Method for accessing ceilometer radar data for WRF-GSI cloud analysis

Abstract

The application discloses a cloud-detecting radar data access method for WRF-GSI cloud analysis and relates to the technical field of numerical weather prediction observation data assimilation. The method comprises the following steps: collecting historical paired observation data of a cloud-detecting radar and a weather radar, and constructing a two-dimensional lookup table through grouped median statistics; for each assimilation time, reading a cloud-detecting radar CSV observation file, performing lookup table conversion through the lookup table to obtain pseudo-weather radar reflectivity factors, generating a pseudo-weather radar NetCDF file through gridding, vertically interpolating the pseudo-weather radar NetCDF file to a standard vertical layer number to obtain a mosaic NetCDF file, and converting the mosaic NetCDF file into a standard input BUFR file; modifying and compiling a GSI cloud analysis source code to enable a cloud analysis module, and after running configuration, assimilating the standard input BUFR file together with conventional observation data to perform cloud analysis on a numerical prediction initial field. The application realizes cloud-detecting radar observation access to WRF-GSI cloud analysis assimilation, and improves the cloud and water vapor analysis quality of an initial field.

Description

Technical Field

[0001] This invention belongs to the field of numerical weather prediction observation data assimilation technology, and in particular relates to a cloud radar data access method for WRF-GSI cloud analysis. Background Technology

[0002] The accuracy of numerical weather prediction (NWP) forecasts heavily relies on the accurate description of clouds and condensates in the initial field. The quality of cloud initialization directly affects the effectiveness of short-term precipitation forecasts, and assimilation of observational data is a core method for improving the quality of the initial cloud field. The gridded statistical interpolation system (GSI) is currently a widely used assimilation system in regional NWPs. Its built-in GSD cloud analysis module can use radar reflectivity mosaic data to correct for condensate variables such as cloud cover, cloud water, and cloud ice in the model background field, thereby improving the cloud structure description in the initial field.

[0003] Current GSI cloud analysis relies primarily on radar observations from operational weather radar networks, which operate in the S-band or C-band. Weather radars, limited by wavelength, are insensitive to weak echo cloud systems such as non-precipitating clouds, thin cirrus clouds, and stratus clouds, exhibiting significant shortcomings in detecting upper-level clouds and thin stratus clouds. Furthermore, operational weather radar networks have limited station coverage, with large blind spots in mountainous and marine areas, making it difficult to provide complete three-dimensional cloud structure information.

[0004] Millimeter-wave cloud-measuring radar operates at a much shorter wavelength than weather radar, is sensitive to cloud particle scattering, and can effectively detect non-precipitating clouds and thin-layer clouds. It also possesses high vertical resolution for detecting the three-dimensional structure of clouds, making it an important supplement to the limitations of weather radar observations. However, due to the wavelength difference, the reflectivity factor of millimeter-wave cloud-measuring radar differs significantly from that of weather radar in both magnitude and physical meaning, making them incomparable. This results in cloud-measuring radar observation data not being directly read and used by the GSI cloud analysis module. Furthermore, the storage format, coordinate system, and vertical hierarchy of the raw cloud-measuring radar observation data are incompatible with the specific BUFR format and standard vertical hierarchy structure required by GSI cloud analysis. Current technology lacks the ability to completely convert raw cloud-measuring radar observations into GSI-readable BUFR files; even after reflectivity factor conversion, the data still cannot enter the assimilation process.

[0005] In addition, the existing GSI cloud analytics module has code compatibility issues during actual deployment. When it is coupled with the WRF-ARW framework, the relevant source code needs to be specifically modified before it can be compiled and enabled normally, which makes the module unusable in actual applications and unable to support the assimilation and access of cloud radar data.

[0006] In summary, existing technologies cannot effectively integrate cloud radar observation data into the WRF-GSI cloud analysis and assimilation system, leaving this high-value, unconventional cloud radar observation idle in numerical forecast assimilation. Summary of the Invention

[0007] In view of this, the present invention aims to propose a cloud radar data access method for WRF-GSI cloud analysis, in order to solve the problem that cloud radar observation data cannot be effectively accessed into the WRF-GSI cloud analysis assimilation system in the prior art, specifically including the following three aspects: First, there is the issue of quantitative conversion of reflectivity factors from heterogeneous radars. Millimeter-wave cloud-measuring radars and operational weather radars have fundamentally different reflectivity factors in terms of magnitude and physical meaning due to their different operating wavelengths. They cannot be directly substituted for each other. Current technology lacks a systematic method for quantitatively converting the reflectivity factors of cloud-measuring radars into equivalent weather radar reflectivity factors, resulting in cloud-measuring radar observation data being unable to be recognized and utilized by the GSI cloud analysis module.

[0008] Secondly, there is the issue of adapting cloud-measuring radar data to the GSI standard input format. The GSI cloud analysis module requires radar observation data to be input in a specific BUFR format and a standard vertical hierarchy structure. However, the storage format, coordinate system, and vertical hierarchy of the raw cloud-measuring radar data do not match these requirements. Existing technology does not have the complete conversion capability from raw cloud-measuring radar observations to GSI-readable BUFR files.

[0009] Thirdly, there are engineering implementation issues with the GSI cloud analytics module within the WRF-ARW framework. In actual deployment, the existing GSI cloud analytics module's source code has several imperfections when coupled with the WRF-ARW framework, requiring specific modifications before it can be compiled and enabled correctly. This renders the module unusable in practical applications, unable to support the assimilation and access of cloud radar data.

[0010] This invention addresses the three issues mentioned above by establishing an end-to-end technical link from raw cloud radar observation data to WRF-GSI cloud analysis and assimilation operations, filling the technical gap in the application of cloud radar in the field of numerical weather prediction assimilation.

[0011] To achieve the above objectives, the technical solution of the present invention is implemented as follows: This invention provides a method for accessing cloud measurement radar data for WRF-GSI cloud analysis, including: Historical paired observation data of cloud-measuring radar and weather radar simultaneously detecting the same atmospheric area were collected. The data were grouped and statistically analyzed according to two dimensions: the reflectivity factor interval of cloud-measuring radar and the detection height layer interval. A number of cells were obtained. The median of the reflectivity factor of the weather radar corresponding to the paired samples in each cell was calculated to construct a two-dimensional lookup table. Cells with insufficient sample size were recorded as null values. For each assimilation cycle, the cloud-measuring radar CSV observation file is read, and the longitude, latitude, detection altitude, and reflectivity factor sequences of each observation point and altitude layer are extracted. A two-dimensional lookup table is loaded to convert the data and obtain the pseudo-weather radar reflectivity factors. Spatial gridding is then performed to generate a pseudo-weather radar NetCDF file. The pseudo-weather radar NetCDF file is vertically interpolated to the standard vertical layer number required by the radar mosaicking program to obtain a mosaicNetCDF file. Finally, the radar mosaicking program is called to convert the mosaicNetCDF file into a standard input BUFR file. Perform cloud analysis assimilation operation, including: modifying and compiling the GSI cloud analysis source code to enable the cloud analysis module, configuring GSI operation and assimilating the standard input BUFR file together with regular observation data, and performing cloud analysis on the initial field of numerical forecast.

[0012] Furthermore, statistical analysis was performed by grouping samples according to two dimensions: the range of cloud-measuring radar reflectivity factors and the range of detection altitude layers. This resulted in several cells. The median of the weather radar reflectivity factors corresponding to paired samples within each cell was calculated to construct a two-dimensional lookup table. Cells with insufficient sample size were recorded as null values, including: Observations less than or equal to -40 dBZ or equal to 0 dBZ are considered invalid and are discarded. The reflectivity factor of the cloud measuring radar is binned at preset intervals, and the detection height layer is binned in km units. Several cells are determined by binning in two dimensions. For all paired samples falling into each cell, the median of the corresponding weather radar reflectivity factor is calculated and used as the mapping value of that cell. A two-dimensional lookup table is constructed with cloud radar reflectivity bins as row indexes and height layer bins as column indexes. Cells with fewer than a preset threshold of paired samples are marked as null values, and the two-dimensional lookup table is saved in a structured form for online access.

[0013] Furthermore, a two-dimensional lookup table is loaded to obtain the pseudo-weather radar reflectivity factor, and spatial gridding is performed to generate a pseudo-weather radar NetCDF file, including: For each valid observation point, i.e. the observation points that are retained after removing observations containing values ​​less than or equal to -40dBZ or equal to 0dBZ, the reflectivity factor value is extracted from the reflectivity factor sequence layer by layer. The corresponding cell is located on the reflectivity bin axis and the height layer bin axis through binary search. The median stored in the cell is taken as the corresponding pseudo-weather radar reflectivity factor. If the cell is empty, it is replaced with a filler value of -999.0. Based on the actual detection range of the cloud-measuring radar, the output grid range is determined by extending outward by 0.5° in both the longitude and latitude directions. The horizontal resolution of the grid is 0.01°×0.01°, and 24 height layers are set in the vertical direction, ranging from 0.5km to 16.0km. The pseudo-weather radar reflectivity factor of each observation point is inserted into the nearest neighbor grid point to generate a three-dimensional pseudo-weather radar reflectivity field and output in NetCDF format to obtain the pseudo-weather radar NetCDF file.

[0014] Furthermore, the pseudo-weather radar NetCDF file is vertically interpolated to the standard number of vertical layers required by the radar mosaicking process to obtain the mosaic NetCDF file, including: The height coordinates of the 24 height layers were converted from km to m, and linear interpolation and extrapolation methods were used to process them to the target height of 31 layers, that is, adding 7 layers of 750m, 1250m, 1750m, 2250m, 2750m, 8500m and 18000m on the basis of the original 24 layers. Null values ​​appearing after interpolation are replaced with padding values, and the output variable dimension is a mosaicNetCDF file with longitude × latitude × vertical hierarchy; the padding values ​​are used by the GSI cloud analysis module to identify and skip missing measurement areas.

[0015] Furthermore, the radar mosaic processing program is invoked to convert the mosaicNetCDF file into a standard input BUFR file, including: Update the namelist configuration file required by the radar mosaic processing program, specifying the current assimilation time (analysis_time), the mosaic data path (dataPath), and the WRF background field file path (bkfile). The radar mosaic processing program process_NSSL_mosaic.x is called to convert the mosaicNetCDF file into a standard input BUFR file, and the standard input BUFR file is named NSSLRefInGSI.bufr.

[0016] Furthermore, the GSI cloud analytics source code was modified and compiled to enable the cloud analytics module, including: The files related to cloud analytics in the GSIv3.7 source code have been corrected, including uncommenting the cloudCover_radar subroutine call in gsdcloudanalysis.F90 to restore the call to the cloudCover_radar subroutine, and fixing code defects in rapidrefresh_cldsurf_mod.f90 and read_radarref_mosaic.f90 that caused compilation failures or runtime errors. Enable the cloud analytics module compilation option in CMakeLists.txt, and add the -DRR_CLOUDANALYSIS preprocessor macro during compilation to complete the compilation, so that the cloud analytics module can be compiled and enabled normally under the WRF-ARW framework.

[0017] Furthermore, GSI operation configuration is performed and the standard input BUFR file is assimilated along with regular observation data, including: Modify the GSI anavinfo configuration file to add the hydrophobic variable entry required by cloud analytics; Enable the cloud analytics switch in comgsi_namelist.sh; Adjust the assimilation time window parameters nhr_assimilation and min_offset according to the actual observation data to ensure that the analysis time corresponding to the standard input BUFR file falls within the assimilation time window; The script links the regular observation prepbufr file and the standard input BUFR file simultaneously, so that both are entered into the assimilation system at the same time.

[0018] Furthermore, the two-dimensional lookup table is trained separately for different stations and different models of cloud-measuring radars, and the historical paired observation data used for training are collected from the observation data of the corresponding stations and the corresponding models of cloud-measuring radars and weather radars that conduct simultaneous detection of the same atmospheric area.

[0019] Furthermore, the two-dimensional lookup table is saved in a structured form for online access, including: saving the two-dimensional lookup table as a structured file, database table, or key-value mapping, with the structured file named lookup_table.pkl.

[0020] Furthermore, historical paired observation data of cloud-measuring radar and weather radar simultaneously detecting the same atmospheric area are collected and a two-dimensional lookup table is constructed, which is an offline, one-time training step; the cloud-measuring radar CSV observation file is read and converted to obtain pseudo-weather radar NetCDF file, mosaicNetCDF file and standard input BUFR file in sequence, and cloud analysis assimilation is performed, which is an online execution step deployed in the assimilation operation; the training step and each online execution step are executed by the same computer device, or by multiple computer devices interconnected through data interfaces.

[0021] Compared with existing technologies, the cloud measurement radar data access method for WRF-GSI cloud analysis described in this invention has the following advantages: Firstly, it fills the gap in the application of cloud-measuring radar observations in the WRF-GSI assimilation system. By connecting the end-to-end three stages of lookup table construction, hierarchical format conversion, and assimilation operation, this invention overcomes the dual obstacles of the incomparability of cloud-measuring radar reflectivity factors and weather radar reflectivity factors in terms of magnitude and physical meaning, and the mismatch between the storage format, coordinate system, and vertical hierarchy of cloud-measuring radar raw data and the GSI standard input requirements. It realizes the effective utilization of cloud-measuring radar data that was previously unable to be accessed by the numerical weather prediction assimilation system, and expands the application scope of unconventional observation data in numerical weather prediction.

[0022] Secondly, it improved the quality of cloud and water vapor analysis in the initial field of numerical weather prediction. Through layer-by-layer lookup transformation and docking with the 31-layer standard vertical hierarchy, the fine vertical cloud structure information detected by the cloud-measuring radar was fully preserved and incorporated into GSI cloud analysis. This corrected for cloud cover, condensate, and relative humidity variables in the model background field, compensating for the lack of cloud structure information caused by the weather radar's insensitivity to weak echo clouds and blind spots in station coverage. Parallel control experiments further showed that the initial field 500hPa water vapor mixing ratio difference was concentrated in the cloud-measuring radar observation coverage area and was negative, with an order of approximately 2 × 10⁻⁶. -5 Up to 3.5×10 -5 The cloud density was approximately 100 kg / kg, and close to zero outside the coverage area, proving that the incremental cloud analysis originated from the introduction of cloud-measuring radar observations and was constrained by its spatial range. In the 24 hours before the forecast lead time, the relative humidity RMSE difference in the upper troposphere, around 500 to 100 hPa, especially around 200 to 300 hPa, remained negative for about 6 hours after the forecast began. This improvement was maintained in the 72-hour forecast lead time at around 100 to 300 hPa. During the same period, the RMSE of zonal and meridional winds in the upper troposphere also decreased in the corresponding time period and level, indicating that the accuracy of the initial field cloud structure description was effectively improved, which is conducive to the improvement of short-term precipitation forecasts and upper-level wind field forecasts.

[0023] Third, a transferable heterogeneous radar reflectivity factor conversion method is proposed. This method utilizes median statistics to suppress extreme value interference, uses null value marking to avoid insufficient sample regions, uses a fill value of -999.0 to mark missing measurements, and employs binary search to ensure efficient online table lookup. Furthermore, lookup tables are trained separately for different sites and different types of cloud-measuring radars. The method framework exhibits good versatility, adapting to data conversion needs under different observation environments, and is suitable for wider application.

[0024] Fourth, the engineering implementation of the GSI cloud analytics module was corrected and improved. Through a systematic review and targeted correction of the relevant source code of the GSI cloud analytics module, coupled with the enabling of compilation options and the -DRR_CLOUDANALYSIS preprocessing macro, as well as improvements to runtime configurations such as anavinfo, comgsi_namelist.sh, and assimilation time windows, the module can now be compiled and run normally under the WRF-ARW framework. This reduces the actual deployment difficulty of the module and provides a reliable system foundation for subsequent related research and business applications.

[0025] Fifth, the separation of offline training and online conversion balances conversion accuracy with business timeliness. The recalculation of statistical mapping relationships is completed in one go during the offline phase, while the online phase only performs lightweight operations such as table lookup, interpolation, and format encapsulation. Furthermore, the phases are decoupled by disk files, allowing any step to be independently reviewed and rerun, which is beneficial for the stable operation of the business system and problem localization. The parallel setup of the control and experimental groups, as well as the quantitative evaluation based on reanalysis data, provide attributable and verifiable experimental support for the technical effects. Attached Figure Description

[0026] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram of the overall technical route provided for an embodiment of the present invention; Figure 2 A schematic diagram of temperature and meteorological elements related to the RMSE difference between two sets of experiments 24 hours before the forecast lead time provided in this embodiment of the invention, as the vertical layer and forecast lead time change. Figure 3 A schematic diagram of the zonal wind meteorological elements in the RMSE difference between two sets of experiments 24 hours before the forecast lead time provided in this embodiment of the invention, as the vertical layer and forecast lead time change. Figure 4 A schematic diagram of meridional wind meteorological elements in the RMSE difference between two sets of experiments 24 hours before the forecast lead time provided in this embodiment of the invention, as the vertical layer and forecast lead time change. Figure 5 A schematic diagram of specific humidity meteorological elements in the RMSE difference between two sets of experiments 24 hours before the forecast lead time provided in this embodiment of the invention, as the vertical layer and forecast lead time change. Figure 6 A schematic diagram of the relative humidity meteorological elements in the changes of the RMSE difference between two sets of experiments 24 hours before the forecast lead time provided in this embodiment of the invention, as the vertical layer and forecast lead time change. Figure 7A schematic diagram of geopotential height meteorological elements in the RMSE difference between two sets of experiments 24 hours before the forecast lead time provided in this embodiment of the invention, as the vertical layer and forecast lead time change. Figure 8 A schematic diagram of temperature and meteorological elements related to the variation of the RMSE difference between two sets of experiments within a 72-hour forecast lead time provided in this embodiment of the invention, as well as the vertical layer and forecast lead time. Figure 9 A schematic diagram of zonal wind meteorological elements in the variation of the RMSE difference between two sets of experiments within a 72-hour forecast lead time provided in this embodiment of the invention, as well as the vertical layer and forecast lead time. Figure 10 A schematic diagram of meridional wind meteorological elements in the RMSE difference between two sets of experiments within a 72-hour forecast lead time provided in this embodiment of the invention, as the vertical layer and forecast lead time change. Figure 11 A schematic diagram of specific humidity meteorological elements in the variation of RMSE difference between two sets of experiments within a 72-hour forecast lead time provided in this embodiment of the invention, as well as the variation of vertical layer and forecast lead time. Figure 12 A schematic diagram of the meteorological elements related to relative humidity in the RMSE difference between two sets of experiments within a 72-hour forecast lead time provided in this embodiment of the invention, as the vertical layer and forecast lead time change. Figure 13 This is a schematic diagram of geopotential height meteorological elements in the RMSE difference between two sets of experiments within a 72-hour forecast lead time provided in this embodiment of the invention, as the vertical layer and forecast lead time change. Detailed Implementation

[0027] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0028] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0029] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0030] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0031] Unless otherwise defined, the technical terms and generic terms used herein have the following meanings: Cloud-measuring radar refers to millimeter-wave cloud-measuring radar, operating in the Ka-band or W-band. Its operating wavelength is much shorter than that of weather radar, making it sensitive to cloud particle scattering. It can effectively detect non-precipitating clouds and thin-layer clouds, and has high vertical resolution for detecting the three-dimensional structure of clouds. Weather radar refers to operational weather radar, operating in the S-band or C-band. Its reflectivity factor mosaic data is the radar observation source that the cloud analysis module routinely relies on. The reflectivity factor is measured in dBZ. GSI refers to a gridded statistical interpolation system. Its built-in cloud analysis module is the GSD cloud analysis module, which can use radar reflectivity mosaic data to correct for variables such as cloud cover, cloud water, and cloud ice in the model background field, thereby improving the initial field's cloud structure description. WRF-GSI refers to a regional numerical weather prediction assimilation system coupled with the mesoscale forecasting system WRF and the gridded statistical interpolation system GSI. WRF-ARW refers to an advanced research version framework of the mesoscale forecasting system WRF. BUFR refers to the binary universal meteorological data representation format, which is the standard observation input format required by the cloud analysis module. The standard input BUFR file referred to in this paper is a BUFR file that can be directly read by the GSI cloud analysis module, and its filename can be NSSLRefInGSI.bufr. NetCDF refers to the network universal data format. CSV refers to a comma-separated text file format. RMSE refers to the root mean square error; the definition of RMSE difference in different figures should be based on the corresponding descriptions of each figure. The prepbufr file refers to the conventional observation data file used for GSI assimilation. The control group referred to in this paper refers to the experiment that only assimilates conventional observation data, i.e., the prepbufr experiment; the experimental group referred to in this paper refers to the experiment that, in addition to the control group, introduces the standard input BUFR file obtained from cloud radar conversion and enables the cloud analysis module, i.e., the radar_prepbufr experiment. ERA5 reanalysis data refers to the fifth-generation atmospheric reanalysis data released by the European Centre for Medium-Range Weather Forecasts (ECMWF). Radiosonde observation refers to upper-air observations acquired by sensors carried by radiosonde balloons. NCEP refers to the National Center for Environmental Prediction (NCEP). GDAS refers to NCEP's Global Data Assimilation System. QVAPOR refers to the water vapor mixing ratio variable. hPa refers to hectopascals. The radar mosaic processing program referred to in this paper refers to the process_NSSL_mosaic.x program called by GSI cloud analysis. The two-dimensional lookup table referred to in this paper is simply called a lookup table. The system referred to in this paper refers to the cloud-measuring radar data access system used to execute the methods of the embodiments of this invention; the module referred to in this paper refers to the functional unit of the system, which can be implemented in software, hardware, or a combination of both.Unless otherwise specified, the above unified designations are used for the same subject throughout the text to avoid multiple names for one subject. The full English name of WRF-GSI is Weather Research and Forecasting - Gridpoint Statistical Interpolation, and its full Chinese name is Weather Research and Forecasting Model - Gridpoint Statistical Interpolation System.

[0032] The cloud detection radar data access method for WRF-GSI cloud analysis provided by the embodiments of the present invention may be executed by a computer device deployed with a regional numerical weather prediction assimilation system, and the computer device may be a server, a workstation or a cluster; the execution subject of each stage may be the same computer device, or may be completed by the cooperation of multiple computer devices interconnected through data interfaces. This method can be applied to the assimilation system of the WRF-ARW framework, and can also be applied to other regional assimilation systems with cloud analysis capabilities.

[0033] Embodiment 1 As shown Figure 1 in the figure, in the cloud detection radar data access method for WRF-GSI cloud analysis provided by the embodiment of the present invention, the overall technical route is divided into three sequentially connected stages, namely the stage of constructing the statistical conversion relationship of reflectivity factors, the stage of converting cloud detection radar data time by time, and the stage of WRF-GSI cloud analysis assimilation operation. The first stage is an offline training process, the second stage is an online processing process, and the third stage is an assimilation application and verification process. The lookup table constructed in the first stage, as the statistical conversion product of offline training, is input into the conversion process of the second stage, and serves as a coupling interface connecting offline training and online processing; the standard input BUFR file output in the second stage is input into the third stage as the observation input for assimilation operation; the third stage is concluded with the parallel assimilation operation and effect verification of the control group and the test group, forming a complete technical closed loop. Thus, Figure 1 the expressed end-to-end technical link is established, that is, starting from the original observation data of the cloud detection radar, through step-by-step conversion of physical magnitude and data format, the complete access of cloud detection radar data for WRF-GSI cloud analysis is finally realized, and the cloud structure of the numerical prediction initial field is improved. The following is a detailed description in combination with each step.

[0034] Stage 1: Construction of the statistical conversion relationship of reflectivity factors.

[0035] This stage is a one-time offline operation, and the purpose is to establish a quantitative mapping relationship between the cloud detection radar reflectivity factor and the weather radar reflectivity factor. As shown in the first stage of Figure 1 the figure, this stage includes three sequentially performed links: collecting historical data of cloud detection radars and weather radars, performing grouping statistics by reflectivity factor and detection height interval, and constructing a two-dimensional lookup table.

[0036] In the data collection phase, historical paired observation data from simultaneous observations of the same atmospheric region by cloud-measuring radar and weather radar were collected. Specifically, historical time periods where the observation periods of the two radars overlapped were selected in the temporal dimension, and atmospheric regions where their detection ranges overlapped in the spatial dimension. The cloud-measuring radar observations and weather radar observations within the same atmospheric region and time period were then correlated according to detection altitude to form a paired sample set. This paired sample set reflects the difference in echo response of the same cloud target under two radars operating at different wavelengths, and forms the factual basis for subsequent statistical mapping relationships.

[0037] In the grouping and statistical stage, paired data are grouped and statistically analyzed along two dimensions. The first dimension is the range of reflectivity factors from cloud-measuring radar, which is binned at certain intervals. The binning interval can be set according to the total historical sample size and statistical stability requirements. For example, the reflectivity factor range can be binned at 1 dBZ intervals. Before binning and statistical analysis, observations less than or equal to -40 dBZ or equal to 0 dBZ are considered invalid observations and are discarded to eliminate the contamination of the statistics by no-echo and noise echoes. The second dimension is the detection altitude range, which is binned in km units. For example, the altitude range can be binned at 0.5 km or 1 km intervals. For all paired samples falling within each cell jointly defined by the reflectivity range and altitude range, the median of the corresponding weather radar reflectivity factor is calculated, and this median is used as the mapping value for that cell. This constructs a lookup table with cloud-measuring radar reflectivity bins as row indexes and altitude range bins as column indexes. Cells with insufficient sample size, i.e., cells with fewer than a preset threshold of paired samples, are marked as null values, indicating that the cell lacks reliable statistical support. It should be noted that the median, rather than the mean, is used as the statistic in this step because paired samples inevitably contain extreme values ​​caused by strong echo cases, anomalous clutter, etc. The median is insensitive to extreme values, reducing the interference of extreme values ​​on the mapping relationship and improving the robustness of the transformation results.

[0038] In the lookup table construction and saving stage, the aforementioned lookup table is saved in a structured form for later use. For example, it can be saved as a structured file named lookup_table.pkl. In other embodiments, it can also be stored as a database table or key-value mapping for subsequent online processing. Since this stage is an offline, one-time operation, its statistical calculation cost does not occupy the assimilation time window, and the same lookup table can be repeatedly called multiple times during assimilation. Furthermore, for different sites and different models of cloud-measuring radar, paired data can be collected separately, and separate lookup tables can be trained. Therefore, this transformation relationship has good portability and versatility.

[0039] Phase Two: Hourly conversion of cloud radar data.

[0040] This phase is executed for each assimilation cycle, sequentially converting the raw cloud radar observation data into the standard input BUFR file required for GSI cloud analysis through three steps. For example... Figure 1 As shown in the second stage, the data format evolves sequentially from the cloud radar CSV observation file to the pseudo-weather radar NetCDF file, then to the mosaic NetCDF file, and finally to the standard input BUFR file. At the same time, the lookup table constructed in the first stage is input into the conversion process in step 1, which, together with the cloud radar CSV observation file, determines the generation of the pseudo-weather radar NetCDF file.

[0041] Step 1: Convert the cloud radar CSV observation file into a pseudo-weather radar NetCDF file.

[0042] The system reads the original CSV observation file from the cloud-measuring radar and extracts the longitude, latitude, detection altitude, and reflectivity factor sequences for each altitude layer of each observation point. For each valid observation point (i.e., the observation point retained after removing invalid observations), the reflectivity factor value of the cloud-measuring radar is extracted layer by layer. A lookup table pre-built in Phase 1 is loaded, and a binary search is used to locate the cell containing the observation value on the reflectivity binning axis and the altitude layer binning axis. The median value stored in that cell is taken as the corresponding pseudo-weather radar reflectivity factor. If the cell containing the value is empty (i.e., insufficient sample), it is filled with -999.0 to retain the missing measurement marker for downstream identification and processing. Using a binary search table for location allows for location in logarithmic time complexity, ensuring the timeliness of hourly online processing.

[0043] After completing the point-to-point lookup table conversion, spatial gridding is performed. Based on the actual detection range of the cloud-measuring radar, the output grid range is determined by extending outward by 0.5° in both the longitude and latitude directions, with a horizontal resolution of 0.01°×0.01°. In the vertical direction, 24 height layers are set from 0.5km to 16.0km, specifically 0.5, 1.0, 1.5, 2.0, 2.5, 3.0, 3.5, 4.0, 4.5, 5.0, 5.5, 6.0, 6.5, 7.0, 7.5, 8.0, 9.0, 10.0, 11.0, 12.0, 13.0, 14.0, 15.0, and 16.0km. The pseudo-weather radar reflectivity factor of each observation point is inserted into the nearest neighbor grid point. This involves assigning the pseudo-weather radar reflectivity factor of each observation point to the output grid point closest to its latitude and longitude location, generating a three-dimensional pseudo-weather radar reflectivity field, and outputting it in NetCDF format. The output file can be pseudo_radar.nc. In other embodiments, an inverse range-weighted average or other gridded assignment method can be used instead of nearest neighbor insertion, without departing from the concept of this embodiment. This file completes the organization from discrete observation points to a regular three-dimensional grid field and the conversion from cloud-measuring radar physical quantities to pseudo-weather radar physical quantities.

[0044] Step 2: Convert the fake weather radar NetCDF file to a mosaic NetCDF file.

[0045] The radar mosaicking processing program `process_NSSL_mosaic.x` requires input data with 31 standard vertical layers, while the pseudo-weather radar NetCDF file output in step 1 has 24 layers. Therefore, vertical interpolation expansion is necessary. Specifically, the 24-layer height coordinates are converted from km to m, and linear interpolation and extrapolation are used to process them to 31 target height layers. That is, seven layers are added to the original 24 layers: 750m, 1250m, 1750m, 2250m, 2750m, 8500m, and 18000m. Null values ​​after interpolation are replaced with a padding value of -999.0. The output is a mosaicNetCDF file that meets the format requirements. The file name can be mosaic.nc, and its variable dimension is longitude × latitude × vertical layer. In other embodiments, linear interpolation can also be implemented using piecewise linear interpolation or monotonic constrained interpolation, etc., without affecting the docking effect with the standard vertical layers. This step eliminates the mismatch between the cloud-measuring radar vertical layers and the GSI standard vertical layers.

[0046] Step 3: Convert the mosaicNetCDF file to a standard input BUFR file.

[0047] The process_NSSL_mosaic.x radar mosaic processing program is updated with its namelist configuration file, specifying the current assimilation time (analysis_time), mosaic data path (dataPath), and WRF background field file path (bkfile). The process_NSSL_mosaic.x program is then called to convert mosaic.nc into a standard input BUFR file. At this point, the raw cloud-measuring radar observations have completed adaptation in terms of physical magnitude, spatial organization, vertical hierarchy, and interface format. The obtained standard input BUFR file serves as the observation input for the third stage, thus completing the integration of cloud-measuring radar data for WRF-GSI cloud analysis.

[0048] It should also be noted that the fill value -999.0 is used consistently in this embodiment of the invention: from the lookup of null cells in step 1, to the interpolation of null values ​​in step 2, all the way to the BUFR encoding stage, the missing test information is always marked with -999.0, so that the GSI cloud analysis module can identify and skip the missing test area, avoid the missing test being mistakenly treated as zero echo in cloud analysis correction, and thus ensure the physical rationality of the correction result.

[0049] Phase 3: WRF-GSI cloud analytics assimilation operation.

[0050] like Figure 1 As shown in the third stage, this stage includes four steps in sequence: GSI cloud analysis code correction, GSI operation configuration, parallel assimilation operation of the control group and the experimental group, and effect verification, corresponding to steps 4 to 6.

[0051] Step 4: GSI cloud analytics source code correction and compilation.

[0052] Several files related to cloud analytics in the GSIv3.7 source code were specifically corrected, primarily involving gsdcloudanalysis.F90, rapidrefresh_cldsurf_mod.f90, and read_radarref_mosaic.f90. Specifically, the call to the cloudCover_radar subroutine in the original code was commented out and thus disabled. This was removed from gsdcloudanalysis.F90, restoring the call to the cloudCover_radar subroutine and allowing the radar cloud coverage analysis logic to execute correctly. Implementation issues (i.e., code defects that caused compilation failures or runtime errors) in rapidrefresh_cldsurf_mod.f90 and read_radarref_mosaic.f90 were corrected. After these corrections, the cloud analytics module compilation option was enabled in CMakeLists.txt, and the -DRR_CLOUDANALYSIS preprocessor macro was added during compilation. The compilation command can be performed as follows: first, generate the compilation configuration using `cmake .. / -DCMAKE_POLICY_VERSION_MINIMUM=3.5-DCMAKE_Fortran_FLAGS=-DRR_CLOUDANALYSIS`, and then complete the compilation using `make-j1`. After the above corrections and compilation, the GSI cloud analysis module can be compiled and enabled normally under the WRF-ARW framework, providing an engineering foundation for subsequent assimilation operations.

[0053] Step 5: GSI Run Configuration.

[0054] Modify the GSI's anavinfo configuration file to add entries for hydrophobic variables required for cloud analysis, defining the hydrophobic variables and their vertical structure identified and analyzed by the assimilation system; enable the cloud analysis switch in comgsi_namelist.sh to activate the cloud analysis module; adjust the assimilation time window parameters nhr_assimilation and min_offset according to the actual observation data to ensure that the analysis time corresponding to the cloud radar converted data falls within the assimilation time window, so that the observation and background field are correctly matched in time; link the regular observation prepbufr file and the standard input BUFR file obtained from the cloud radar conversion in the running script simultaneously, so that regular observations and cloud radar converted observations can enter the assimilation system at the same time.

[0055] Step 6: Assimilation process and effect verification.

[0056] Two parallel experiments were conducted. The control group assimilated only conventional prepbufr observation data; the experimental group, in addition to the control group, introduced a standard input BUFR file converted from cloud-measuring radar and enabled the GSI cloud analysis module. The two groups were identical in configuration except for the inclusion of cloud-measuring radar cloud analysis, ensuring that any experimental differences were attributable to the introduction of cloud-measuring radar cloud analysis. Using ERA5 reanalysis data or radiosonde observations as a reference, the root mean square error (RMSE) of relative humidity, cloud water, and other variables within the forecast lead time was calculated for both groups to evaluate the improvement effect of cloud-measuring radar cloud analysis on the initial field of numerical forecasting. The experimental results show that after introducing cloud-measuring radar cloud analysis, in areas where cloud analysis produces significant water vapor increments, the RMSE of relative humidity within the forecast lead time is significantly reduced compared to the control group, proving that the technical link established in this embodiment can effectively improve the quality of cloud and water vapor analysis in the initial field of numerical forecasting. In one embodiment, the above assimilation operation and effect verification can be performed with reference to the experimental setup, evaluation method, and experimental results described in Embodiment 4, and combined with… Figures 2 to 13 I understand.

[0057] To further clarify, Phase 1 is the offline training process, while Phases 2 and 3 are the online execution processes deployed in the assimilation business. Steps 4 and 5 belong to the system deployment and configuration phases, which can be completed in advance before or in parallel with Phase 2. The execution order of each stage and each step within each phase is based on the sequential dependencies in the text. Contents not explicitly defined in the text should not be interpreted as undue restrictions on the execution order.

[0058] Example 2 This invention also provides a cloud radar data access system, including a lookup table construction module, an hourly conversion module, and an assimilation and operation module. In one deployment configuration, the three modules are deployed as software functional modules on the same computer device; in another deployment configuration, the three modules are distributed across different computer devices, which are interconnected through data interfaces, and a lookup table and intermediate files serve as the data interfaces between the modules, together forming the system.

[0059] The lookup table construction module is used to collect historical paired observation data of cloud-measuring radar and weather radar simultaneously detecting the same atmospheric area. It is grouped and statistically analyzed according to two dimensions: reflectivity factor interval and detection height layer interval. The median of the reflectivity factor of the weather radar corresponding to the paired sample in each cell is calculated to construct the lookup table. Cells with insufficient sample size are marked as null values.

[0060] The time-by-time conversion module is used to read the cloud radar CSV observation file for each assimilation time, extract the reflectivity factors of each observation point at each altitude layer, load the lookup table to convert the pseudo-weather radar reflectivity factors, perform spatial gridding to generate a pseudo-weather radar NetCDF file, perform vertical interpolation to the standard vertical layer number to obtain a mosaic NetCDF file, and call the radar mosaic processing program to convert the mosaic NetCDF file into a standard input BUFR file.

[0061] The assimilation and execution module is used to modify and compile the GSI cloud analysis source code to enable the cloud analysis module, configure GSI execution, assimilate the standard input BUFR file and regular observation data together, and perform cloud analysis on the initial field of numerical forecast.

[0062] Furthermore, the system also includes a data storage unit for storing lookup tables, pseudo-weather radar NetCDF files, mosaic NetCDF files, and standard input BUFR files, so that the various stages are decoupled by disk files, and any stage can be independently reviewed and rerun. The specific implementation of each module is consistent with Embodiment 1, and will not be repeated here.

[0063] Example 3 This invention also provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of the method provided in Embodiment 1. This invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method provided in Embodiment 1. The aforementioned computer device and computer-readable storage medium serve as deployment carriers for the method of Embodiment 1 and the system of Embodiment 2, enabling the aforementioned methods and systems to be deployed in software form in numerical weather prediction operational systems or research platforms.

[0064] Example 4 To verify the technical effects of the embodiments of the present invention, this embodiment provides a set of parallel control experiments based on the method of Embodiment 1 and the system of Embodiment 2, and combines them with... Figures 2 to 13 This is explained below. It should be noted that the test setup and results in this embodiment are for illustrative purposes only and do not constitute a limitation on the scope of protection.

[0065] Regarding the experimental setup, a WRF-GSI coupled system was adopted, where WRF was used for atmospheric dynamic integration and physical process forecasting, and GSI was used for variational assimilation analysis of multi-source observation data. Taking the initial field reported from 00 UTC on May 30, 2025 as an example, two assimilation strategies were set up: a control group that assimilated only ground observation data (prepbufr experiment), and an experimental group that assimilated ground data plus cloud radar analysis (radar_prepbufr experiment). The analysis fields obtained from the two assimilation strategies were used as the initial fields, and 72-hour forecasts were performed respectively. The ground observation data was taken from the prepbufr file provided by GDAS, mainly including ground observations from land weather stations, ships, buoys, and airport reports, with elements such as air pressure, temperature, humidity, and wind. In this embodiment, the cloud radar observation data was generated by UAV-dropped data and converted into a standard input BUFR file using the method in Example 1 for assimilation by the experimental group. In terms of evaluation methods, firstly, for the initial field, the effective implementation of cloud analysis by the cloud-measuring radar is verified by comparing the differences in the analytical fields obtained by the two assimilation strategies; secondly, for the forecast field, the RMSE of each meteorological element in the forecast field obtained by the two assimilation strategies is calculated and the difference is made using ERA5 reanalysis data as the true reference field, to evaluate the improvement effect of cloud analysis by the cloud-measuring radar on the simulation effect of the forecast field.

[0066] Regarding the impact of cloud analysis on the initial field, the cloud-measuring radar does not directly assimilate dynamic and thermodynamic variables such as wind and temperature. Instead, it determines cloud formation through reflectivity. When observations indicate the presence of clouds, the cloud analysis module performs a humidification correction on the background field humidity, bringing the cloud area closer to saturation and thus correcting the water vapor mixing ratio. Variables such as wind and temperature are not directly corrected by this module; therefore, the difference between the two sets of analysis fields is primarily due to the increase in humidity. Taking a 500 hPa humidity field as an example, the difference between the two sets of analysis fields can be used as the initial 500 hPa water vapor mixing ratio QVAPOR. This difference is defined as the control group analysis field minus the experimental group analysis field. It can be seen that the difference does not appear uniformly throughout the entire model area but is concentrated in the area covered by the cloud-measuring radar; the difference in this area is negative, with an order of approximately 2 × 10⁻⁶. -5 Up to 3.5×10 -5 The result of kg / kg indicates that after introducing cloud analysis from the cloud-measuring radar, the water vapor mixing ratio in the corresponding area is higher than that in the experiment that only assimilated ground observations; the difference outside the coverage area is close to zero, indicating that the cloud analysis increment is constrained by the spatial range of radar observations and does not produce additional adjustments for areas without cloud-measuring radar data. The above results verify that the cloud analysis increment is indeed generated by the introduction of cloud-measuring radar observations, and its spatial distribution is consistent with the observation coverage area.

[0067] Regarding the impact of cloud analytics on forecast fields. Figures 2 to 7The figure shows the changes in the RMSE difference between the two experimental groups with vertical layer and forecast lead time 24 hours before the forecast lead time. The RMSE difference is defined as the experimental group minus the control group. Negative values ​​in the figure indicate that the RMSE of the experimental group is smaller, i.e., the forecast error is lower. The analysis range is limited to the area in the initial field where cloud analysis by cloud radar causes a significant increase. Figures 2 to 7 The six figures correspond to temperature, zonal wind, meridional wind, specific humidity, relative humidity, and geopotential height, respectively. The results show that the improvement in relative humidity is most significant among all meteorological elements: in the upper troposphere, around 500 to 100 hPa, especially around 200 to 300 hPa, the difference remains negative for approximately 6 hours after the initial forecast, indicating that cloud analysis corrections to the humidity field can reduce the forecast error of relative humidity during this period. Simultaneously, the RMSE of zonal and meridional winds in the upper troposphere also decreased, meaning that the initial field adjusted for humidity also has a positive effect on upper-level wind field forecasting.

[0068] Figures 8 to 13 This shows the variation of the RMSE difference between the two sets of experiments with vertical layers and forecast lead time over the entire 72-hour forecast lead time. The definition of the difference is the same as the analysis range. Figures 2 to 7 . Figures 8 to 13 The six graphs correspond to temperature, zonal wind, meridional wind, specific humidity, relative humidity, and geopotential height, respectively. Figures 8 to 13 It is evident that the improvement in relative humidity persists in the upper troposphere at approximately 100 to 300 hPa, indicating that the positive effect of cloud analysis on the humidity field is not limited to the short period following forecast initiation. Regarding the wind field, zonal winds show predominantly negative RMSE differences at approximately 24 to 72 hours and around 100 to 300 hPa, while meridional winds show similar differences at approximately 12 to 48 hours and around 100 to 300 hPa. These results demonstrate that the improvement in humidity field brought about by cloud analysis from cloud-measuring radar can be maintained along the forecast lead time and provides some improvement to the forecast of wind fields in the upper troposphere.

[0069] Based on the above experiments, this embodiment verifies that the technical link established in Embodiment 1 can effectively improve the quality of cloud and water vapor analysis in the initial field of numerical forecasts, and that the improvement effect has spatial locality consistent with the observation coverage and continuity along the forecast lead time.

[0070] It should be understood that the technical features of the above embodiments and alternative embodiments can be combined with each other without conflicting with each other, and the resulting technical solutions are also part of the disclosure of the embodiments of the present invention.

[0071] In conjunction with the above specific embodiments, the beneficial effects of the present invention include at least the following aspects.

[0072] Firstly, it fills the gap in the application of cloud-measuring radar observations in the WRF-GSI assimilation system. By connecting the end-to-end three stages of lookup table construction, hierarchical format conversion, and assimilation operation, this invention overcomes the dual obstacles of the incomparability of cloud-measuring radar reflectivity factors and weather radar reflectivity factors in terms of magnitude and physical meaning, and the mismatch between the storage format, coordinate system, and vertical hierarchy of cloud-measuring radar raw data and the GSI standard input requirements. It realizes the effective utilization of cloud-measuring radar data that was previously unable to be accessed by the numerical weather prediction assimilation system, and expands the application scope of unconventional observation data in numerical weather prediction.

[0073] Secondly, it improved the quality of cloud and water vapor analysis in the initial field of numerical weather prediction. Through layer-by-layer lookup transformation and docking with the 31-layer standard vertical hierarchy, the fine vertical cloud structure information detected by the cloud-measuring radar was fully preserved and incorporated into GSI cloud analysis. This corrected for cloud cover, condensate, and relative humidity variables in the model background field, compensating for the lack of cloud structure information caused by the weather radar's insensitivity to weak echo clouds and blind spots in station coverage. Parallel control experiments in Example 4 further demonstrated that the initial field 500hPa water vapor mixing ratio difference was concentrated in the cloud-measuring radar observation coverage area and was negative, with an order of approximately 2 × 10⁻⁶. -5 Up to 3.5×10 -5 The cloud density was approximately 100 kg / kg, and close to zero outside the coverage area, proving that the incremental cloud analysis originated from the introduction of cloud-measuring radar observations and was constrained by its spatial range. In the 24 hours before the forecast lead time, the relative humidity RMSE difference in the upper troposphere, around 500 to 100 hPa, especially around 200 to 300 hPa, remained negative for about 6 hours after the forecast began. This improvement was maintained in the 72-hour forecast lead time at around 100 to 300 hPa. During the same period, the RMSE of zonal and meridional winds in the upper troposphere also decreased in the corresponding time period and level, indicating that the accuracy of the initial field cloud structure description was effectively improved, which is conducive to the improvement of short-term precipitation forecasts and upper-level wind field forecasts.

[0074] Third, a transferable heterogeneous radar reflectivity factor conversion method is proposed. This method utilizes median statistics to suppress extreme value interference, uses null value marking to avoid insufficient sample regions, uses a fill value of -999.0 to mark missing measurements, and employs binary search to ensure efficient online table lookup. Furthermore, lookup tables are trained separately for different sites and different types of cloud-measuring radars. The method framework exhibits good versatility, adapting to data conversion needs under different observation environments, and is suitable for wider application.

[0075] Fourth, the engineering implementation of the GSI cloud analytics module was corrected and improved. Through a systematic review and targeted correction of the relevant source code of the GSI cloud analytics module, coupled with the enabling of compilation options and the -DRR_CLOUDANALYSIS preprocessing macro, as well as improvements to runtime configurations such as anavinfo, comgsi_namelist.sh, and assimilation time windows, the module can now be compiled and run normally under the WRF-ARW framework. This reduces the actual deployment difficulty of the module and provides a reliable system foundation for subsequent related research and business applications.

[0076] Fifth, the separation of offline training and online conversion balances conversion accuracy with business timeliness. The recalculation of statistical mapping relationships is completed in one go during the offline stage, while the online stage only performs lightweight operations such as table lookup, interpolation, and format encapsulation. Furthermore, the stages are decoupled by disk files, allowing any step to be independently reviewed and rerun, which is beneficial for the stable operation of the business system and problem localization. The parallel setup of the control and experimental groups, as well as the quantitative evaluation based on reanalysis data as described in Example 4, provide attributable and verifiable experimental support for the technical effects.

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

Claims

1. A cloud-measuring radar data access method for WRF-GSI cloud analysis, characterized in that, include: Historical paired observation data of cloud-measuring radar and weather radar simultaneously detecting the same atmospheric area were collected. The data were grouped and statistically analyzed according to two dimensions: the reflectivity factor interval of cloud-measuring radar and the detection height layer interval. A number of cells were obtained. The median of the reflectivity factor of the weather radar corresponding to the paired samples in each cell was calculated to construct a two-dimensional lookup table. Cells with insufficient sample size were recorded as null values. For each assimilation cycle, the cloud-measuring radar CSV observation file is read, and the longitude, latitude, detection altitude, and reflectivity factor sequences of each observation point and altitude layer are extracted. A two-dimensional lookup table is loaded to convert the data and obtain the pseudo-weather radar reflectivity factors. Spatial gridding is then performed to generate a pseudo-weather radar NetCDF file. The pseudo-weather radar NetCDF file is vertically interpolated to the standard vertical layer number required by the radar mosaicking program to obtain a mosaicNetCDF file. Finally, the radar mosaicking program is called to convert the mosaicNetCDF file into a standard input BUFR file. Perform cloud analysis assimilation operation, including: modifying and compiling the GSI cloud analysis source code to enable the cloud analysis module, configuring GSI operation and assimilating the standard input BUFR file together with regular observation data, and performing cloud analysis on the initial field of numerical forecast.

2. The cloud measurement radar data access method for WRF-GSI cloud analysis according to claim 1, characterized in that, The data is grouped and statistically analyzed along two dimensions: the range of reflectivity factors for cloud-measuring radar and the range of detection altitude layers. This results in a number of cells. The median of the weather radar reflectivity factor corresponding to paired samples within each cell is calculated to construct a two-dimensional lookup table. Cells with insufficient sample size are marked as null values, including: Observations less than or equal to -40 dBZ or equal to 0 dBZ are considered invalid and are discarded. The reflectivity factor of the cloud measuring radar is binned at preset intervals, and the detection height layer is binned in km units. Several cells are determined by binning in two dimensions. For all paired samples falling into each cell, the median of the corresponding weather radar reflectivity factor is calculated and used as the mapping value of that cell. A two-dimensional lookup table is constructed with cloud radar reflectivity bins as row indexes and height layer bins as column indexes. Cells with fewer than a preset threshold of paired samples are marked as null values, and the two-dimensional lookup table is saved in a structured form for online access.

3. The cloud measurement radar data access method for WRF-GSI cloud analysis according to claim 1, characterized in that, Loading a two-dimensional lookup table and performing a lookup transformation yields the pseudo-weather radar reflectivity factor. Spatial gridding is then performed to generate a pseudo-weather radar NetCDF file, including: For each valid observation point, i.e. the observation points that are retained after removing observations containing values ​​less than or equal to -40dBZ or equal to 0dBZ, the reflectivity factor value is extracted from the reflectivity factor sequence layer by layer. The corresponding cell is located on the reflectivity bin axis and the height layer bin axis through binary search. The median stored in the cell is taken as the corresponding pseudo-weather radar reflectivity factor. If the cell is empty, it is replaced with a filler value of -999.

0. Based on the actual detection range of the cloud-measuring radar, the output grid range is determined by extending outward by 0.5° in both the longitude and latitude directions. The horizontal resolution of the grid is 0.01°×0.01°, and 24 height layers are set in the vertical direction, ranging from 0.5km to 16.0km. The pseudo-weather radar reflectivity factor of each observation point is inserted into the nearest neighbor grid point to generate a three-dimensional pseudo-weather radar reflectivity field and output in NetCDF format to obtain the pseudo-weather radar NetCDF file.

4. The cloud measurement radar data access method for WRF-GSI cloud analysis according to claim 3, characterized in that, Vertical interpolation is performed on the pseudo-weather radar NetCDF file to the standard number of vertical layers required by the radar mosaicking process, resulting in a mosaicNetCDF file, including: The height coordinates of the 24 height layers were converted from km to m, and linear interpolation and extrapolation methods were used to process them to the target height of 31 layers, that is, adding 7 layers of 750m, 1250m, 1750m, 2250m, 2750m, 8500m and 18000m on the basis of the original 24 layers. Null values ​​appearing after interpolation are replaced with padding values, and the output variable dimension is a mosaicNetCDF file with longitude × latitude × vertical hierarchy; the padding values ​​are used by the GSI cloud analysis module to identify and skip missing measurement areas.

5. The cloud-measuring radar data access method for WRF-GSI cloud analysis according to claim 1, characterized in that, The radar mosaicking process converts the mosaicNetCDF file into a standard input BUFR file, including: Update the namelist configuration file required by the radar mosaic processing program, specifying the current assimilation time (analysis_time), the mosaic data path (dataPath), and the WRF background field file path (bkfile). The radar mosaic processing program process_NSSL_mosaic.x is called to convert the mosaicNetCDF file into a standard input BUFR file, and the standard input BUFR file is named NSSLRefInGSI.bufr.

6. The cloud-measuring radar data access method for WRF-GSI cloud analysis according to claim 1, characterized in that, Modify and compile the GSI cloud analytics source code to enable the cloud analytics module, including: The files related to cloud analytics in the GSIv3.7 source code have been corrected, including uncommenting the cloudCover_radar subroutine call in gsdcloudanalysis.F90 to restore the call to the cloudCover_radar subroutine, and fixing code defects in rapidrefresh_cldsurf_mod.f90 and read_radarref_mosaic.f90 that caused compilation failures or runtime errors. Enable the cloud analytics module compilation option in CMakeLists.txt, and add the -DRR_CLOUDANALYSIS preprocessor macro during compilation to complete the compilation, so that the cloud analytics module can be compiled and enabled normally under the WRF-ARW framework.

7. The cloud measurement radar data access method for WRF-GSI cloud analysis according to claim 1, characterized in that, Configure GSI operation and assimilate the standard input BUFR file along with regular observation data, including: Modify the GSI anavinfo configuration file to add the hydrophobic variable entry required by cloud analytics; Enable the cloud analytics switch in comgsi_namelist.sh; Adjust the assimilation time window parameters nhr_assimilation and min_offset according to the actual observation data to ensure that the analysis time corresponding to the standard input BUFR file falls within the assimilation time window; The script links the regular observation prepbufr file and the standard input BUFR file simultaneously, so that both are entered into the assimilation system at the same time.

8. The cloud measurement radar data access method for WRF-GSI cloud analysis according to claim 1, characterized in that, The two-dimensional lookup table was trained separately for different stations and different models of cloud-measuring radars. The historical paired observation data used for training were collected from the observation data of the corresponding stations and the corresponding models of cloud-measuring radars and weather radars that conducted simultaneous detection of the same atmospheric area.

9. The cloud-measuring radar data access method for WRF-GSI cloud analysis according to claim 2, characterized in that, The two-dimensional lookup table is saved in a structured form for online access, including: saving the two-dimensional lookup table as a structured file, database table, or key-value mapping. The structured file is named lookup_table.pkl.

10. The cloud-measuring radar data access method for WRF-GSI cloud analysis according to claim 1, characterized in that, Historical paired observation data from simultaneous observations of the same atmospheric region by cloud-measuring radar and weather radar are collected and a two-dimensional lookup table is constructed, which is an offline, one-time training step. The cloud-measuring radar CSV observation file is read and converted to obtain pseudo-weather radar NetCDF file, mosaicNetCDF file and standard input BUFR file in sequence, and cloud analysis assimilation is performed, which is an online execution step deployed in the assimilation operation. The training step and each online execution step are executed by the same computer device, or by multiple computer devices interconnected through data interfaces.

Images