satellite data, and the historical satellite data includes solar zenith angle data and multi-channel imager data, and the multi-channel imager data is used to determine the band characteristic data; a screening module, for screening strong convection samples in the first data set, and using the screened strong convection samples as a second data set, wherein the strong convection samples are samples whose radar reflectivity data meet a preset intensity condition; a dividing module, for dividing the second data set into a first training set and a second training set according to the solar zenith angle data. A set and a second training set, wherein the solar zenith angle of the historical satellite data in the first training set is less than or equal to a first preset threshold, and the solar zenith angle of the historical satellite data in the second training set is greater than the first preset threshold, the first training set includes solar zenith angle data, terrain height data, first band feature data and second band feature data, the second training set includes solar zenith angle data, terrain height data and second band feature data, the bands of the first band feature data include visible light band and shortwave infrared band, and the bands of the second band feature data include water vapor band and medium and long wave infrared band; a training module is used to train a radar reflectivity prediction model based on the first training set and the second training set, wherein the radar reflectivity prediction model is used to invert radar reflectivity under target conditions, and the target conditions include target time point and target area.

[0017] According to another aspect of an embodiment of the present application, a non-volatile storage medium is provided, in which a program is stored. When the program is executed, the device where the non-volatile storage medium is located is controlled to execute a radar reflectivity inversion method.

[0018] According to another aspect of an embodiment of the present application, an electronic device is provided, including: a memory and a processor, wherein the processor is configured to run a program stored in the memory, wherein the radar reflectivity inversion method is executed when the program is run.

[0019] According to another aspect of an embodiment of the present application, a computer program product is further provided, including a computer program, which implements a radar reflectivity inversion method when executed by a processor.

[0020] In an embodiment of the present application, time information and geographic location information of historical radar reflectivity data are used to match historical satellite data corresponding to the historical radar reflectivity data to obtain a first data set, wherein the first data set includes one-to-one corresponding historical radar reflectivity data and historical satellite data, and the historical satellite data includes solar zenith angle data and multi-channel imager data, and the multi-channel imager data is used to determine band characteristic data; strong convection samples in the first data set are screened, and the screened strong convection samples are used as the second data set, wherein the strong convection samples are samples whose radar reflectivity data meets a preset intensity condition; the second data set is divided into a first training set and a second training set based on the solar zenith angle data, wherein the solar zenith angle of the historical satellite data in the first training set is less than or equal to a first preset threshold, and the solar zenith angle of the historical satellite data in the second training set is greater than the first preset threshold, and the first training set includes solar zenith angle data, terrain height data, first band characteristic data and second band Characteristic data, the second training set includes solar zenith angle data, terrain height data and second band characteristic data, the bands of the first band characteristic data include visible light band and short-wave infrared band, and the bands of the second band characteristic data include water vapor band and medium- and long-wave infrared band; the radar reflectivity prediction model is trained based on the first training set and the second training set, wherein the radar reflectivity prediction model is used to invert the radar reflectivity under target conditions, and the target conditions include target time point and target area. By screening the strong convection sample training model, establishing an adaptive loss function, and supplementing the solar zenith angle and terrain height as characteristic channel data, the purpose of improving the inversion accuracy of strong echo areas, expanding the applicable time range of visible light channel information, and enhancing the adaptability to the geographical environment is achieved, thereby achieving the technical effect of improving the accuracy and versatility of intelligent inversion of radar reflectivity, and thus solving the technical problem of inaccurate inversion of radar reflectivity due to the low versatility of the existing radar reflectivity inversion method based on satellite observation. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0022] Figure 1 is a structural diagram of a computer terminal provided according to an embodiment of the present application;

[0023] Figure 2 1 is a flow chart of a radar reflectivity inversion method provided according to an embodiment of the present application;

[0024] Figure 3 This is a schematic diagram of the cutting and screening effect of a strong convection sample provided in an embodiment of the present application;

[0025] Figure 4 Schematic diagram of a U-Net model provided according to an embodiment of the present application;

[0026] Figure 5 This is a schematic diagram of a Gaussian weighted code effect provided according to an embodiment of the present application;

[0027] Figure 6 1 is a flow chart of another radar reflectivity inversion method provided according to an embodiment of the present application;

[0028] Figure 7 3 is a schematic structural diagram of a radar reflectivity inversion device provided according to an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0030] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in a sequence other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0031] The radar reflectivity factor measures the intensity of the scattered echo from a precipitation target to electromagnetic waves transmitted by a weather radar. It is related to the size, number, and phase of precipitation particles per unit volume of the precipitation target. Radar reflectivity can be used to determine the intensity, distribution, movement, and evolution of precipitation in the atmosphere. Areas with radar echoes greater than 15dBZ typically indicate significant precipitation, while areas with strong echoes greater than 35dBZ generally correspond to heavy to torrential rain. However, ground-based radar observations often lack effective coverage in remote areas and over ocean regions, and are susceptible to interference from ground objects in complex terrain, impacting observation quality.

[0032] Compared to ground-based radar, geostationary satellites have a wider observation range and are not restricted by surface information, effectively compensating for the lack of meteorological observations in areas such as oceans, plateaus, mountains, and deserts. The difficulty in retrieving radar reflectivity from geostationary satellite data lies in the lack of a one-to-one correspondence between the two types of observational data. Satellite data do not directly observe precipitation particles, but instead scan and image radiation from above clouds in different wavelengths. Therefore, it is difficult to derive an accurate correspondence between satellite data and radar reflectivity using physical formulas.

[0033] Artificial intelligence technology has developed rapidly in recent years, and deep learning-based models of the relationship between satellite and radar data have become a promising approach. Current research commonly uses multi-channel data from satellite imagers, using U-Net architectures to construct radar emissivity inversion simulation models. While this research has achieved some success, the practical application of existing models still faces the following three challenges:

[0034] First, the sample imbalance problem makes inversion difficult in strong echo areas. Although existing studies have partially alleviated the sample imbalance problem by artificially setting error weights for different echo intensities, this weight setting lacks objectivity and universality.

[0035] Second, the information of visible light and shortwave infrared channels is insufficiently utilized. Studies have shown that the introduction of higher-precision visible light and shortwave infrared channels is beneficial to improving the inversion details of daytime models, but currently it is basically only used in situations where the solar zenith angle is greater than 70° or 65°.

[0036] Third, the inversion model lacks characteristics of different geographical environments. Studies have shown that there are differences in the inversion effects in inland, coastal, offshore and ocean areas, but there is currently a lack of special treatment for related characteristics.

[0037] In order to solve the above problems, relevant solutions are provided in the embodiments of the present application, which are described in detail below.

[0038] According to an embodiment of the present application, a method embodiment of a radar reflectivity inversion method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0039] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 The hardware structure block diagram of a computer terminal for implementing the radar reflectivity inversion method is shown in FIG. Figure 1As shown, the computer terminal 10 may include one or more (illustrated as 102a, 102b, ..., 102n in the figure) processors 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It will be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components than shown, or with Figure 1 Different configurations shown.

[0040] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuitry." The data processing circuitry may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuitry may be a single, independent processing module, or may be incorporated in whole or in part into any of the other components of the computer terminal 10. As described in the embodiments of the present application, the data processing circuitry serves as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0041] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the radar reflectivity inversion method in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby implementing the above-mentioned radar reflectivity inversion method. The memory 104 may include a high-speed random access memory and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely located relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0042] The transmission device 106 is configured to receive or transmit data via a network. A specific example of the aforementioned network may include a wireless network provided by the communications provider of the computer terminal 10. In one embodiment, the transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the Internet. In another embodiment, the transmission device 106 may be a radio frequency (RF) module, which is configured to communicate with the Internet wirelessly.

[0043] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 .

[0044] In the above operating environment, the embodiment of the present application provides a radar reflectivity inversion method, such as Figure 2 As shown, the method includes the following steps:

[0045] Step S202: Match historical satellite data corresponding to the historical radar reflectivity data based on the time information and geographic location information of the historical radar reflectivity data to obtain a first data set, wherein the first data set includes a one-to-one correspondence between the historical radar reflectivity data and the historical satellite data, the historical satellite data includes solar zenith angle data and multi-channel imager data, and the multi-channel imager data is used to determine the band characteristic data.

[0046] Alternatively, first obtain L1-level data (historical satellite data) from the satellite platform, which has been calibrated and geolocated by geostationary satellite radiometry. Specifically, read the observation data from the multi-channel scanning imaging radiometer and convert the raw digital signal into band reflectivity using a lookup table. Then, read the matching satellite observation positioning information (including solar zenith angle data and row and column numbers) and use the full-disk row and column number to convert longitude and latitude lookup table to obtain the longitude and latitude information of the satellite data. Next, based on the longitude and latitude range (i.e., geographic location information) of the historical radar reflectivity data, clip the satellite data for the area corresponding to the historical radar reflectivity data.

[0047] Optionally, before matching historical radar reflectivity data with historical satellite data, quality control is also included: first, probability distribution analysis is performed on the multi-channel imager data and solar zenith angle data of the satellite data to remove problematic samples containing invalid values ​​and obvious discrete values; second, after the radar reflectivity data is imaged, samples with obvious clutter are manually screened and eliminated.

[0048] Optionally, matching the historical satellite data corresponding to the historical radar reflectivity data based on the time information and geographic location information of the historical radar reflectivity data includes: in terms of time matching, matching the historical radar reflectivity data with the nearest satellite observation data one by one according to the observation time (i.e., the time information of the historical radar reflectivity data); if the time difference between the radar observation and the satellite observation is more than 3 minutes, removing the sample, thereby obtaining time-matched radar reflectivity and satellite observation samples; in terms of space matching, interpolating the satellite observation data to the latitude and longitude grid of the radar observation (i.e., the geographic location information of the historical radar reflectivity data) through the nearest interpolation method. Previous studies have basically matched to a preset specification grid in the opposite or compromise manner, where the preset specification can be 2 km or 4 km; thereby obtaining a spatiotemporal matching sample set D m (ie the first data set).

[0049] Step S204 , screening the strong convection samples in the first data set, and using the screened strong convection samples as the second data set, wherein the strong convection samples are samples whose radar reflectivity data meet a preset intensity condition.

[0050] As an optional implementation, screening the strong convection samples in the first data set includes: cropping the radar reflectivity data in the first data set according to a preset window size to obtain a radar image block, wherein the radar image block includes a plurality of grid points; screening valid grid points according to preset screening rules, wherein the valid grid points are grid points whose radar reflectivity data meet the preset screening rules; determining a valid radar image block based on the valid grid points, wherein the valid radar image block only includes valid grid points; determining a first valid radar image block with the largest average reflectivity from the valid radar image blocks, and removing a second valid radar image block from the first data set, wherein the second valid radar image block The block is a valid radar image block other than the first valid radar image block and includes a central grid point of the first valid radar image block; a new first valid radar image block is again determined in the third valid radar image block, wherein the third valid radar image block is a valid radar image block other than the first valid radar image block in the first dataset; a second valid radar image block corresponding to the new first valid radar image block is removed from the first dataset, the third valid radar image block is updated, and a new first valid radar image block is again determined in the new third valid radar image block, until no second valid radar image block corresponding to the new first valid radar image block exists in the first dataset.

[0051] Optionally, cropping the radar reflectivity data in the first dataset according to a preset window size includes: setting a specified window size (i.e., the preset window size), slidingly cropping the original image from the upper left corner to the right and downward at a certain distance; if the sliding window cannot cover the last row or column, then taking a row or column from the lower right corner to the left and upward to obtain the sliding cropping window start and end indexes of all sliding windows. During model training, the sliding cropping window only retains windows where all grid points are valid grid points; during model testing, the sliding cropping window only retains windows with valid grid points; and when the model is spliced ​​and output, all sliding cropping windows are retained.

[0052] Optionally, D m The radar data and satellite data are cut sample by sample, and the severe convection sample set D is obtained by screening. p Including: Assume that the total number of retained cropping windows (i.e., valid radar image blocks) is N p , among which, for D m The i-th sample The specific operations are as follows:

[0053] (1) According to N p The start and end indexes of the sliding cropping window, The radar data is clipped to obtain N p radar image blocks;

[0054] (2) According to the distribution of radar reflectivity of the actual mission, the threshold for judging strong convection samples is set, and the non-strong convection radar image blocks in the image blocks obtained in (1) are eliminated;

[0055] (3) Take the image block P with the strongest average reflectivity max (i.e. the first valid radar image block), and then remove the image blocks containing P max The radar image block of the center grid point (i.e., the second valid radar image block);

[0056] (4) In the remaining radar image blocks (excluding the P max , i.e. the third effective radar image block), repeat the above step (3) until there is no radar image block to be removed, and then select the image block according to the position of the remaining image block. Cut the corresponding satellite data and terrain altitude data to obtain Strong convection sample, cropping effect reference Figure 3 Example, where the fill color reflects the intensity of radar reflectivity, the white areas correspond to invalid data grid points, and the rectangular box is the effect image after cropping and filtering.

[0057] Optionally, filtering valid grid points according to preset filtering rules includes: counting the number of valid data for each grid point, wherein the number of valid data is the number of times the radar reflectivity is greater than a second preset threshold in the historical radar reflectivity data corresponding to the grid point; and determining the grid point whose number of valid data is greater than a third preset threshold as a valid grid point.

[0058] Optionally, a radar intensity threshold (i.e., a second preset threshold) is set as a representative of valid data, and then all radar data are traversed to obtain a spatial distribution map of the valid data frequency (i.e., the number of valid data), and the grid points whose frequency exceeds the set threshold (i.e., the third preset threshold) are taken as valid grid points (i.e., valid grid points).

[0059] Step S206: Divide the second data set into a first training set and a second training set based on the solar zenith angle data, wherein the solar zenith angle of the historical satellite data in the first training set is less than or equal to a first preset threshold, and the solar zenith angle of the historical satellite data in the second training set is greater than the first preset threshold. The first training set includes solar zenith angle data, terrain height data, first band feature data and second band feature data; the second training set includes solar zenith angle data, terrain height data and second band feature data; the bands of the first band feature data include a visible light band and a shortwave infrared band; and the bands of the second band feature data include a water vapor band and a medium- and long-wave infrared band.

[0060] Optionally, the solar zenith angle θ maximum threshold Th θ (i.e., the first preset threshold), which is generally set to 65° or 70° in previous studies. After testing, using the method of the embodiment of the present application, Th θ The highest setting can be 85°. The solar zenith angle is mainly affected by time and directly reflects the time information. 85° roughly corresponds to around 6:00 am Beijing time. p The maximum solar zenith angle is greater than Th θ The night sample set D p_night (i.e. the second training set), D p The maximum solar zenith angle is less than Th θ The daytime sample set D p_day The development of severe convection has spatiotemporal continuity. To avoid similar samples in the validation set, test set, and training set, the validation set, training set, and test set must be divided to ensure that the sample dates in the subsets do not overlap.

[0061] It should be noted that the satellite characteristic channels used in the night sample set include 3.75μm (low), 6.25μm, 7.1μm, 8.5μm, 10.7μm, 12.0μm, 13.5μm (i.e., second-band characteristic data), and the solar zenith angle. The satellite characteristic channels used during the daytime include four additional observation channels: 0.65μm, 1.375μm, 1.61μm, and 2.25μm (i.e., first-band characteristic data). The input variables for model training include satellite characteristic channels and terrain height, and the label is radar reflectivity.

[0062] Step S208 : training a radar reflectivity prediction model based on the first training set and the second training set, wherein the radar reflectivity prediction model is used to invert radar reflectivity under target conditions, where the target conditions include a target time point and a target area.

[0063] In the technical solution provided in step S208, before training the radar reflectivity prediction model based on the first training set and the second training set, the method also includes: normalizing the first band feature data and the second band feature data in the first training set; normalizing the second band feature data in the second training set; cosine encoding the solar zenith angle data in the first training set or the second training set, and normalizing the cosine-encoded solar zenith angle data; updating the value of the terrain height data in the first training set or the second training set that is lower than the fourth preset threshold to the fourth preset threshold, and normalizing the updated terrain height data.

[0064] Optionally, the normalization process refers to querying the maximum and minimum values ​​of each variable in the sample set for the geostationary satellite data and the terrain data, and performing normalization according to the following formula:

[0065]

[0066] Among them, Var new 、Var org 、Var max 、Var min They are the normalized data of a variable, the original data before normalization, the maximum value of the original data, and the minimum value of the original data.

[0067] It should be noted that, considering that the underwater terrain height in the ocean area has little effect on the rise and condensation of water vapor, the minimum terrain height threshold (i.e., the fourth preset threshold) is set first, and the height below the fourth preset threshold is directly set to the fourth preset threshold before normalization.

[0068] In addition, the solar zenith angle θ needs to be cosine coded (S θ ) and then normalized, the encoding formula is as follows:

[0069]

[0070] In the technical solution provided in step S208, the radar reflectivity prediction model includes a first radar reflectivity prediction model and a second radar reflectivity prediction model, and training the radar reflectivity prediction model based on the first training set and the second training set includes: inputting the first training set into the first radar reflectivity prediction model to be trained; inputting the second training set into the second radar reflectivity prediction model to be trained; training multiple radar reflectivity prediction models to be trained based on a weighted loss function to obtain multiple radar reflectivity prediction models to be evaluated, wherein the weights of the weighted loss function include a priori weights based on probability density distribution and an error weight based on the mean absolute error in training; determining the scores of the radar reflectivity prediction models to be evaluated based on preset scoring rules; determining the model with the highest score in the first radar reflectivity prediction model to be evaluated as the first radar reflectivity prediction model; and determining the model with the highest score in the second radar reflectivity prediction model to be evaluated as the second radar reflectivity prediction model.

[0071] Optionally, the radar reflectivity prediction model uses Figure 4 In the U-Net model shown, adding a residual network structure during downsampling and upsampling can effectively improve the model effect. It should be noted that it is not recommended to add a spatial attention mechanism when applying the method embodiment of this application. Although previous research on radar inversion has proposed that the spatial attention mechanism is beneficial for the inversion of strong convection areas, in the method embodiment of this application, effective strong convection sample screening has been performed before model training. Experiments have shown that the use of the spatial attention mechanism in this method will have the opposite effect, and too many parameters can easily lead to overfitting.

[0072] Optionally, the weighted loss function is as follows: Among them, n b is the total number of grid points corresponding to the current training batch samples, y′ i and y i The predicted value corresponding to the i-th grid point and the true value corresponding to the i-th grid point, respectively, The current training batch samples are divided into n c +1 interval, is the total number of grid points corresponding to the k-th interval sample in the second data set, is the total number of grid points corresponding to the kth interval sample in the current training batch.

[0073] Optionally, the improved adaptive loss function (i.e., weighted loss function) is obtained by weighting the mean square error, and the weight includes a priori weight w based on the probability density distribution. pand the error weight w based on the mean absolute error (MAE) during training e , this loss function helps the model training focus on both the small sample interval and the interval with greater prediction difficulty. Specifically, according to the radar reflectivity intensity, the radar reflectivity data is divided into n c +1 interval, marked from weak to strong The recommended range for C0 is radar reflectivity ≤ 0dBZ.

[0074] For Category C k (k∈[0,n c ]), the prior weight calculation formula of its probability density distribution is as follows:

[0075]

[0076] in is a strong sample set D p All radar reflectivity data in category C k The total number of data points is the total number of grid points corresponding to a certain type of sample.

[0077] For Category C k (k∈[0,n c ]), the mean absolute error is calculated as follows:

[0078] When y i ∈C k

[0079] Among them, y′ i and y i are the predicted value and the true value respectively, is the category C in the current training batch sample k The total number of data points.

[0080] For Category C k (k∈[0,n c ]), the error weight calculation formula is as follows:

[0081] When y∈C k

[0082] Note that formula w e It is calculated based on the prediction error of the current training batch, and each batch w e is not fixed. During model training, the adaptive loss function for the current batch is calculated as follows:

[0083]

[0084] Among them, y′ i and yi are the predicted value and the true value respectively, n b is the total number of data points in the current training batch.

[0085] As an optional implementation, determining the score of the radar reflectivity prediction model to be evaluated according to a preset scoring rule includes: determining the category of the prediction result of the radar reflectivity prediction model to be evaluated according to a preset target interval range, wherein the categories include: the prediction result is in the target interval and the corresponding historical radar reflectivity is in the target interval, the prediction result is in the target interval and the corresponding historical radar reflectivity is not in the target interval, the prediction result is not in the target interval and the corresponding historical radar reflectivity is in the target interval, and the prediction result is not in the target interval and the corresponding historical radar reflectivity is not in the target interval; counting the number of prediction results in each category, and determining the first score of the radar reflectivity prediction model to be evaluated corresponding to the preset target interval according to a preset scoring formula; determining the average of the first scores corresponding to each preset target interval range as the score of the radar reflectivity prediction model to be evaluated.

[0086] Optionally, when selecting a model, the Heidke skill score (HSS) of the true value of the model radar reflectivity in the range of ≥15dBZ, ≥35dBZ and ≥45dBZ (i.e., the preset target range) is calculated based on the validation set to obtain the HSS 15 、HSS 35 and HSS 45 , then select the model f with the highest average value from the trained models f (i.e., the radar reflectivity prediction model to be evaluated) * , the average of the first scores corresponding to each preset target range is determined by the following formula:

[0087]

[0088] Optionally, a first score corresponding to the radar reflectivity prediction model to be evaluated and the preset target interval is determined according to the following formula: Among them, HSS is the first score, N A N is the number of samples whose prediction results are in the target interval and whose corresponding historical radar reflectivity is in the target interval. B N is the number of samples whose prediction results are in the target interval and whose corresponding historical radar reflectivity is not in the target interval. C If the prediction result is not in the target interval and the corresponding historical radar reflectivity is in the target interval, N D is the number of samples whose prediction results are not in the target interval and whose corresponding historical radar reflectivity is not in the target interval.

[0089] Optionally, the HSS is calculated as follows:

[0090]

[0091] Optionally, N A 、N B 、N C and N D They are the total number of data points that each sample meets the classification requirements. The classification method is as follows:

[0092]

[0093] In the technical solution provided in step S208, after training the radar reflectivity prediction model based on the first training set and the second training set, the method further includes: determining target satellite data based on the target time point and the target area, wherein the difference between the observation time point of the target satellite data and the target time point is less than a preset interval, and the observation area of ​​the target satellite data is the same as the target area; cropping the target satellite data based on a preset window size to obtain target satellite data blocks and block indexes; classifying the target satellite data blocks based on solar zenith angle data of the target satellite data blocks to obtain a first target satellite data block and a second target satellite data block, wherein the solar zenith angle of the first target satellite data block is less than or equal to a first preset threshold, and the solar zenith angle of the second target satellite data block is greater than the first preset threshold; obtaining prediction results for the first target satellite data block and the second target satellite data block based on the radar reflectivity prediction model, wherein the prediction results include a radar reflectivity corresponding to the first target satellite data block or the second target satellite data block; and performing superposition calculation on each prediction result according to a preset calculation rule to determine the radar reflectivity corresponding to the target condition.

[0094] Optionally, determining the target satellite data based on the target time point and the target area includes matching the closest satellite observation data according to the target time point, and retaining satellite data whose time difference with the satellite observation at the target time point is within 3 minutes; in spatial matching, retaining satellite data with the same longitude and latitude as the target area, and using the satellite data retained after matching as the target satellite data.

[0095] Optionally, a specified window size is set, and the original image is slidingly cropped from the upper left corner to the right and downward according to a certain distance. If the sliding window cannot cover the last row or column, a row or column is taken from the lower right corner to the left and upward to obtain the sliding cropping window start and end indexes of all sliding windows.

[0096] Optionally, the number of target satellites is normalized according to the following formula:

[0097]

[0098] Among them, Var new 、Var org 、Varmax 、Var min They are the normalized data of a variable, the original data before normalization, the maximum value of the original data, and the minimum value of the original data.

[0099] It should be noted that, considering that the underwater terrain height in the ocean area has little effect on the rise and condensation of water vapor, the minimum terrain height threshold (i.e., the fourth preset threshold) is set first, and the height below the fourth preset threshold is directly set to the fourth preset threshold before normalization.

[0100] In addition, the solar zenith angle θ needs to be cosine coded (S θ ) and then normalized, the encoding formula is as follows:

[0101]

[0102] As an optional implementation, the radar reflectivity prediction model includes a first radar reflectivity prediction model and a second radar reflectivity prediction model, the first radar reflectivity prediction model is trained based on the first training set, and the second radar reflectivity prediction model is trained based on the second training set; obtaining prediction results of the first target satellite data block and the second target satellite data block based on the radar reflectivity prediction model includes: inputting first feature data of the first target satellite data block into the first radar reflectivity prediction model to obtain a prediction result, wherein the first feature data includes solar zenith angle data, terrain height data, first band feature data and second band feature data, the bands of the first band feature data include solar zenith angle data, terrain height data, visible light band and shortwave infrared band, and the bands of the second band feature data include water vapor band and medium and long wave infrared band; inputting second feature data of the second target satellite data block into the second radar reflectivity prediction model to obtain a prediction result, wherein the second feature data includes second band feature data.

[0103] Optionally, set the solar zenith angle to be less than or equal to Th θ (i.e., the first preset threshold) satellite data block (i.e., the first target satellite data block) is input (i.e. the first radar reflectivity prediction model), greater than Th θ Input of (second target satellite data block) (i.e. the second radar reflectivity prediction model), and obtain the block inversion radar reflectivity y′ p (i.e. predicted results).

[0104] As an optional implementation, performing a superposition calculation on each prediction result according to a preset calculation rule to determine the radar reflectivity corresponding to the target condition includes: multiplying each prediction result by a preset weight; initializing a first all-zero array of the same size as the target satellite data; adding the prediction result multiplied by the preset weight to the corresponding position of the first all-zero array according to a block index to obtain a first array; initializing a second all-zero array of the same size as the target satellite data; adding the preset weight to the corresponding position of the second all-zero array according to the block index to obtain a second array; dividing the first array by the second array, and determining the calculation result as the radar reflectivity corresponding to the target condition.

[0105] Optionally, set an array of preset window size with all values ​​0 except the center point being 1, and use Gaussian filtering algorithm to calculate the weight code w G (i.e. preset weights), to adapt to the radar inversion task, obtain w G It is recommended that the standard deviation of the Gaussian filter be set to 0.25 times the window size, and the array is expanded by copying the pattern of the nearest data when the filter exceeds the boundary. The resulting w G Effects such as Figure 5 shown.

[0106] Set two all-zero arrays y′ of the same shape and size as the output area all (i.e. the first all-zero array) and w all (i.e. the second all-zero array), the block data and weights are superimposed and calculated in sequence, and the i-th block data y′ pi and the corresponding index The calculation formula is as follows:

[0107]

[0108] Then through y′ all Divide by w all Regularization is achieved to obtain the final predicted value y′ fnl (i.e. radar reflectivity).

[0109] The embodiment of the present application provides a radar reflectivity inversion method, such as Figure 6 As shown, the method includes the following steps:

[0110] S602. Obtain historical geostationary satellite data and radar reflectivity data, and after preprocessing, quality control, and spatiotemporal matching, establish a dataset D of satellite data and radar observations that are spatiotemporally matched for the target area. m .

[0111] First, the FY-4A geostationary satellite L1 data is acquired by reading the observation data from the multi-channel scanning imaging radiometer. A lookup table is used to convert the raw digital signals into actual physical quantities such as reflectivity. The matching satellite observation positioning information, including the solar zenith angle and row and column numbers, is then read. This information is then combined with a full-disk row and column number conversion lookup table to obtain the satellite data's longitude and latitude. The satellite data is then cropped to capture the longitude and latitude of the target area, and problematic samples containing invalid or significantly outliers are removed. Furthermore, the acquired radar composite reflectivity data (spatial accuracy of 0.01° × 0.01°, array shape of 1001 × 1251) is processed for visualization, and data with significant radar clutter is removed.

[0112] Then, the quality-controlled data is time-space matched, and the radar reflectivity data are matched one by one to obtain satellite observation data with observation time differences within 3 minutes. Based on the nearest neighbor interpolation method, the satellite observation data are interpolated to the latitude and longitude grid of the radar observation. Thus, the radar satellite time-space matching dataset D is obtained. m .

[0113] S604. Determine the valid data grid point location and the start and end indexes of the sliding clipping window, slide-clip the satellite data and radar observation data, and filter according to the echo intensity to obtain strong convection samples. Each sample is matched with the corresponding terrain data to obtain the strong convection sample set D. p .

[0114] In order to pay more attention to the strong convection samples, the radar reflectivity ≥ 35dBZ is taken as the valid data and the D m For all radar data, grid points with a frequency of ≥35 dBZ and a cumulative frequency greater than 1000 are considered valid grid points.

[0115] To D m The variable data of each sample in the training set are cropped with a cropping window of 192×192 and a sliding distance of 38. The original shape of 1001×1251 will be cropped into 667 windows. The windows with invalid grid points are removed from the cropped windows, resulting in 214 training set windows.

[0116] According to the index position of the 214 training set windows, D m The radar data and satellite data are clipped sample by sample. For example, 214 radar reflectivity image blocks are obtained by cropping, and then the image blocks with radar reflectivity ≥ 30dBZ and the number of grid points less than 1% are deleted, and then the average radar reflectivity of the remaining image blocks is calculated. max (image block with the highest average radar reflectivity), and remove the image blocks containing P maxThe radar image block of the center grid point is then removed again from the remaining image blocks. Finally, the position of the remaining image blocks is Cut the corresponding satellite data and terrain altitude data to obtain The strong convection sample is obtained Severe convection samples.

[0117] S606. According to the set solar zenith angle maximum value threshold, the severe convection samples sampled in S604 are divided into a daytime sample set and a nighttime sample set, and are divided into a training set, a validation set, and a test set respectively.

[0118] Optionally, the severe convection sample set D p The maximum solar zenith angle less than or equal to 85° is classified as daytime samples, and the maximum solar zenith angle greater than 85° is classified as nighttime samples. For daytime samples and nighttime samples, those occurring on the 5th of each month are used as the test set, those occurring on the 10th of each month are used as the validation set, and samples on other days are divided into training sets.

[0119] S608. Construct a deep learning neural network, input the satellite feature channel data and corresponding terrain data selected from the daytime training set and the nighttime training set respectively, normalize the data and use the improved adaptive loss function to train the models separately, and select the daytime model with the best inversion effect score and night model

[0120] Optionally, a neural network structure is constructed by combining the U-Net model and the residual network. When training the night model, seven channel data, including 3.75μm (low), 6.25μm, 7.1μm, 8.5μm, 10.7μm, 12.0μm, and 13.5μm, as well as solar zenith angle and terrain height data, are input. When training the day model, four additional observation channels, 0.65μm, 1.375μm, 1.61μm, and 2.25μm, are added based on the night data input. After data normalization, the terrain height (in meters) less than or equal to -50 is set to -50, and then the terrain height is normalized. The model optimization training is performed separately using the adaptive loss function, and the model inversion score is calculated using the formula based on the validation set data, and the daytime model with the highest score is selected. and night model The test set data is used to calculate the HSS score for model evaluation.

[0121] S610. Obtain the characteristic channel data and corresponding terrain data of the regional geostationary satellite at the time to be inverted, and after the consistent preprocessing in S602, perform sliding window cutting to obtain block data x p and the start and end indexes of the cropping window I p .

[0122] Optionally, after variable reading and spatial interpolation of the satellite data according to the method of step S602, an array of N (number of satellite channels) × 1001 × 1251 is obtained. Together with the terrain height data, a 192 × 192 cropping window and a sliding distance of 38 are used to crop the data into 667 blocks x p and the start and end indexes of the cropping window I p .

[0123] S612. Get x from S610 p Normalize in accordance with the method described in S608, and select the value of the solar zenith angle obtained in S608 or Get the inverted radar reflectivity y′ of the block area p .

[0124] Optionally, get the x p Normalize the data in the image, and then input the block data with the original maximum solar zenith angle less than 85° into the image. Block data input greater than 85° Get the inverted radar reflectivity y′ of the block area p ;

[0125] S614. y′ obtained in S612 p Gaussian weight w G Weighted, then superimposed calculation is performed according to the start and end index of the window, and divided by the weight sum of the corresponding position to achieve regularization, and the final target area inversion radar reflectivity y′ is obtained fnl .

[0126] Optionally, a Gaussian weight w of size 192×192 is obtained based on a Gaussian filtering scheme. G , then set two all-zero arrays y′ of 1001×1251 all and w all Then, the 667 block inversion data are stacked and calculated from 1 to 667. Then, y′ all Divide by w all , and obtain the final target area inversion radar reflectivity y′ fnl .

[0127] Through the above steps, it is possible to realize intelligent inversion of radar reflectivity based on geostationary satellite data, improve the inversion accuracy in strong echo areas, expand the applicable time range of high-precision visible light and short-wave infrared channel data, and enhance the applicability of the inversion model under different geographical environment characteristics. By screening the strong convection sample training model, establishing an adaptive loss function, and supplementing the solar zenith angle and terrain height as characteristic channel data, the purpose of improving the inversion accuracy in strong echo areas, expanding the applicable time range of visible light channel information, and enhancing the adaptability to geographical environments is achieved, thereby achieving the technical effect of improving the accuracy and versatility of intelligent inversion of radar reflectivity, and thus solving the technical problem of the inability to accurately invert radar reflectivity due to the low versatility of existing radar reflectivity inversion methods based on satellite observations. Specifically, the method embodiment of the present application has the following advantages:

[0128] 1. In terms of sample screening, a strong convection sample screening strategy was developed. This effectively alleviates the sample imbalance problem while ensuring sample diversity, improves the inversion accuracy in strong echo areas, and helps save memory during model training and output.

[0129] 2. In terms of channel selection, the solar zenith angle θ was added to supplement the temporal information of the input features. This expands the applicable range of high-precision visible and shortwave infrared channel data from the traditional conditions of θ ≤ 65° or θ ≤ 70° to θ ≤ 85°, thus expanding the applicable temporal range of high-precision visible and shortwave infrared channel data. Furthermore, by adding normalized terrain data, the applicability of the inversion model to different terrain environments is enhanced.

[0130] 3. In terms of loss function, an adaptive weighted loss function is proposed by combining real-time prediction error and data probability distribution. This takes into account both small sample intervals and intervals with greater prediction difficulty, effectively improving inversion accuracy.

[0131] 4. In terms of network structure adaptation, for training strong samples after screening, there is no need to use the spatial attention mechanism, which simplifies the model network structure;

[0132] 5. In terms of model selection, combined with the HSS score, an evaluation formula was designed that balancedly considers no (weak) convection and strong convection. This can effectively screen models with better strong convection inversion effects while ensuring the no (weak) convection inversion effect.

[0133] 6. In terms of model output, the radar reflectivity splicing is optimized and the output accuracy is improved by setting the Gaussian weight code suitable for radar inversion splicing output.

[0134] The embodiment of the present application provides a radar reflectivity inversion device, Figure 7 It is a schematic diagram of the structure of the device. Figure 7As can be seen in the figure, the device includes: a matching module 70, which is used to match the historical satellite data corresponding to the historical radar reflectivity data based on the time information and geographical location information of the historical radar reflectivity data to obtain a first data set, wherein the first data set includes one-to-one corresponding historical radar reflectivity data and historical satellite data, and the historical satellite data includes solar zenith angle data and multi-channel imager data, and the multi-channel imager data is used to determine the band feature data; a screening module 72, which is used to screen the strong convection samples in the first data set, and use the screened strong convection samples as the second data set, wherein the strong convection samples are samples whose radar reflectivity data meets the preset intensity condition; a division module 74, which is used to divide the second data set into a first training set and a second training set based on the solar zenith angle data. Among them, the solar zenith angle of the historical satellite data in the first training set is less than or equal to the first preset threshold, and the solar zenith angle of the historical satellite data in the second training set is greater than the first preset threshold. The first training set includes solar zenith angle data, terrain height data, first band characteristic data and second band characteristic data. The second training set includes solar zenith angle data, terrain height data and second band characteristic data. The bands of the first band characteristic data include visible light band and shortwave infrared band, and the bands of the second band characteristic data include water vapor band and medium and long wave infrared band; the training module 76 is used to train the radar reflectivity prediction model based on the first training set and the second training set, wherein the radar reflectivity prediction model is used to invert the radar reflectivity under target conditions, and the target conditions include target time point and target area.

[0135] In some embodiments of the present application, the screening module 72 screens the severe convection samples in the first data set, including: cropping the radar reflectivity data in the first data set according to a preset window size to obtain a radar image block, wherein the radar image block includes a plurality of grid points; screening valid grid points according to preset screening rules, wherein the valid grid points are grid points whose radar reflectivity data meet the preset screening rules; determining a valid radar image block based on the valid grid points, wherein the valid radar image block only includes the valid grid points; determining a first valid radar image block with the largest average reflectivity from the valid radar image blocks, and removing a second valid radar image block from the first data set, wherein the second valid radar image block is a plurality of grid points; The method comprises the steps of: determining a new first valid radar image block from a third valid radar image block, wherein the third valid radar image block is a valid radar image block in the first dataset other than the first valid radar image block, removing a second valid radar image block corresponding to the new first valid radar image block from the first dataset, updating the third valid radar image block, and determining a new first valid radar image block from the new third valid radar image block, until no second valid radar image block corresponding to the new first valid radar image block exists in the first dataset.

[0136] In some embodiments of the present application, the screening module 72 screens valid grid points according to preset screening rules, including: counting the number of valid data for each grid point, where the number of valid data is the number of times the radar reflectivity is greater than a second preset threshold in the historical radar reflectivity data corresponding to the grid point; and determining the grid point whose number of valid data is greater than a third preset threshold as a valid grid point.

[0137] In some embodiments of the present application, before training the radar reflectivity prediction model based on the first training set and the second training set, the training module 76 also includes: normalizing the first band feature data and the second band feature data in the first training set; normalizing the second band feature data in the second training set; cosine encoding the solar zenith angle data in the first training set or the second training set, and normalizing the cosine-encoded solar zenith angle data; updating the value of the terrain height data in the first training set or the second training set that is lower than the fourth preset threshold to the fourth preset threshold, and normalizing the updated terrain height data.

[0138] In some embodiments of the present application, the radar reflectivity prediction model includes a first radar reflectivity prediction model and a second radar reflectivity prediction model, and the training module 76 trains the radar reflectivity prediction model based on the first training set and the second training set, including: inputting the first training set into the first radar reflectivity prediction model to be trained; inputting the second training set into the second radar reflectivity prediction model to be trained; training multiple radar reflectivity prediction models to be trained based on a weighted loss function to obtain multiple radar reflectivity prediction models to be evaluated, wherein the weights of the weighted loss function include a priori weights based on probability density distribution and error weights based on mean absolute error in training; determining the scores of the radar reflectivity prediction models to be evaluated based on preset scoring rules; determining the model with the highest score in the first radar reflectivity prediction model to be evaluated as the first radar reflectivity prediction model; and determining the model with the highest score in the second radar reflectivity prediction model to be evaluated as the second radar reflectivity prediction model.

[0139] In some embodiments of the present application, the training module 76 determines the score of the radar reflectivity prediction model to be evaluated based on a preset scoring rule, including: determining the category of the prediction result of the radar reflectivity prediction model to be evaluated based on a preset target interval range, wherein the categories include: the prediction result is in the target interval and the corresponding historical radar reflectivity is in the target interval, the prediction result is in the target interval and the corresponding historical radar reflectivity is not in the target interval, the prediction result is not in the target interval and the corresponding historical radar reflectivity is in the target interval, and the prediction result is not in the target interval and the corresponding historical radar reflectivity is not in the target interval; counting the number of prediction results in each category, and determining the first score corresponding to the radar reflectivity prediction model to be evaluated and the preset target interval according to a preset scoring formula; determining the average of the first scores corresponding to each preset target interval range as the score of the radar reflectivity prediction model to be evaluated.

[0140] In some embodiments of the present application, the weighted loss function is as follows: Among them, n b is the total number of grid points corresponding to the current training batch samples, y′ i and y i The predicted value corresponding to the i-th grid point and the true value corresponding to the i-th grid point, respectively, The current training batch samples are divided into n c +1 interval, is the total number of grid points corresponding to the k-th interval sample in the second data set, is the total number of grid points corresponding to the kth interval sample in the current training batch.

[0141] In some embodiments of the present application, the training module 76 determines the first score corresponding to the radar reflectivity prediction model to be evaluated and the preset target interval according to the following formula: Among them, HSS is the first score, N A N is the number of samples whose prediction results are in the target interval and whose corresponding historical radar reflectivity is in the target interval. B N is the number of samples whose prediction results are in the target interval and whose corresponding historical radar reflectivity is not in the target interval. C If the prediction result is not in the target interval and the corresponding historical radar reflectivity is in the target interval, N D is the number of samples whose prediction results are not in the target interval and whose corresponding historical radar reflectivity is not in the target interval.

[0142] In some embodiments of the present application, after the training module 76 trains the radar reflectivity prediction model based on the first training set and the second training set, the further steps include: determining target satellite data based on the target time point and the target area, wherein the difference between the observation time point of the target satellite data and the target time point is less than a preset interval, and the observation area of ​​the target satellite data is the same as the target area; cropping the target satellite data based on a preset window size to obtain target satellite data blocks and block indexes; classifying the target satellite data blocks based on solar zenith angle data of the target satellite data blocks to obtain a first target satellite data block and a second target satellite data block, wherein the solar zenith angle of the first target satellite data block is less than or equal to a first preset threshold, and the solar zenith angle of the second target satellite data block is greater than the first preset threshold; obtaining prediction results for the first target satellite data block and the second target satellite data block based on the radar reflectivity prediction model, wherein the prediction results include the radar reflectivity corresponding to the first target satellite data block or the second target satellite data block; and performing superposition calculation on each prediction result according to a preset calculation rule to determine the radar reflectivity corresponding to the target condition.

[0143] In some embodiments of the present application, performing a superposition calculation on each prediction result according to a preset calculation rule to determine the radar reflectivity corresponding to the target condition includes: multiplying each prediction result by a preset weight; initializing a first all-zero array of the same size as the target satellite data; adding the prediction result multiplied by the preset weight to the corresponding position of the first all-zero array according to a block index to obtain a first array; initializing a second all-zero array of the same size as the target satellite data; adding the preset weight to the corresponding position of the second all-zero array according to the block index to obtain a second array; dividing the first array by the second array, and determining the calculation result as the radar reflectivity corresponding to the target condition.

[0144] In some embodiments of the present application, the radar reflectivity prediction model includes a first radar reflectivity prediction model and a second radar reflectivity prediction model, the first radar reflectivity prediction model is obtained by training based on the first training set, and the second radar reflectivity prediction model is obtained by training based on the second training set; obtaining prediction results of the first target satellite data block and the second target satellite data block based on the radar reflectivity prediction model includes: inputting first feature data of the first target satellite data block into the first radar reflectivity prediction model to obtain a prediction result, wherein the first feature data includes solar zenith angle data, terrain height data, first band feature data and second band feature data, the bands of the first band feature data include visible light band and shortwave infrared band, and the bands of the second band feature data include water vapor band and medium and long wave infrared band; inputting second feature data of the second target satellite data block into the second radar reflectivity prediction model to obtain a prediction result, wherein the second feature data includes solar zenith angle data, terrain height data and second band feature data.

[0145] It should be noted that the various modules in the above-mentioned radar reflectivity inversion device can be program modules (for example, a set of program instructions that implement a certain specific function) or hardware modules. For the latter, it can be expressed in the following forms, but is not limited to this: the expression form of each of the above-mentioned modules is a processor, or the functions of each of the above-mentioned modules are implemented by a processor.

[0146] An embodiment of the present application provides a non-volatile storage medium, in which a program is stored, wherein when the program is running, the device where the non-volatile storage medium is located is controlled to execute the following radar reflectivity inversion method: according to the time information and geographic location information of the historical radar reflectivity data, the historical satellite data corresponding to the historical radar reflectivity data are matched to obtain a first data set, wherein the first data set includes one-to-one corresponding historical radar reflectivity data and historical satellite data, and the historical satellite data includes solar zenith angle data and multi-channel imager data, and the multi-channel imager data is used to determine the band characteristic data; the strong convection samples in the first data set are screened, and the strong convection samples obtained by screening are used as the second data set, wherein the strong convection samples are samples whose radar reflectivity data meet the preset intensity conditions; according to the solar zenith angle data, the strong convection samples are screened ... The second data set is divided into a first training set and a second training set, wherein the solar zenith angle of the historical satellite data in the first training set is less than or equal to a first preset threshold, and the solar zenith angle of the historical satellite data in the second training set is greater than the first preset threshold. The first training set includes solar zenith angle data, terrain height data, first band characteristic data and second band characteristic data, and the second training set includes solar zenith angle data, terrain height data and second band characteristic data. The bands of the first band characteristic data include visible light band and shortwave infrared band, and the bands of the second band characteristic data include water vapor band and medium and long wave infrared band. The radar reflectivity prediction model is trained based on the first training set and the second training set, wherein the radar reflectivity prediction model is used to invert the radar reflectivity under target conditions, and the target conditions include target time point and target area.

[0147] An embodiment of the present application provides an electronic device, comprising: a memory and a processor, the processor being configured to run a program stored in the memory, wherein the following radar reflectivity inversion method is executed when the program is run: historical satellite data corresponding to the historical radar reflectivity data is matched based on time information and geographic location information of historical radar reflectivity data to obtain a first data set, wherein the first data set includes one-to-one corresponding historical radar reflectivity data and historical satellite data, the historical satellite data includes solar zenith angle data and multi-channel imager data, and the multi-channel imager data is used to determine band characteristic data; strong convection samples in the first data set are screened, and the screened strong convection samples are used as a second data set, wherein the strong convection samples are samples whose radar reflectivity data meet a preset intensity condition; the first data set is matched based on the solar zenith angle data to obtain a first data set. The two data sets are divided into a first training set and a second training set, wherein the solar zenith angle of the historical satellite data in the first training set is less than or equal to a first preset threshold, and the solar zenith angle of the historical satellite data in the second training set is greater than the first preset threshold. The first training set includes solar zenith angle data, terrain height data, first band feature data and second band feature data, and the second training set includes solar zenith angle data, terrain height data and second band feature data. The bands of the first band feature data include visible light band and shortwave infrared band, and the bands of the second band feature data include water vapor band and medium and long wave infrared band; the radar reflectivity prediction model is trained based on the first training set and the second training set, wherein the radar reflectivity prediction model is used to invert the radar reflectivity under target conditions, and the target conditions include target time point and target area.

[0148] An embodiment of the present application provides a computer program product, including a computer program, which implements the following radar reflectivity inversion method when executed by a processor: matching historical satellite data corresponding to the historical radar reflectivity data based on time information and geographic location information of the historical radar reflectivity data to obtain a first data set, wherein the first data set includes one-to-one corresponding historical radar reflectivity data and historical satellite data, and the historical satellite data includes solar zenith angle data and multi-channel imager data, and the multi-channel imager data is used to determine band characteristic data; screening strong convection samples in the first data set, and using the screened strong convection samples as a second data set, wherein the strong convection samples are samples whose radar reflectivity data meets a preset intensity condition; dividing the second data set into A first training set and a second training set, wherein the solar zenith angle of the historical satellite data in the first training set is less than or equal to a first preset threshold, and the solar zenith angle of the historical satellite data in the second training set is greater than the first preset threshold, the first training set includes solar zenith angle data, terrain height data, first band feature data and second band feature data, the second training set includes solar zenith angle data, terrain height data and second band feature data, the bands of the first band feature data include visible light band and shortwave infrared band, and the bands of the second band feature data include water vapor band and medium and long wave infrared band; a radar reflectivity prediction model is trained based on the first training set and the second training set, wherein the radar reflectivity prediction model is used to invert radar reflectivity under target conditions, and the target conditions include target time point and target area.

[0149] In the above embodiments of the present application, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0150] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only exemplary. For example, the division of the units can be a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0151] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0152] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0153] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the relevant technology or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.

[0154] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A radar reflectivity inversion method, characterized in that: include: Matching historical satellite data corresponding to the historical radar reflectivity data based on time information and geographic location information of the historical radar reflectivity data to obtain a first data set, wherein the first data set includes the historical radar reflectivity data and the historical satellite data in a one-to-one correspondence, the historical satellite data includes solar zenith angle data and multi-channel imager data, and the multi-channel imager data is used to determine band characteristic data; screening the strong convection samples in the first data set, and using the screened strong convection samples as the second data set, wherein the strong convection samples are samples whose radar reflectivity data meet a preset intensity condition; The second data set is divided into a first training set and a second training set based on the solar zenith angle data, wherein the solar zenith angle of the historical satellite data in the first training set is less than or equal to a first preset threshold, and the solar zenith angle of the historical satellite data in the second training set is greater than the first preset threshold, the first training set includes solar zenith angle data, terrain height data, first band characteristic data and second band characteristic data, the second training set includes solar zenith angle data, terrain height data and second band characteristic data, the bands of the first band characteristic data include a visible light band and a shortwave infrared band, and the bands of the second band characteristic data include a water vapor band and a medium- and longwave infrared band; A radar reflectivity prediction model is trained based on the first training set and the second training set, wherein the radar reflectivity prediction model is used to invert radar reflectivity under target conditions, and the target conditions include a target time point and a target area.

2. The radar reflectivity inversion method according to claim 1, characterized in that: Screening the severe convection samples in the first dataset includes: cropping the radar reflectivity data in the first data set according to a preset window size to obtain a radar image block, wherein the radar image block includes a plurality of grid points; Screening valid grid points according to preset screening rules, wherein the valid grid points are grid points whose radar reflectivity data meet the preset screening rules; determining a valid radar image block according to the valid grid points, wherein the valid radar image block only includes the valid grid points; determining a first valid radar image block having a maximum average reflectivity from the valid radar image blocks, and removing a second valid radar image block from the first data set, wherein the second valid radar image block is the valid radar image block other than the first valid radar image block and including a central grid point of the first valid radar image block; A new first valid radar image block is again determined in a third valid radar image block, wherein the third valid radar image block is a valid radar image block in the first data set excluding the first valid radar image block; a second valid radar image block corresponding to the new first valid radar image block is removed from the first data set, the third valid radar image block is updated, and a new first valid radar image block is again determined in the new third valid radar image block, until no second valid radar image block corresponding to the new first valid radar image block exists in the first data set.

3. The radar reflectivity inversion method according to claim 2, characterized in that: Valid grid points are screened based on preset screening rules, including: Counting the number of valid data for each of the grid points, wherein the number of valid data is the number of times the radar reflectivity is greater than a second preset threshold in the historical radar reflectivity data corresponding to the grid point; The grid point having the number of valid data greater than a third preset threshold is determined as a valid grid point.

4. The radar reflectivity inversion method according to claim 1, characterized in that: Before training the radar reflectivity prediction model based on the first training set and the second training set, the method further includes: performing normalization processing on the first band feature data and the second band feature data in the first training set; performing normalization processing on the second band feature data in the second training set; performing cosine encoding on the solar zenith angle data in the first training set or the second training set, and performing normalization processing on the cosine-encoded solar zenith angle data; The values ​​of the terrain height data in the first training set or the second training set that are lower than a fourth preset threshold are updated to the fourth preset threshold, and the updated terrain height data are normalized.

5. The radar reflectivity inversion method according to claim 1, characterized in that: The radar reflectivity prediction model includes a first radar reflectivity prediction model and a second radar reflectivity prediction model, and training the radar reflectivity prediction model according to the first training set and the second training set includes: Inputting the first training set into a first radar reflectivity prediction model to be trained; inputting the second training set into a second radar reflectivity prediction model to be trained; Training a plurality of radar reflectivity prediction models to be trained based on a weighted loss function to obtain a plurality of radar reflectivity prediction models to be evaluated, wherein the weights of the weighted loss function include a priori weights based on a probability density distribution and an error weight based on a mean absolute error during training; Determining a score for the radar reflectivity prediction model to be evaluated according to a preset scoring rule; Determine the first radar reflectivity prediction model with the highest score among the first radar reflectivity prediction models to be evaluated as the first radar reflectivity prediction model; The one with the highest score among the second radar reflectivity prediction models to be evaluated is determined as the second radar reflectivity prediction model.

6. The radar reflectivity inversion method according to claim 5, characterized in that: Determining the score of the radar reflectivity prediction model to be evaluated according to a preset scoring rule includes: Determining the category of the prediction result of the radar reflectivity prediction model to be evaluated based on the preset target interval range, wherein the categories include: the prediction result is within the target interval and the corresponding historical radar reflectivity is within the target interval, the prediction result is within the target interval and the corresponding historical radar reflectivity is not within the target interval, the prediction result is not within the target interval and the corresponding historical radar reflectivity is within the target interval, and the prediction result is not within the target interval and the corresponding historical radar reflectivity is not within the target interval; Counting the number of prediction results in each category, and determining a first score corresponding to the radar reflectivity prediction model to be evaluated and the preset target interval according to a preset scoring formula; An average of the first scores corresponding to the preset target intervals is determined as the score of the radar reflectivity prediction model to be evaluated.

7. The radar reflectivity inversion method according to claim 5, characterized in that: The weighted loss function is as follows: Among them, n b is the total number of grid points corresponding to the current training batch samples, y i ′ and y i The predicted value corresponding to the i-th grid point and the true value corresponding to the i-th grid point, respectively, The current training batch samples are divided into n c +1 interval, is the total number of grid points corresponding to the k-th interval sample in the second data set, y i ∈C k , is the total number of grid points corresponding to the kth interval sample in the current training batch.

8. The radar reflectivity inversion method according to claim 1, characterized in that: After training the radar reflectivity prediction model based on the first training set and the second training set, the method further includes: determining target satellite data based on the target time point and the target area, wherein a difference between the observation time point of the target satellite data and the target time point is less than a preset interval, and the observation area of ​​the target satellite data is the same as the target area; The target satellite data is cropped according to a preset window size to obtain a target satellite data block and a block index; classifying the target satellite data block according to the solar zenith angle data of the target satellite data block to obtain a first target satellite data block and a second target satellite data block, wherein the solar zenith angle of the first target satellite data block is less than or equal to the first preset threshold, and the solar zenith angle of the second target satellite data block is greater than the first preset threshold; Obtaining prediction results of the first target satellite data block and the second target satellite data block according to the radar reflectivity prediction model, wherein the prediction results include radar reflectivities corresponding to the first target satellite data block or the second target satellite data block; The prediction results are superimposed and calculated according to a preset calculation rule to determine the radar reflectivity corresponding to the target condition.

9. The radar reflectivity inversion method according to claim 8, characterized in that: Performing a superposition calculation on each of the prediction results according to a preset calculation rule to determine the radar reflectivity corresponding to the target condition includes: Multiplying each of the prediction results by a preset weight; Initializing a first all-zero array having the same size as the target satellite data; Adding the prediction result multiplied by the preset weight to the corresponding position of the first all-zero array according to the block index to obtain a first array; Initializing a second all-zero array having the same size as the target satellite data; Adding the preset weight to the corresponding position of the second all-zero array according to the block index to obtain a second array; The first array is divided by the second array, and a calculation result is determined as a radar reflectivity corresponding to the target condition.

10. The radar reflectivity inversion method according to claim 8, characterized in that: The radar reflectivity prediction model includes a first radar reflectivity prediction model and a second radar reflectivity prediction model, wherein the first radar reflectivity prediction model is obtained by training based on a first training set, and the second radar reflectivity prediction model is obtained by training based on a second training set; Obtaining prediction results of the first target satellite data block and the second target satellite data block according to the radar reflectivity prediction model includes: Inputting first feature data of the first target satellite data block into a first radar reflectivity prediction model to obtain a prediction result, wherein the first feature data includes solar zenith angle data, terrain height data, first band feature data and second band feature data, the bands of the first band feature data include a visible light band and a shortwave infrared band, and the bands of the second band feature data include a water vapor band and a medium and longwave infrared band; Inputting second characteristic data of the second target satellite data block into a second radar reflectivity prediction model to obtain a prediction result, wherein the second characteristic data includes solar zenith angle data, terrain height data and second band characteristic data.

11. A radar reflectivity inversion device, characterized in that: include: a matching module, configured to match historical satellite data corresponding to the historical radar reflectivity data based on time information and geographic location information of the historical radar reflectivity data to obtain a first data set, wherein the first data set includes the historical radar reflectivity data and the historical satellite data in a one-to-one correspondence, the historical satellite data includes solar zenith angle data and multi-channel imager data, and the multi-channel imager data is used to determine band characteristic data; a screening module, configured to screen the strong convection samples in the first data set, and use the screened strong convection samples as the second data set, wherein the strong convection samples are samples whose radar reflectivity data meets a preset intensity condition; a partitioning module, configured to divide the second data set into a first training set and a second training set based on the solar zenith angle data, wherein the solar zenith angle of the historical satellite data in the first training set is less than or equal to a first preset threshold, and the solar zenith angle of the historical satellite data in the second training set is greater than the first preset threshold, the first training set includes solar zenith angle data, terrain height data, first band feature data and second band feature data, the second training set includes solar zenith angle data, terrain height data and second band feature data, the bands of the first band feature data include a visible light band and a shortwave infrared band, and the bands of the second band feature data include a water vapor band and a medium- and longwave infrared band; A training module is used to train a radar reflectivity prediction model based on the first training set and the second training set, wherein the radar reflectivity prediction model is used to invert the radar reflectivity under target conditions, and the target conditions include a target time point and a target area.

12. A non-volatile storage medium, characterized in that: The non-volatile storage medium stores a program, wherein when the program is running, the device where the non-volatile storage medium is located is controlled to execute the radar reflectivity inversion method according to any one of claims 1 to 10.

13. An electronic device, characterized in that: include: A memory and a processor, wherein the processor is configured to run a program stored in the memory, wherein the radar reflectivity inversion method according to any one of claims 1 to 10 is executed when the program is run.

14. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the radar reflectivity inversion method according to any one of claims 1 to 10 is implemented.