A numerical prediction method and system for complex terrain areas based on FY4B

Abstract

The application provides a complex terrain area numerical prediction method and system based on FY4B, and belongs to the technical field of meteorological numerical prediction. The method first acquires FY4B satellite real-time observation, complex terrain area background field assistance and short-time prediction background field data; inputs full-channel brightness temperature and auxiliary data into a preset radiation transmission model screening channel to obtain channel subset brightness temperature data; combines multiple types of data to complete dynamic bias correction and acquire corrected brightness temperature data; fuses information such as a digital elevation model to carry out cloud-terrain quality control and obtain effective brightness temperature observation data; constructs a three-dimensional analysis field by relying on the effective brightness temperature observation data and a high-frequency cycle assimilation algorithm, and then carries out high-resolution prediction, and finally obtains complex terrain area numerical prediction results. The method improves satellite data utilization rate, strengthens initial field quality and improves the overall numerical weather prediction effect in complex terrain areas through multi-link data quality control and efficient assimilation processing.

Description

Technical Field

[0001] This invention belongs to the field of meteorological numerical forecasting technology, specifically relating to a numerical forecasting method and system for complex terrain regions based on FY4B. Background Technology

[0002] Complex topographic regions such as the Qinghai-Tibet Plateau are challenging areas for global numerical weather prediction. The dramatic topographic relief and extremely sparse conventional meteorological observation network in this region lead to a severe "hole effect" in the initial fields of numerical models, making it difficult to accurately depict the initiation and evolution of mesoscale weather systems such as plateau vortices and shear lines. Traditional methods mainly rely on temperature and humidity profile data from polar-orbiting meteorological satellites, but these satellites only pass over the target area 1-2 times per day, resulting in insufficient spatiotemporal resolution and an inability to capture convective systems with lifespans of only a few hours. This leads to short lead times and high false negative rates for severe weather events such as heavy precipitation. The GIIRS (Geospectral Infrared Observatory) on China's Fengyun-4B satellite (FY4B) offers advantages such as minute-level high-frequency observations and over a thousand channels of spectral resolution, providing new data support to fill the observation gaps in complex topographic regions. However, the complex underlying surface of the plateau (such as snow and bare soil) and frequent shallow cloud systems severely interfere with satellite radiation signals, and directly assimilating raw brightness and temperature data introduces significant errors. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a numerical weather prediction method and system for complex terrain areas based on FY4B, in order to solve the aforementioned problems. This method improves the utilization rate of satellite data and enhances the quality of the initial field through multi-stage data quality control and efficient assimilation processing, thereby improving the overall performance of numerical weather prediction in complex terrain areas.

[0004] To address the aforementioned technical problems, this invention provides a numerical prediction method for complex terrain areas based on FY4B, comprising the following steps: Acquire real-time observation datasets from FY4B satellites, auxiliary datasets of background fields in complex terrain areas, and short-term forecast background fields; The full-channel brightness temperature dataset and the background field auxiliary dataset of complex terrain areas in the real-time observation dataset of the FY4B satellite are input into the preset radiative transfer model so that the radiative transfer model can perform channel filtering based on the full-channel brightness temperature dataset and the background field auxiliary dataset of complex terrain areas to obtain a channel subset brightness temperature dataset. Based on the aforementioned channel subset brightness temperature dataset, the complex terrain area background field auxiliary dataset, the short-term forecast background field, and the satellite scan angle and cloud mask data in the FY4B satellite real-time observation dataset, dynamic bias correction is performed to obtain the corrected brightness temperature dataset. Based on the corrected brightness temperature dataset, the preset digital elevation model, and the cloud image data and satellite observation geometric information in the FY4B satellite real-time observation dataset, cloud-topography quality control is performed to obtain an effective brightness temperature observation dataset. A three-dimensional assimilation analysis field is constructed based on the effective brightness temperature observation dataset and the preset high-frequency cyclic assimilation algorithm. High-resolution forecasts are performed based on the assimilated three-dimensional analysis field to obtain numerical forecast results for complex terrain areas.

[0005] In the above scheme, the full-channel brightness temperature dataset from the FY4B satellite real-time observation dataset and auxiliary background field data for complex terrain areas are first input into a preset radiative transfer model for channel filtering to obtain a subset brightness temperature dataset, reducing noise interference from the underlying surface of complex terrain. Then, dynamic bias correction is performed by combining satellite scan angle and cloud mask data to obtain a corrected brightness temperature dataset, improving the accuracy of the observation data. Subsequently, cloud-topography quality control is implemented by integrating a preset digital elevation model, cloud image data, and satellite observation geometric information to eliminate invalid observations and form a valid brightness temperature observation dataset, ensuring data reliability. Based on this, a high-frequency cyclic assimilation algorithm is used to construct an assimilated three-dimensional analysis field, significantly optimizing the initial forecast field and effectively improving its spatiotemporal resolution and accuracy. Finally, high-resolution forecasts are conducted based on this optimized analysis field, resulting in accurate numerical weather prediction results for complex terrain areas. This scheme, through multi-stage data quality control and efficient assimilation processing, improves the utilization rate of satellite data, enhances the quality of the initial field, and comprehensively improves the numerical weather prediction effect for complex terrain areas.

[0006] Further, the full-channel brightness temperature dataset and the auxiliary dataset of the complex terrain background field from the FY4B satellite real-time observation dataset are input into a preset radiative transfer model, so that the radiative transfer model performs channel filtering based on the full-channel brightness temperature dataset and the auxiliary dataset of the complex terrain background field to obtain a channel subset brightness temperature dataset, including: In the preset radiative transfer model, the following steps are performed: Based on the background field auxiliary dataset of the complex terrain region, typical atmospheric profile data and land surface type data of the complex terrain region are obtained; For each channel brightness temperature data in the full-channel brightness temperature dataset, the weighted peak height of that channel is obtained based on the brightness temperature data of that channel and the typical atmospheric profile data of complex terrain areas. Peak values ​​are filtered based on the peak height of several channel weights and the corresponding brightness temperature data of several channels to obtain the first channel subset; Based on the full-channel brightness temperature dataset, typical atmospheric profile data of complex terrain areas, and surface type data, a surface emissivity sensitivity analysis was performed to obtain the second channel subset. The brightness temperature dataset of the channel subset is obtained by taking the intersection of the first channel subset and the second channel subset.

[0007] In the above scheme, based on a pre-set radiative transfer model, typical atmospheric profile data and surface type data of complex terrain areas are obtained using a background field auxiliary dataset. For the brightness temperature data of each channel in the FY4B satellite full-channel brightness temperature dataset, the channel weight peak height is calculated and filtered to obtain the first channel subset. Simultaneously, surface emissivity sensitivity analysis is conducted by combining the full-channel brightness temperature dataset, typical atmospheric profile data of complex terrain areas, and surface type data to obtain the second channel subset. By taking the intersection of the first and second channel subsets, the final channel subset brightness temperature dataset is obtained. This scheme can accurately eliminate observation channels significantly affected by the underlying surface of complex terrain, reduce noise interference from the underlying surface of complex terrain, ensure the effectiveness and stability of the channel subset brightness temperature data, and provide reliable data support for subsequent dynamic bias correction, quality control, and assimilation analysis, thereby improving the adaptability and utilization of satellite observation data in numerical forecasting of complex terrain areas.

[0008] Furthermore, based on the full-channel brightness temperature dataset, typical atmospheric profile data of complex terrain areas, and surface type data, surface emissivity sensitivity analysis is performed to obtain a second channel subset, including: For each channel brightness temperature data in the full-channel brightness temperature dataset, the basic emissivity value is determined based on the channel brightness temperature data and the surface type data. The first simulated brightness temperature is obtained based on the brightness temperature data of this channel and the basic emissivity value; The perturbation emission rate value is obtained based on the base emission rate value and the preset perturbation amplitude; The second simulated brightness temperature is obtained based on the brightness temperature data of this channel and the perturbation emissivity value; Based on the first simulated brightness temperature and the second simulated brightness temperature, obtain the absolute value of the brightness temperature change corresponding to the brightness temperature data of this channel; Based on a preset change threshold, several channel brightness temperature data and corresponding absolute values ​​of brightness temperature change, the channels are filtered to obtain the second channel subset.

[0009] In the above scheme, for the brightness temperature data of each channel in the full-channel brightness temperature dataset, the basic emissivity value is first determined based on the brightness temperature data of that channel and the surface type data. The first simulated brightness temperature corresponding to the basic emissivity value and the second simulated brightness temperature corresponding to the perturbed emissivity value obtained by adjusting the basic emissivity value by a preset perturbation amplitude are then obtained. Next, the absolute value of the brightness temperature change corresponding to the brightness temperature data of that channel is calculated using the first and second simulated brightness temperatures, thus characterizing the degree of influence of surface emissivity changes on the channel brightness temperature. Finally, channel selection is completed based on a preset change threshold to obtain a second subset of channels. This scheme can effectively select channels with weaker responses to surface emissivity changes, reduce simulation errors caused by surface emissivity fluctuations in complex terrain areas, ensure the accuracy and stability of the selected channel data, provide a reliable data foundation for subsequent satellite observation data processing, assimilation analysis, and numerical prediction, and improve the application effect of satellite data in numerical prediction in complex terrain areas.

[0010] Furthermore, based on the aforementioned channel subset brightness temperature dataset, the auxiliary dataset of the background field in complex terrain areas, the short-term forecast background field, and the satellite scan angle and cloud mask data in the FY4B satellite real-time observation dataset, dynamic bias correction is performed to obtain the corrected brightness temperature dataset, including: Land surface type data are obtained based on the background field auxiliary dataset of the complex terrain area; Based on the channel subset brightness temperature dataset, surface type data, and satellite scan angle simulation of theoretical brightness temperature, a theoretical brightness temperature dataset is obtained; The observation bias is calculated based on the theoretical brightness temperature dataset and the channel subset brightness temperature dataset to obtain the original observation bias dataset. A multiple linear regression model is constructed based on the original observation bias dataset, satellite scan angle, land surface type data, and cloud mask data. A systematic bias estimate is obtained based on the multiple linear regression model and the original observation bias dataset. Brightness temperature correction is performed based on the original observation bias dataset and the systematic bias estimate to obtain the corrected brightness temperature dataset.

[0011] In the above scheme, surface type data is obtained based on the background field auxiliary dataset for complex terrain areas. Theoretical brightness temperature dataset is then obtained by simulating theoretical brightness temperature using a channel subset brightness temperature dataset, surface type data, and satellite scan angle. The observation bias is calculated using the theoretical brightness temperature dataset and the channel subset brightness temperature dataset to obtain the original observation bias dataset. A multiple linear regression model is then constructed using the original observation bias dataset, satellite scan angle, surface type data, and cloud mask data to obtain accurate estimates of systematic bias. Finally, brightness temperature is corrected based on the original observation bias dataset and the systematic bias estimates to obtain the corrected brightness temperature dataset. This scheme is well-suited to the observation environment of complex terrain areas, effectively eliminating systematic observation bias caused by satellite scan angle, surface type, and cloud mask factors, significantly improving the accuracy and reliability of brightness temperature data. It provides high-quality observation data for subsequent cloud-terrain quality control and high-frequency cyclic assimilation, laying the foundation for optimizing the initial field accuracy of numerical forecasts in complex terrain areas.

[0012] Furthermore, the cloud-topography quality control based on the corrected brightness temperature dataset, the preset digital elevation model, and the cloud image data and satellite observation geometric information in the FY4B satellite real-time observation dataset, to obtain an effective brightness temperature observation dataset, includes: Thick cloud identification is performed based on the cloud map data and a preset multispectral threshold algorithm to obtain thick cloud detection results; Based on the cloud map data and the preset local binary pattern texture analysis algorithm, broken and shallow clouds are identified to obtain broken and shallow cloud detection results. Cloud contamination labeling results are obtained based on the thick cloud detection results and the broken, shallow cloud detection results. Based on the satellite observation geometric information and the preset digital elevation model, terrain occlusion is identified to obtain terrain occlusion marking results; Based on the cloud pollution labeling results and terrain occlusion labeling results, data is removed from the corrected brightness temperature dataset to obtain the effective brightness temperature observation dataset.

[0013] In the above scheme, based on cloud image data from the FY4B satellite real-time observation dataset, thick cloud detection results are obtained by using a preset multispectral threshold algorithm for thick cloud identification, and broken and thin cloud detection results are obtained by using a preset local binary mode texture analysis algorithm for broken and thin clouds. Cloud contamination labeling results are obtained based on the two types of detection results. Simultaneously, terrain occlusion identification is performed by combining satellite observation geometric information and a preset digital elevation model to obtain terrain occlusion labeling results. On this basis, data is removed from the corrected brightness temperature dataset according to the cloud contamination labeling results and terrain occlusion labeling results, finally obtaining the effective brightness temperature observation dataset. The above scheme can accurately eliminate invalid observations caused by cloud contamination and terrain occlusion, ensuring the reliability and applicability of observation data, providing high-quality observation data for subsequent construction of assimilated three-dimensional analysis fields, improving the application quality of satellite data in complex terrain areas, and laying a stable data foundation for high-resolution numerical weather prediction.

[0014] Further, the data removal process based on the cloud contamination labeling results and terrain occlusion labeling results to obtain the effective brightness temperature observation dataset includes: The satellite zenith angle is obtained based on the aforementioned satellite observation geometric information; Based on the satellite zenith angle and the preset zenith angle threshold, the modified brightness temperature dataset is initially purged to obtain the initial brightness temperature observation dataset. Based on the cloud pollution labeling results and terrain occlusion labeling results, the initial brightness temperature observation dataset is purged to obtain the effective brightness temperature observation dataset.

[0015] In the above scheme, the satellite zenith angle is obtained based on satellite observation geometric information. The modified brightness temperature dataset is then initially purged based on the satellite zenith angle and a preset zenith angle threshold to obtain an initial brightness temperature observation dataset. Further data purging is performed on the initial brightness temperature observation dataset based on cloud contamination and terrain occlusion markers, ultimately yielding a valid brightness temperature observation dataset. This scheme employs a hierarchical and progressive filtering logic. First, abnormal data with poor observation geometry is excluded using the zenith angle threshold. Then, invalid observations are removed by combining cloud contamination and terrain occlusion markers. This achieves multi-dimensional and multi-level data filtering, effectively improving the validity and reliability of the brightness temperature observation data. It provides high-quality observation data for the subsequent construction of an assimilated three-dimensional analysis field using a high-frequency cyclic assimilation algorithm, ensuring the accuracy of the initial field construction for numerical forecasting in complex terrain areas and providing reliable data support for subsequent high-resolution forecasts.

[0016] Furthermore, based on the effective brightness temperature observation dataset and the preset high-frequency cyclic assimilation algorithm, an assimilation three-dimensional analysis field is constructed, including: Get the default observation error covariance matrix; For each observation point in the effective brightness temperature observation dataset, cloud cover data, terrain complexity index, and satellite zenith angle are obtained based on the observation point. The observation error adjustment factor is obtained based on the cloud cover data, terrain complexity index, and satellite zenith angle. The observation error covariance matrix corresponding to the observation point is obtained based on the observation error adjustment factor and the default observation error covariance matrix. Assimilation analysis is performed based on the preset high-frequency cyclic assimilation algorithm, the effective brightness temperature observation dataset, and the observation error covariance matrix corresponding to several observation points to obtain the assimilation three-dimensional analysis field.

[0017] In the above scheme, a default observation error covariance matrix is ​​first obtained. For each observation point in the effective brightness temperature observation dataset, the cloud cover data, terrain complexity index, and satellite zenith angle corresponding to that observation point are obtained, and an observation error adjustment factor is derived based on this information. Then, the observation error adjustment factor is combined with the default observation error covariance matrix to obtain the observation error covariance matrix corresponding to that observation point. Subsequently, based on a preset high-frequency cyclic assimilation algorithm, assimilation analysis is performed using the effective brightness temperature observation dataset and the observation error covariance matrices corresponding to each observation point to construct an assimilated three-dimensional analysis field. This scheme can adaptively match observation errors by combining cloud cover, terrain complexity, and satellite observation geometry, achieving fine-grained control of observation data errors, improving the reliability of assimilation analysis, and optimizing the accuracy and quality of the assimilated three-dimensional analysis field. This lays a solid foundation for subsequent high-resolution forecasting of complex terrain areas and effectively enhances the application effect of satellite data in numerical weather prediction.

[0018] Furthermore, the high-resolution forecasting based on the assimilated three-dimensional analysis field to obtain numerical forecasting results for complex terrain areas includes: Acquire large-scale boundary condition data and conventional initial fields; High-resolution forecasts are performed based on the large-scale boundary condition data, the assimilated three-dimensional analysis field, and the preset regional numerical forecasting model to obtain high-precision forecast results. High-resolution forecasts are performed based on the large-scale boundary condition data, conventional initial fields, and preset regional numerical forecast models to obtain conventional forecast results. Numerical forecast results for the complex terrain region are obtained based on the high-precision forecast results and the conventional forecast results.

[0019] In the above scheme, large-scale boundary condition data and a conventional initial field are acquired. High-resolution forecasts are then performed based on the assimilated 3D analysis field, large-scale boundary condition data, and a pre-set regional numerical weather prediction model, yielding high-precision forecast results. Simultaneously, high-resolution forecasts are also performed based on the conventional initial field, large-scale boundary condition data, and the pre-set regional numerical weather prediction model, yielding conventional forecast results. Finally, numerical weather prediction results for complex terrain areas are obtained based on the high-precision forecast results and the conventional forecast results. This scheme employs consistent large-scale boundary conditions and a regional numerical weather prediction model, using only the initial field as the sole variable. This allows for a direct demonstration of the improvement effect of the assimilated 3D analysis field on the forecast results, ensuring the accuracy and comparability of numerical weather prediction results for complex terrain areas. It fully leverages the effect of satellite data assimilation and optimization of the initial field, effectively improving the precision of numerical weather predictions for complex terrain areas and providing more reliable data support for meteorological forecasts in these regions.

[0020] Furthermore, the numerical prediction method for complex terrain areas based on FY4B, as described above, also includes: Acquire real-time observation data, and obtain high-precision forecast results and conventional forecast results based on the numerical forecast results of the complex terrain area; Based on the high-precision forecast results, conventional forecast results, and actual observation data, traditional skill scoring is performed to obtain high-precision forecast scoring results and conventional forecast scoring results; Based on the high-precision forecast scoring results and the conventional forecast scoring results, a significant difference analysis was conducted to obtain the key areas for forecast improvement and the key forecast lead time. Based on the forecast improvement key areas and key forecast lead times, high-precision three-dimensional meteorological element fields corresponding to the regions and lead times are extracted from the high-precision forecast results; Based on the forecast improvement key areas and key forecast lead times, the conventional three-dimensional meteorological element fields corresponding to the regions and lead times are extracted from the high-precision forecast results; Based on the high-precision three-dimensional meteorological element field and the conventional three-dimensional meteorological element field, multi-physical quantity collaborative diagnosis is performed to obtain physical mechanism diagnosis results.

[0021] In the above scheme, real-time observation data is acquired, and high-precision forecast results and conventional forecast results are extracted based on numerical forecast results for complex terrain areas. High-precision forecast scores and conventional forecast scores are then calculated using traditional skill scoring methods. A significant difference analysis is conducted based on the two types of forecast scores to identify key areas and key forecast lead times for forecast improvement. Based on these key areas and key forecast lead times, high-precision three-dimensional meteorological element fields for the corresponding areas and lead times are extracted from the high-precision forecast results, and conventional three-dimensional meteorological element fields for the corresponding areas and lead times are extracted from the conventional forecast results. Finally, multi-physical quantity collaborative diagnosis is performed based on the high-precision and conventional three-dimensional meteorological element fields to obtain physical mechanism diagnosis results. This scheme can accurately locate the key spatiotemporal range for forecast improvement, clarify the intrinsic mechanism of forecast optimization through multi-physical quantity collaborative diagnosis, provide objective basis for the iterative optimization of numerical forecasting methods for complex terrain areas based on the FY4B satellite, and further enhance the scientific rigor and practicality of the forecasting method.

[0022] This invention also provides a numerical prediction system for complex terrain areas based on FY4B, comprising: The data acquisition module is used to acquire the real-time observation dataset of FY4B satellite, the auxiliary dataset of background field in complex terrain areas, and the short-term forecast background field; The channel filtering module is used to input the full-channel brightness temperature dataset and the background field auxiliary dataset of complex terrain areas from the real-time observation dataset of the FY4B satellite into the preset radiative transfer model, so that the radiative transfer model can perform channel filtering based on the full-channel brightness temperature dataset and the background field auxiliary dataset of complex terrain areas to obtain a channel subset brightness temperature dataset. The dynamic deviation correction module is used to perform dynamic deviation correction based on the channel subset brightness temperature dataset, the background field auxiliary dataset of complex terrain area, the short-term forecast background field, and the satellite scan angle and cloud mask data in the FY4B satellite real-time observation dataset, to obtain the corrected brightness temperature dataset. The quality control module is used to perform cloud-topography quality control based on the corrected brightness temperature dataset, the preset digital elevation model, and the cloud image data and satellite observation geometric information in the FY4B satellite real-time observation dataset, so as to obtain an effective brightness temperature observation dataset. The assimilation analysis module is used to construct an assimilation three-dimensional analysis field based on the effective brightness temperature observation dataset and a preset high-frequency cyclic assimilation algorithm; The numerical prediction module is used to perform high-resolution prediction based on the assimilated three-dimensional analysis field to obtain numerical prediction results for complex terrain areas.

[0023] In the above scheme, the FY4B-based numerical weather prediction system for complex terrain areas acquires real-time FY4B satellite observation datasets, auxiliary datasets for the background field of complex terrain areas, and short-term forecast background fields through a data acquisition module. A channel selection module uses a preset radiative transfer model to filter the full-channel brightness temperature dataset, obtaining a subset brightness temperature dataset. A dynamic bias correction module combines multiple types of data with satellite scan angle and cloud mask data to perform dynamic bias correction, obtaining a corrected brightness temperature dataset. A quality control module performs cloud-topography quality control based on a digital elevation model, cloud image data, and satellite observation geometry information, obtaining an effective brightness temperature observation dataset. An assimilation analysis module constructs an assimilation three-dimensional analysis field based on the effective brightness temperature observation dataset and a preset high-frequency cyclic assimilation algorithm. The numerical weather prediction module performs high-resolution forecasts based on this analysis field, obtaining numerical weather prediction results for complex terrain areas. The various modules of the system are interconnected and operate collaboratively. Through full-process data quality control and efficient assimilation processing, the system improves the utilization rate of satellite data, enhances the quality of the initial field, and comprehensively improves the accuracy and reliability of numerical weather prediction for complex terrain areas. Attached Figure Description

[0024] Figure 1 A schematic diagram of a numerical prediction method for complex terrain areas based on FY4B provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a numerical prediction system architecture for complex terrain regions based on FY4B, provided as an embodiment of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] It should be noted that the complex terrain areas described in this invention include, but are not limited to, areas with large topographic relief and strong underlying surface heterogeneity, such as the Qinghai-Tibet Plateau, mountains, and areas where plateaus and basins intersect. The technical solution of this invention can be adapted to numerical forecasting scenarios for all areas with such terrain characteristics, and is not limited to a specific complex terrain area.

[0027] Please see Figure 1 This embodiment provides a numerical prediction method for complex terrain areas based on FY4B, including the following steps: Step S1: Obtain the FY4B satellite real-time observation dataset, the background field auxiliary dataset for complex terrain areas, and the short-term forecast background field; Step S2: Input the full-channel brightness temperature dataset and the background field auxiliary dataset of complex terrain area from the real-time observation dataset of FY4B satellite into the preset radiative transfer model, so that the radiative transfer model can perform channel filtering based on the full-channel brightness temperature dataset and the background field auxiliary dataset of complex terrain area to obtain the channel subset brightness temperature dataset. Step S3: Based on the channel subset brightness temperature dataset, the complex terrain background field auxiliary dataset, the short-term forecast background field, and the satellite scan angle and cloud mask data in the FY4B satellite real-time observation dataset, perform dynamic bias correction to obtain the corrected brightness temperature dataset; Step S4: Based on the corrected brightness temperature dataset, the preset digital elevation model, and the cloud image data and satellite observation geometric information in the FY4B satellite real-time observation dataset, cloud-topography quality control is performed to obtain an effective brightness temperature observation dataset; Step S5: Construct an assimilation three-dimensional analysis field based on the effective brightness temperature observation dataset and the preset high-frequency cyclic assimilation algorithm; Step S6: Perform high-resolution forecasting based on the assimilated three-dimensional analysis field to obtain numerical forecasting results for complex terrain areas.

[0028] In this embodiment, the full-channel brightness temperature dataset from the FY4B satellite real-time observation dataset and auxiliary background field data for complex terrain areas are first input into a preset radiative transfer model for channel filtering to obtain a subset brightness temperature dataset, reducing noise interference from the underlying surface of complex terrain. Then, dynamic bias correction is performed by combining satellite scan angle and cloud mask data to obtain a corrected brightness temperature dataset, improving the accuracy of the observation data. Subsequently, cloud-topography quality control is implemented by fusing a preset digital elevation model, cloud image data, and satellite observation geometric information to eliminate invalid observations and form a valid brightness temperature observation dataset, ensuring data reliability. Based on this, a high-frequency cyclic assimilation algorithm is used to construct an assimilated three-dimensional analysis field, significantly optimizing the initial forecast field and effectively improving its spatiotemporal resolution and accuracy. Finally, high-resolution forecasts are conducted based on this optimized analysis field, resulting in accurate numerical weather prediction results for complex terrain areas. This embodiment improves satellite data utilization, enhances initial field quality, and comprehensively improves the numerical weather prediction effect for complex terrain areas through multi-stage data quality control and efficient assimilation processing.

[0029] It should be noted that the acquisition methods of the FY4B satellite real-time observation dataset described in this invention include real-time satellite downlink reception and retrieval from the satellite data archive. The temporal resolution of the dataset can be adjusted according to forecast requirements. Furthermore, the technical steps of this invention can be parallelized based on a high-performance computing cluster. Real-time data interaction and feedback are supported between each step. The execution order of the steps can be reasonably adjusted according to the actual forecast scenario, as long as the preprocessing, assimilation, and high-resolution forecasting of satellite data can be achieved.

[0030] Further, the full-channel brightness temperature dataset and the auxiliary dataset of the complex terrain background field from the FY4B satellite real-time observation dataset are input into a preset radiative transfer model, so that the radiative transfer model performs channel filtering based on the full-channel brightness temperature dataset and the auxiliary dataset of the complex terrain background field to obtain a channel subset brightness temperature dataset, including: In the preset radiative transfer model, the following steps are performed: Based on the background field auxiliary dataset of the complex terrain region, typical atmospheric profile data and land surface type data of the complex terrain region are obtained; For each channel brightness temperature data in the full-channel brightness temperature dataset, the weighted peak height of that channel is obtained based on the brightness temperature data of that channel and the typical atmospheric profile data of complex terrain areas. Peak values ​​are filtered based on the peak height of several channel weights and the corresponding brightness temperature data of several channels to obtain the first channel subset; Based on the full-channel brightness temperature dataset, typical atmospheric profile data of complex terrain areas, and surface type data, a surface emissivity sensitivity analysis was performed to obtain the second channel subset. The brightness temperature dataset of the channel subset is obtained by taking the intersection of the first channel subset and the second channel subset.

[0031] In this embodiment, based on a preset radiative transfer model, typical atmospheric profile data and surface type data for complex terrain areas are obtained using a background field auxiliary dataset. For the brightness temperature data of each channel in the FY4B satellite full-channel brightness temperature dataset, the channel weight peak height is calculated and filtered to obtain a first channel subset. Simultaneously, surface emissivity sensitivity analysis is conducted by combining the full-channel brightness temperature dataset, typical atmospheric profile data for complex terrain areas, and surface type data to obtain a second channel subset. By taking the intersection of the first and second channel subsets, the final channel subset brightness temperature dataset is obtained. This embodiment can accurately eliminate observation channels significantly affected by the underlying surface of complex terrain, reduce noise interference from the underlying surface of complex terrain, ensure the effectiveness and stability of the channel subset brightness temperature data, and provide reliable data support for subsequent dynamic bias correction, quality control, and assimilation analysis, thereby improving the adaptability and utilization of satellite observation data in numerical forecasting for complex terrain areas.

[0032] It should be noted that the preset radiative transfer model includes, but is not limited to, the Fast Radiative Transfer Model (RTTOV) and the Common Radiative Transfer Model (CRTM). Any radiative transfer model that can simulate atmospheric radiative transfer, calculate channel weight functions, and perform surface emissivity sensitivity analysis is applicable to this invention. The typical atmospheric profile data of the complex terrain area can be obtained through climatological statistics, field radiosonde observations, reanalysis data interpolation, etc. The surface type data can adopt global or regional land cover classification datasets, and the two types of data can be localized and corrected according to the underlying surface characteristics of the complex terrain area.

[0033] Furthermore, based on the full-channel brightness temperature dataset, typical atmospheric profile data of complex terrain areas, and surface type data, surface emissivity sensitivity analysis is performed to obtain a second channel subset, including: For each channel brightness temperature data in the full-channel brightness temperature dataset, the basic emissivity value is determined based on the channel brightness temperature data and the surface type data. The first simulated brightness temperature is obtained based on the brightness temperature data of this channel and the basic emissivity value; The perturbation emission rate value is obtained based on the base emission rate value and the preset perturbation amplitude; The second simulated brightness temperature is obtained based on the brightness temperature data of this channel and the perturbation emissivity value; Based on the first simulated brightness temperature and the second simulated brightness temperature, obtain the absolute value of the brightness temperature change corresponding to the brightness temperature data of this channel; Based on a preset change threshold, several channel brightness temperature data and corresponding absolute values ​​of brightness temperature change, the channels are filtered to obtain the second channel subset.

[0034] In this embodiment, for the brightness temperature data of each channel in the full-channel brightness temperature dataset, a base emissivity value is first determined based on the brightness temperature data of that channel and the surface type data. A first simulated brightness temperature corresponding to the base emissivity value and a second simulated brightness temperature corresponding to the perturbed emissivity value obtained by adjusting the base emissivity value by a preset perturbation amplitude are then obtained. Next, the absolute value of the brightness temperature change corresponding to the brightness temperature data of that channel is calculated using the first and second simulated brightness temperatures, thus characterizing the degree of influence of surface emissivity changes on the channel brightness temperature. Finally, channel selection is completed based on a preset change threshold to obtain a second subset of channels. This embodiment can effectively select channels with weaker responses to surface emissivity changes, reduce simulation errors caused by surface emissivity fluctuations in complex terrain areas, ensure the accuracy and stability of the selected channel data, provide a reliable data foundation for subsequent satellite observation data processing, assimilation analysis, and numerical prediction, and improve the application effect of satellite data in numerical prediction in complex terrain areas.

[0035] It should be noted that the preset disturbance amplitude is not limited to a fixed value, but can be adaptively adjusted within the range of ±0.005 to ±0.02 according to the surface type characteristics of complex terrain areas; the preset change threshold can be set within the range of 0.5K to 2.0K according to the forecast accuracy requirements, and any threshold value that can effectively eliminate surface emissivity sensitive channels is applicable to this invention; the basic emissivity value can be the measured emissivity corresponding to the surface type, the default emissivity of the radiative transfer model, or the emissivity dataset after localized correction.

[0036] In one embodiment, a numerical weather prediction method for complex terrain areas based on FY4B is provided, the implementation of which depends on the following implementation environment and data preparation. The hardware environment adopts a high-performance computing cluster with a central processing unit (CPU) core count of no less than 512 and memory of no less than 2TB; the software environment includes a Linux operating system, a Weather Research and Forecasting Model (WRF, V4.0 and above) numerical weather prediction model and its three-dimensional variational (3DVAR) assimilation system or ensemble Kalman filter (EnKF) assimilation system, as well as a fast radiative transfer model (RTTOV) or a universal radiative transfer model (CRTM). The data sources mainly include: radiation brightness temperature data from the L1-level hyperspectral infrared detector (GIIRS) of the FY-4B satellite, with a time resolution of 15 minutes and more than a thousand spectral channels; cloud image products from the Advanced Geostationary Radiation Imager (AGRI) of the FY-4B satellite, used for cloud detection; a high-precision digital elevation model, using internationally recognized spaceborne radar topographic mapping data, with a resolution of 30 meters; conventional observation data, including radiosonde and ground station data, as well as fifth-generation reanalysis data released by authoritative international meteorological agencies, used for verifying forecast results; in addition, the initial and boundary conditions of the WRF model also need to be prepared, such as global forecast field data released by mainstream global forecast systems in the industry or internationally renowned meteorological forecast agencies.

[0037] It should be noted that the configuration parameters of the hardware environment can be upgraded or adaptively adjusted according to the range, resolution, and forecast timeliness requirements of the forecast area, supporting distributed computing of multi-node clusters; the version of the WRF mode in the software environment is not limited to V4.0 and above, and any mode version that can achieve high-resolution numerical forecasting in complex terrain areas is applicable to this invention, and the assimilation system includes, but is not limited to, 3DVAR and EnKF, and a hybrid assimilation algorithm can also be used; the resolution of the high-precision digital elevation model can be selected in the range of 10 meters to 90 meters according to the forecast refinement, and the reanalysis data is not limited to the fifth generation, and any reanalysis data that can realize forecast result verification and background field construction is applicable to this invention.

[0038] First, input data is acquired, including the GIIRS full-channel brightness temperature dataset, typical atmospheric profiles of complex terrain regions, and land surface type data. The typical atmospheric profiles of complex terrain regions represent the vertical temperature and humidity structure of complex terrain areas such as plateaus, while the land surface type data characterizes different land cover classifications such as snow, bare soil, and vegetation. A channel selection process is performed within a pre-defined radiative transfer model. On one hand, based on the typical atmospheric profiles of complex terrain regions, the weight function of each channel is calculated using the radiative transfer model, and its peak height is analyzed. Channels with peak values ​​between 700 hPa and 400 hPa are selected to form the first channel subset. Channels in this subset primarily sense atmospheric information and are less affected by surface interference. On the other hand, based on the full-channel brightness temperature dataset, the typical atmospheric profile data of complex terrain regions, and the land surface type data, surface emissivity sensitivity analysis is performed to obtain the second channel subset. Specifically, for the brightness temperature data of each channel in the full-channel brightness temperature dataset, the baseline emissivity value is first determined based on the corresponding land surface type data. Based on this baseline emissivity value, a first simulated brightness temperature is obtained using a radiative transfer model. Then, a preset perturbation amplitude, such as ±0.01, is applied to the baseline emissivity value to obtain a perturbed emissivity value, and a second simulated brightness temperature is obtained based on this perturbed emissivity value. The absolute value of the brightness temperature change for each channel is calculated based on the first and second simulated brightness temperatures; this absolute value characterizes the degree of influence of land surface emissivity changes on the channel brightness temperature. Afterwards, based on a preset change threshold, such as 1.0K, all channels are filtered, and channels with absolute brightness temperature changes greater than this threshold are removed. The remaining channels constitute the second subset of channels that are insensitive to land surface emissivity.

[0039] It should be noted that the weighted peak height screening range of the first channel subset is not limited to 700hPa to 400hPa. It can be adaptively adjusted within the range of 300hPa to 850hPa according to the atmospheric boundary layer characteristics of complex terrain areas and the atmospheric layers of interest in the forecast. Any height range that can screen out the main atmospheric information and is less affected by surface interference is applicable to this invention.

[0040] In this embodiment, observation channels are screened from two dimensions—atmospheric sensitivity and surface interference—through the aforementioned weighting function peak height analysis and surface emissivity sensitivity analysis. The former ensures that the selected channels can effectively capture information from the upper atmosphere, while the latter excludes channels that introduce significant simulation errors due to surface emissivity fluctuations in complex terrain areas. Finally, the intersection of the first and second channel subsets is taken to obtain the channel subset brightness temperature dataset. The channel subset selected in this way can effectively sense atmospheric conditions and has low sensitivity to changes in the emissivity of the underlying surface in complex terrain, thereby reducing surface noise interference in the subsequent assimilation process. This provides reliable data support for dynamic bias correction, quality control, and assimilation analysis, improving the adaptability and utilization of satellite observation data in numerical prediction in complex terrain areas.

[0041] Furthermore, based on the aforementioned channel subset brightness temperature dataset, the auxiliary dataset of the background field in complex terrain areas, the short-term forecast background field, and the satellite scan angle and cloud mask data in the FY4B satellite real-time observation dataset, dynamic bias correction is performed to obtain the corrected brightness temperature dataset, including: Land surface type data are obtained based on the background field auxiliary dataset of the complex terrain area; Based on the channel subset brightness temperature dataset, surface type data, and satellite scan angle simulation of theoretical brightness temperature, a theoretical brightness temperature dataset is obtained; The observation bias is calculated based on the theoretical brightness temperature dataset and the channel subset brightness temperature dataset to obtain the original observation bias dataset. A multiple linear regression model is constructed based on the original observation bias dataset, satellite scan angle, land surface type data, and cloud mask data. A systematic bias estimate is obtained based on the multiple linear regression model and the original observation bias dataset. Brightness temperature correction is performed based on the original observation bias dataset and the systematic bias estimate to obtain the corrected brightness temperature dataset.

[0042] In this embodiment, surface type data is obtained based on the background field auxiliary dataset for complex terrain areas. Theoretical brightness temperature datasets are then obtained by simulating theoretical brightness temperatures using channel subset brightness temperature datasets, surface type data, and satellite scan angles. Observational bias datasets are calculated using the theoretical and channel subset brightness temperature datasets to obtain the original observational bias dataset. A multiple linear regression model is then constructed using the original observational bias dataset, satellite scan angles, surface type data, and cloud mask data to obtain accurate estimates of systematic bias. Finally, brightness temperature is corrected based on the original observational bias dataset and the systematic bias estimates to obtain the corrected brightness temperature dataset. This embodiment is well-suited to the observation environment of complex terrain areas, effectively eliminating systematic observational biases caused by satellite scan angles, surface types, and cloud mask factors. It significantly improves the accuracy and reliability of brightness temperature data, providing high-quality observational data for subsequent cloud-terrain quality control and high-frequency cyclic assimilation processes, and laying the foundation for optimizing the initial field accuracy of numerical forecasts in complex terrain areas.

[0043] It should be noted that the short-term forecast background field can be obtained through short-term integration of numerical forecast models, reanalysis data interpolation, etc., and the background field can be processed by temporal and spatial interpolation according to the forecast time to match the spatiotemporal resolution of satellite observations. The input factors of the multiple linear regression model are not limited to satellite scan angle, surface type data, and cloud mask data. Satellite azimuth angle, atmospheric column water vapor content, terrain height, and other factors can be added according to the observation characteristics of complex terrain areas. Any combination of factors that can effectively fit the systematic bias is applicable to this invention.

[0044] In one embodiment, the dynamic bias correction is performed as follows: First, input data is acquired, including a channel subset brightness temperature dataset, the WRF short-term forecast background field, satellite scan angle, surface type data, and cloud mask data from the Advanced Geostationary Radiometric Imager (AGRI) on the FY-4B satellite, where the cloud mask data provides cloud cover (CF) information for each observation point. Based on the above inputs, a theoretical brightness temperature dataset is simulated using either the Fast Radiative Transfer Mode (RTTOV) or the Universal Radiative Transfer Mode (CRTM), with the WRF short-term forecast background field as input. Subsequently, the brightness temperature data of the channel subset was used as the observed brightness temperature. Calculate the observation bias This yields the original observation bias dataset.

[0045] Based on this, a multiple linear regression model is established using a sliding time window (e.g., 7 days) to estimate systematic bias. The multiple linear regression model is expressed as: in, The value is the estimated systematic bias to be subtracted from the observations, in Kelvin (K); θ is the satellite scan angle. This is a land cover type index used to distinguish different land cover categories; CF represents cloud cover. , , , These are the regression coefficients obtained by fitting the observation bias data within a sliding time window. The original observation bias is then fitted using the aforementioned multiple linear regression model to obtain an estimate of the systematic bias.

[0046] It should be noted that the duration of the sliding time window is not limited to 7 days, but can be adaptively adjusted within the range of 3 to 14 days according to the weather system evolution characteristics of complex terrain areas and the amount of satellite observation data; the fitting method of the multiple linear regression model includes, but is not limited to, the least squares method. Any fitting algorithm that can effectively solve the regression coefficients is applicable to this invention, and the fitted model can be subjected to significance testing to remove factors that are not statistically significant.

[0047] In this embodiment, due to the strong heterogeneity of the underlying surface and complex cloud conditions in complex terrain areas, satellite-observed brightness temperatures often contain systematic biases related to scan angle, surface type, and cloud cover. By constructing a multiple linear regression model that includes these factors, such systematic biases can be separated from the original observation biases. Finally, by subtracting the estimated value of the systematic biases from the original observed brightness temperatures, a corrected brightness temperature dataset is obtained. This embodiment effectively eliminates systematic errors introduced by satellite scanning geometry, surface inhomogeneity, and cloud interference, significantly improving the accuracy and reliability of brightness temperature data. It provides high-quality observation data for subsequent cloud-topography quality control and high-frequency cyclic assimilation, thus laying the foundation for optimizing the initial field accuracy of numerical forecasts in complex terrain areas.

[0048] Furthermore, the cloud-topography quality control based on the corrected brightness temperature dataset, the preset digital elevation model, and the cloud image data and satellite observation geometric information in the FY4B satellite real-time observation dataset, to obtain an effective brightness temperature observation dataset, includes: Thick cloud identification is performed based on the cloud map data and a preset multispectral threshold algorithm to obtain thick cloud detection results; Based on the cloud map data and the preset local binary pattern texture analysis algorithm, broken and shallow clouds are identified to obtain broken and shallow cloud detection results. Cloud contamination labeling results are obtained based on the thick cloud detection results and the broken, shallow cloud detection results. Based on the satellite observation geometric information and the preset digital elevation model, terrain occlusion is identified to obtain terrain occlusion marking results; Based on the cloud pollution labeling results and terrain occlusion labeling results, data is removed from the corrected brightness temperature dataset to obtain the effective brightness temperature observation dataset.

[0049] In this embodiment, based on cloud image data from the FY4B satellite real-time observation dataset, thick cloud detection results are obtained through a preset multispectral threshold algorithm for thick cloud identification, and broken, shallow cloud detection results are obtained through a preset local binary mode texture analysis algorithm for broken, shallow cloud identification. Cloud contamination labeling results are then obtained based on these two types of detection results. Simultaneously, terrain occlusion identification is performed by combining satellite observation geometric information with a preset digital elevation model, resulting in terrain occlusion labeling results. Based on these results, the corrected brightness temperature dataset is purged according to the cloud contamination labeling and terrain occlusion labeling results, ultimately yielding a valid brightness temperature observation dataset. This embodiment can accurately eliminate invalid observations caused by cloud contamination and terrain occlusion, ensuring the reliability and applicability of the observation data. It provides high-quality observation data for the subsequent construction of an assimilated three-dimensional analysis field, improves the application quality of satellite data in complex terrain areas, and lays a stable data foundation for high-resolution numerical weather prediction.

[0050] It should be noted that the preset multispectral threshold algorithm can be localized based on the spectral channel characteristics of the FY4B satellite AGRI cloud image, and the thick cloud identification algorithm is not limited to the multispectral threshold method, but can also use cloud top height detection, infrared brightness temperature difference method, etc.; the local binary mode (LBP) texture analysis algorithm can be replaced by other texture analysis algorithms such as gray-level co-occurrence matrix and wavelet transform. Any algorithm that can effectively identify broken and shallow clouds is applicable to this invention.

[0051] Further, the data removal process based on the cloud contamination labeling results and terrain occlusion labeling results to obtain the effective brightness temperature observation dataset includes: The satellite zenith angle is obtained based on the aforementioned satellite observation geometric information; Based on the satellite zenith angle and the preset zenith angle threshold, the modified brightness temperature dataset is initially purged to obtain the initial brightness temperature observation dataset. Based on the cloud pollution labeling results and terrain occlusion labeling results, the initial brightness temperature observation dataset is purged to obtain the effective brightness temperature observation dataset.

[0052] In this embodiment, the satellite zenith angle is obtained based on satellite observation geometry information. The modified brightness temperature dataset is initially purged based on the satellite zenith angle and a preset zenith angle threshold to obtain an initial brightness temperature observation dataset. Then, based on cloud contamination and terrain occlusion marking results, further data purging is performed on the initial brightness temperature observation dataset to finally obtain a valid brightness temperature observation dataset. This embodiment employs a hierarchical and progressive filtering logic. First, abnormal data with poor observation geometry is excluded using the zenith angle threshold. Then, invalid observations are removed by combining cloud contamination and terrain occlusion markings. This achieves multi-dimensional and multi-level data filtering, effectively improving the validity and reliability of the brightness temperature observation data. This provides high-quality observation data for the subsequent construction of an assimilated three-dimensional analysis field using a high-frequency cyclic assimilation algorithm, ensuring the accuracy of the initial field construction for numerical forecasting in complex terrain areas and providing reliable data support for subsequent high-resolution forecasts.

[0053] It should be noted that the preset zenith angle threshold is not limited to a fixed value, but can be set within the range of 50° to 70° according to the accuracy characteristics of satellite observations and the geographical latitude of complex terrain areas. Any threshold value that can eliminate abnormal data with poor observation geometry is applicable to this invention. The data elimination method includes directly eliminating observation points marked as invalid, or marking invalid observation points and reducing their weight in the subsequent assimilation process. Both methods are within the protection scope of this invention.

[0054] In one embodiment, the cloud-terrain quality control in a numerical prediction method for complex terrain areas based on FY4B is specifically performed in the following manner.

[0055] First, obtain the input data, including the corrected brightness temperature dataset output after dynamic bias correction. The data includes cloud images from the Advanced Geostationary Radiometric Imager (AGRI) of the FY-4B satellite, a pre-defined digital elevation model (DEM), and satellite observation geometric information, including the latitude and longitude of each observation point, the satellite zenith angle θ, and the satellite azimuth angle φ.

[0056] Based on the above inputs, cloud contamination labeling is performed first. Cloud detection consists of two parts: thick cloud identification and fragmented shallow cloud identification. On one hand, thick clouds are identified based on cloud image data and a preset multispectral threshold algorithm, yielding thick cloud detection results. On the other hand, for fragmented shallow clouds, a preset Local Binary Pattern (LBP) texture analysis algorithm is used to extract infrared cloud image texture features for identification, yielding fragmented shallow cloud detection results. During this process, the satellite zenith angle θ can be used to assist in the judgment, for example, by adjusting the threshold for observations with larger zenith angles to compensate for signal attenuation caused by the increase in the slant path. The cloud contamination labeling results are obtained by combining the thick cloud detection results and the fragmented shallow cloud detection results.

[0057] Simultaneously, terrain occlusion is marked. Based on the observation point's latitude and longitude, satellite azimuth angle φ, and zenith angle θ from the satellite observation geometry, as well as pre-set digital elevation model (DEM) data, ray tracing is performed backward from the observation point along the satellite's line of sight, where the satellite's line of sight direction is defined by both the satellite azimuth angle φ and the zenith angle θ. Specifically, the line of sight is discretized into a series of step sizes. At each step, the terrain height at the corresponding horizontal position is queried. If the terrain height is greater than the height of the line of sight at that step size, it is determined to be terrain-occluded; otherwise, it is determined to be unoccluded. The output is a Boolean value, i.e., "unoccluded" or "occluded," thus obtaining the terrain occlusion marking result.

[0058] After completing cloud contamination and terrain occlusion labeling, data culling is performed to obtain a valid brightness temperature observation dataset. First, the satellite zenith angle θ is obtained based on satellite observation geometry. Initial culling of the corrected brightness temperature dataset is then performed based on the satellite zenith angle and a preset zenith angle threshold (e.g., 60°), removing observation points with zenith angles greater than this threshold, resulting in the initial brightness temperature observation dataset. Subsequently, based on the cloud contamination and terrain occlusion labeling results, further data culling is performed on the initial brightness temperature observation dataset, removing observation points that meet any of the following criteria: thick clouds, fragmented clouds with uncertain texture features, or severe terrain occlusion.

[0059] It should be noted that complex terrain areas are often accompanied by fragmented and shallow cloud distributions and obstruction of satellite views by complex terrain. Single cloud detection or terrain filtering methods are insufficient to effectively eliminate all invalid observations. This embodiment employs a layered and progressive filtering logic: first, outlier data with poor observation geometry is eliminated using a zenith angle threshold; then, a cloud detection method combining multispectral thresholding and texture analysis is used to identify different types of cloud contamination; simultaneously, ray tracing technology is used to accurately determine terrain occlusion, achieving multi-dimensional and multi-level data quality control. Finally, a valid brightness temperature observation dataset is output for assimilation. This accurately eliminates invalid observations caused by cloud contamination and terrain occlusion, ensuring the reliability and applicability of the observation data. It provides high-quality observation data for subsequent construction of an assimilated three-dimensional analysis field, improves the application quality of satellite data in complex terrain areas, and lays a stable data foundation for high-resolution numerical weather prediction.

[0060] Furthermore, based on the effective brightness temperature observation dataset and the preset high-frequency cyclic assimilation algorithm, an assimilation three-dimensional analysis field is constructed, including: Get the default observation error covariance matrix; For each observation point in the effective brightness temperature observation dataset, cloud cover data, terrain complexity index, and satellite zenith angle are obtained based on the observation point. The observation error adjustment factor is obtained based on the cloud cover data, terrain complexity index, and satellite zenith angle. The observation error covariance matrix corresponding to the observation point is obtained based on the observation error adjustment factor and the default observation error covariance matrix. Assimilation analysis is performed based on the preset high-frequency cyclic assimilation algorithm, the effective brightness temperature observation dataset, and the observation error covariance matrix corresponding to several observation points to obtain the assimilation three-dimensional analysis field.

[0061] In this embodiment, a default observation error covariance matrix is ​​first obtained. For each observation point in the effective brightness temperature observation dataset, the cloud cover data, terrain complexity index, and satellite zenith angle corresponding to that observation point are obtained, and an observation error adjustment factor is obtained based on the above information. Then, the observation error adjustment factor is combined with the default observation error covariance matrix to obtain the observation error covariance matrix corresponding to that observation point. Subsequently, based on a preset high-frequency cyclic assimilation algorithm, assimilation analysis is performed in combination with the effective brightness temperature observation dataset and the observation error covariance matrix corresponding to each observation point to construct an assimilated three-dimensional analysis field. This embodiment can adaptively match observation errors by combining cloud cover, terrain complexity, and satellite observation geometry, achieving fine-grained control of observation data errors, improving the reliability of assimilation analysis, optimizing the accuracy and quality of the assimilated three-dimensional analysis field, laying a solid foundation for subsequent high-resolution forecasting of complex terrain areas, and effectively improving the application effect of satellite data in numerical weather prediction.

[0062] It should be noted that the default observation error covariance matrix can be the default value of the radiative transfer model, the statistical error of the observation data, or the error matrix after local correction, and the matrix can be spatially and temporally interpolated according to the observation characteristics of complex terrain areas; the terrain complexity index can be calculated by the terrain undulation, slope, roughness and other indicators of the digital elevation model, and all indicators that can characterize the degree of terrain ruggedness are applicable to the calculation of the terrain complexity index of this invention.

[0063] In one embodiment, an assimilation three-dimensional analysis field is constructed based on the effective brightness temperature observation dataset and a preset high-frequency cyclic assimilation algorithm, specifically as follows: First, input data is acquired, including the effective brightness temperature observation dataset output in step S1 and the default observation error covariance matrix. Cloud mask data and terrain complexity index The cloud mask data comes from cloud detection products generated using AGRI cloud maps, providing cloud cover (CF) information for each observation point; the terrain complexity index... It is calculated from a preset digital elevation model (DEM) and is used to characterize the ruggedness of the terrain.

[0064] For data integration, the effective brightness temperature observation dataset is input into the three-dimensional variational assimilation system (WRF-DA) of the WRF numerical weather prediction model or the China Meteorological Administration assimilation system (CMA). The observation operator configuration adopts either the Fast Radiative Transfer Model (RTTOV) or the Universal Radiative Transfer Model (CRTM) to achieve the conversion from model variables to satellite-observed brightness temperature.

[0065] During the adaptive optimization of observation errors, for each observation point in the effective brightness temperature observation dataset, the cloud cover CF value is first read from its corresponding cloud mask data, and the terrain complexity index of that observation point is also obtained. And the satellite zenith angle θ. Based on the above information, calculate the observation error adjustment factor α for each observation point: in, and CF is a non-negative weighting coefficient, and CF is cloud cover (the value usually ranges from 0 to 1). Here, θ represents the terrain complexity index, and θ is the satellite zenith angle. Subsequently, this adjustment factor is used to adjust the default observation error covariance matrix: This yields the observation error covariance matrix corresponding to the observation point.

[0066] It should be noted that the non-negative weighting coefficients , The assimilation system can be determined through numerical experiments and statistical fitting based on the cloud conditions and topographic features of complex terrain areas. Differentiated weighting coefficients can be set for different surface types and different cloud cover ranges. The assimilation system is not limited to WRF-DA and CMA. Any numerical forecast assimilation system that can achieve satellite radiation brightness temperature data assimilation is applicable to this invention. Furthermore, the configuration of the observation operator can be parameterized and optimized according to the type of radiation transfer model.

[0067] In this embodiment, it should be noted that the greater the cloud cover, the more complex the terrain, and the larger the satellite zenith angle, the higher the uncertainty of the observed brightness temperature data, and its assimilation weight should be reduced accordingly. By introducing the aforementioned observation error adjustment factor, when CF increases, Increase or As the value of α increases, the observation error at that point is amplified, thus automatically reducing its weight in the assimilation analysis. This embodiment can achieve fine-grained control over the error of observation data by combining cloud cover, terrain complexity, and satellite observation geometry.

[0068] After completing the adaptive optimization of observation errors, high-frequency cyclic assimilation is performed. The assimilation period is set to 1 hour. Based on the preset high-frequency cyclic assimilation algorithm (such as three-dimensional variational assimilation analysis), assimilation analysis is performed by combining the effective brightness temperature observation dataset and the observation error covariance matrix corresponding to each observation point. Finally, a high-precision three-dimensional analysis field after assimilation is output, including meteorological elements such as temperature, humidity, wind field, and air pressure.

[0069] It should be noted that the preset high-frequency cyclic assimilation algorithm includes, but is not limited to, three-dimensional variational assimilation analysis. It can also employ algorithms such as ensemble Kalman filtering and mixed data assimilation. Any algorithm that can achieve high-frequency cyclic assimilation is applicable to this invention. Furthermore, the meteorological elements of the assimilation analysis are not limited to temperature, humidity, wind field, and air pressure, but can also include specific humidity, geopotential height, etc.

[0070] This embodiment improves the reliability of assimilation analysis and the accuracy of the analysis field by adaptively matching observation errors, laying a solid foundation for subsequent high-resolution forecasting of complex terrain areas and effectively enhancing the application effect of satellite data in numerical forecasting.

[0071] Furthermore, the high-resolution forecasting based on the assimilated three-dimensional analysis field to obtain numerical forecasting results for complex terrain areas includes: Acquire large-scale boundary condition data and conventional initial fields; High-resolution forecasts are performed based on the large-scale boundary condition data, the assimilated three-dimensional analysis field, and the preset regional numerical forecasting model to obtain high-precision forecast results. High-resolution forecasts are performed based on the large-scale boundary condition data, conventional initial fields, and preset regional numerical forecast models to obtain conventional forecast results. Numerical forecast results for the complex terrain region are obtained based on the high-precision forecast results and the conventional forecast results.

[0072] In this embodiment, large-scale boundary condition data and a conventional initial field are acquired. High-resolution forecasts are then performed based on the assimilated 3D analysis field, large-scale boundary condition data, and a preset regional numerical weather prediction model, yielding high-precision forecast results. Simultaneously, high-resolution forecasts are also performed based on the conventional initial field, large-scale boundary condition data, and the preset regional numerical weather prediction model, resulting in conventional forecast results. Finally, numerical weather prediction results for complex terrain areas are obtained based on the high-precision and conventional forecast results. This embodiment employs consistent large-scale boundary conditions and a regional numerical weather prediction model, using only the initial field as the sole variable. This allows for a direct demonstration of the improvement effect of the assimilated 3D analysis field on the forecast results, ensuring the accuracy and comparability of numerical weather prediction results for complex terrain areas. It fully leverages the effect of satellite data assimilation and optimization of the initial field, effectively improving the precision of numerical weather predictions for complex terrain areas and providing more reliable data support for meteorological forecasts in these regions.

[0073] It should be noted that the conventional initial field is an analysis field obtained by assimilating only conventional observation data. Conventional observation data includes, but is not limited to, observation data from radiosondes, ground stations, and radar. The construction method of the conventional initial field is consistent with the construction method of the assimilated three-dimensional analysis field, except that FY4B satellite data is not incorporated. The numerical prediction results for the complex terrain area are obtained by directly using high-precision prediction results or by combining and correcting the differences between high-precision prediction results and conventional prediction results. Both methods are within the protection scope of this invention.

[0074] Furthermore, the numerical prediction method for complex terrain areas based on FY4B, as described above, also includes: Acquire real-time observation data, and obtain high-precision forecast results and conventional forecast results based on the numerical forecast results of the complex terrain area; Based on the high-precision forecast results, conventional forecast results, and actual observation data, traditional skill scoring is performed to obtain high-precision forecast scoring results and conventional forecast scoring results; Based on the high-precision forecast scoring results and the conventional forecast scoring results, a significant difference analysis was conducted to obtain the key areas for forecast improvement and the key forecast lead time. Based on the forecast improvement key areas and key forecast lead times, high-precision three-dimensional meteorological element fields corresponding to the regions and lead times are extracted from the high-precision forecast results; Based on the forecast improvement key areas and key forecast lead times, the conventional three-dimensional meteorological element fields corresponding to the regions and lead times are extracted from the high-precision forecast results; Based on the high-precision three-dimensional meteorological element field and the conventional three-dimensional meteorological element field, multi-physical quantity collaborative diagnosis is performed to obtain physical mechanism diagnosis results.

[0075] In this embodiment, real-time observation data is acquired, and high-precision forecast results and conventional forecast results are extracted based on numerical forecast results for complex terrain areas. High-precision forecast scores and conventional forecast scores are obtained through traditional skill scoring. A significant difference analysis is conducted based on the two types of forecast scores to identify key areas and key forecast lead times for forecast improvement. Based on these key areas and key forecast lead times, high-precision three-dimensional meteorological element fields for the corresponding areas and lead times are extracted from the high-precision forecast results, and conventional three-dimensional meteorological element fields for the corresponding areas and lead times are extracted from the conventional forecast results. Then, multi-physical quantity collaborative diagnosis is performed based on the high-precision and conventional three-dimensional meteorological element fields to obtain physical mechanism diagnosis results. This embodiment can accurately locate the key spatiotemporal range for forecast improvement, clarify the intrinsic mechanism of forecast optimization through multi-physical quantity collaborative diagnosis, provide objective basis for the iterative optimization of numerical forecasting methods for complex terrain areas based on the FY4B satellite, and further enhance the scientific rigor and practicality of the forecasting method.

[0076] It should be noted that the real-time observation data includes, but is not limited to, multi-source observation data such as radiosonde, ground station, satellite, and radar. The multi-source real-time observation data can be quality controlled and fused before being used for scoring calculation. The difference significance analysis can use statistical methods such as t-test and F-test. Any analytical method that can effectively identify key areas for forecast improvement and key forecast lead times is applicable to this invention. The physical quantities used for multi-physical quantity collaborative diagnosis can be selected according to the characteristics of the weather system of interest in the forecast, and are not limited to water vapor flux, convective available potential energy, wind field, etc., mentioned later in this invention.

[0077] In one embodiment, the high-resolution forecasting based on the assimilated three-dimensional analysis field to obtain numerical forecasting results for complex terrain areas can be achieved in the following way: Large-scale boundary condition data and a conventional initial field are acquired. In this embodiment, the large-scale boundary condition data is provided by a global forecast model, such as a mainstream global forecast system or a forecast product from an authoritative international meteorological forecasting agency. Specifically, it includes meteorological field data that changes over time during the forecast period, i.e., the three-dimensional spatial distribution of meteorological elements such as temperature, humidity, horizontal wind field, air pressure, or geopotential height at regular time intervals (e.g., 6 hours) along the regional model grid boundary. The conventional initial field is an analytical field obtained by assimilating only conventional observational data. Based on this, two sets of high-resolution forecast experiments are run in parallel. Both sets of experiments use the same large-scale boundary condition data and a preset regional numerical forecast model, with the initial field as the only variable. Specifically, high-resolution forecasts are performed based on the large-scale boundary condition data, the assimilated three-dimensional analytical field, and the preset regional numerical forecast model to obtain high-precision forecast results; simultaneously, high-resolution forecasts are performed based on the large-scale boundary condition data, the conventional initial field, and the preset regional numerical forecast model to obtain conventional forecast results.

[0078] It should be noted that the preset regional numerical forecasting model described in this embodiment can adopt the WRF model and use a three-layer nested grid configuration, such as a nested grid of 27km, 9km, and 3km, with the innermost layer covering complex terrain areas (such as the Qinghai-Tibet Plateau region). The meteorological elements of the large-scale boundary condition data are not limited to temperature, humidity, horizontal wind field, air pressure, or geopotential height, but may also include specific humidity, vertical velocity, etc. Any meteorological elements that can provide effective boundary driving for regional numerical forecasting are applicable to this invention.

[0079] Before model operation, the assimilated 3D analysis field or the conventional initial field is set as the 3D initial field at the start of model integration, and the large-scale boundary condition data is prepared in a model-readable format and configured as the external driving conditions required by the model during integration. During model integration, the model dynamically reads and interpolates the values ​​of meteorological elements on the regional grid boundary at the current and next time steps from the large-scale boundary condition file according to a preset time step. These boundary values ​​are applied as strong constraints in the boundary processing scheme of the regional model, thereby ensuring that the forecast results within the region are consistent with the large-scale environmental evolution. Finally, the numerical forecast results for the complex terrain area are obtained based on the high-precision forecast results and the conventional forecast results.

[0080] It should be noted that the boundary processing schemes of the regional model include, but are not limited to, nudging, sponge boundary, and gradient boundary schemes. Any boundary processing scheme that can effectively couple large-scale boundary conditions with the regional model is applicable to this invention. The time step of the model integration can be set adaptively according to the resolution of the nested grid, following the CFL condition of numerical prediction to ensure the stability of the model integration.

[0081] This embodiment demonstrates the effect of assimilation of the three-dimensional analysis field on the forecast results by setting up two sets of control experiments with only different initial fields. This ensures the accuracy and comparability of numerical forecast results in complex terrain areas and effectively improves the precision of numerical forecasts in complex terrain areas.

[0082] Furthermore, to verify the improved effect of the above forecasting method and reveal its physical mechanism, this embodiment also includes a diagnostic analysis step for the forecast results. Specifically, real-time observation data is acquired, and high-precision forecast results and conventional forecast results are obtained based on the numerical forecast results for the complex terrain area. Traditional skill scoring is performed based on the high-precision forecast results, conventional forecast results, and real-time observation data to obtain high-precision forecast scoring results and conventional forecast scoring results. A significant difference analysis is performed on the high-precision forecast scoring results and conventional forecast scoring results to obtain key areas for forecast improvement and key forecast lead times. Based on the key areas for forecast improvement and key forecast lead times, corresponding high-precision three-dimensional meteorological element fields and conventional three-dimensional meteorological element fields are extracted from the high-precision forecast results and conventional forecast results, respectively. Multi-physical quantity collaborative diagnosis is performed based on the high-precision three-dimensional meteorological element fields and conventional three-dimensional meteorological element fields to obtain physical mechanism diagnostic results. This embodiment accurately locates the key spatiotemporal range for forecast improvement through difference significance analysis, and clarifies the intrinsic mechanism of forecast optimization by using multi-physical quantity collaborative diagnosis. It provides an objective basis for the iterative optimization of numerical forecasting methods for complex terrain regions based on FY4B, and further enhances the scientificity and practicality of the forecasting method.

[0083] In another embodiment, to further verify the improved effect of the aforementioned forecasting method and reveal its physical mechanism, this embodiment also includes a forecast result verification and diagnostic analysis step, which can be implemented in the following way: Two sets of forecast products and real-time observation data were acquired. The two sets of forecast products included the aforementioned high-precision forecast results and conventional forecast results. The real-time observation data included radiosonde observations, ground station observations, and satellite precipitation observations. Based on this, traditional skills assessment and verification, as well as physical diagnosis of key weather processes, were conducted.

[0084] In this embodiment, the conventional skills assessment aims to quantitatively evaluate the forecast improvement effect from a statistical perspective. Specifically, based on the high-precision forecast results, conventional forecast results, and actual observation data, the root mean square error and bias of temperature, humidity, and wind field are calculated, and the TS score and ETS score of precipitation forecast are calculated simultaneously. The formula for calculating the TS score is as follows: Among them, M represents the number of grid points where precipitation events occurred in both forecast and observation (hit), M represents the number of grid points where precipitation was observed but not predicted (missed), and F represents the number of grid points where precipitation was predicted but not observed (false alarm). The TS score ranges from 0 to 1, with a larger value indicating a better forecast.

[0085] It should be noted that the traditional skill scoring is not limited to root mean square error, bias, TS score, and ETS score, but may also include correlation coefficient, hit rate, false alarm rate, missed alarm rate, BIAS score, etc. Any statistical scoring index that can quantitatively evaluate the effectiveness of numerical forecasting is applicable to this invention. The TS score and ETS score of precipitation forecast can be calculated in stages according to the precipitation level, including different levels such as light rain, moderate rain, heavy rain, and rainstorm, so as to achieve a refined evaluation of the forecast effectiveness of precipitation of different intensities.

[0086] The ETS score, by subtracting the random hit probability from the TS score, provides a more objective reflection of forecasting skill. Based on these scoring results, a "Forecast Improvement Assessment Report" is generated. This report includes tables comparing the root mean square error and bias of temperature, humidity, and wind field at key forecast lead times (e.g., 24 hours, 48 ​​hours) and key levels (e.g., surface, 850 hPa, 500 hPa), as well as a comparison table of precipitation TS and ETS scores. It also calculates the difference between the experimental and control groups to visually demonstrate the extent of improvement. For example, the report can reach the following quantitative conclusion: compared to the control group, the experimental group reduced the root mean square error of 850 hPa temperature by 15% in the 48-hour forecast, reduced the humidity bias by 0.5 g / kg, and reduced the root mean square error of wind field by 10%. The scoring results of these basic meteorological elements, combined with the precipitation score, constitute a comprehensive, multi-dimensional, objective, and quantitative assessment of the forecast improvement effect, confirming the positive correction effect of FY-4B data assimilation on the initial field and subsequent forecasts in terms of dynamic and thermal structure.

[0087] It should be noted that the physical diagnosis of key weather processes aims to reveal the underlying reasons for forecast improvements at the mechanistic level. This embodiment extracts representative meteorological element vertical profiles for the target area at the same forecast time from the high-precision forecast results and conventional forecast results. These profiles include temperature, specific humidity, and air pressure, typically obtained from model output or obtained by interpolating model layer data to the standard barosphere. Based on this, water vapor flux diagnosis, energy diagnosis, and wind field correction analysis are performed.

[0088] For water vapor flux diagnosis, the integral water vapor flux throughout the entire layer is calculated using the following formula: Where Q is the integral water vapor flux vector of the entire layer. It is the acceleration due to gravity. Surface air pressure, This refers to the pressure at the top of the atmosphere. ρ is the specific humidity (representing the mass of water vapor per unit mass of moist air), V is the horizontal wind vector, and the integral variable dp represents the integration over air pressure.

[0089] By comparing the differences in water vapor transport between the experimental and control groups, the correction effect of FY-4B data assimilation on water vapor transport flux was clarified. For energy diagnostics, the convective effective potential energy was calculated using the following formula: CAPE stands for Convective Effective Potential Energy, which represents the buoyant energy that an air parcel can obtain from the environment above the free convection height, and is measured in J / kg. The gas constant for dry air; The virtual temperature of the gas parcel represents the temperature of the gas parcel after taking into account the influence of water vapor. The virtual temperature of the ambient atmosphere; LFC is the free convection height, representing the initial height at which the air parcel begins to gain positive buoyancy; EL is the equilibrium height, representing the height at which the buoyancy of the air parcel disappears. This indicates integrating the logarithm of the air pressure.

[0090] Calculate the convective effective potential energy values ​​for the experimental group and the control group respectively, and calculate the difference between them: Thus, the spatial distribution reflecting the correction of unstable energy is obtained.

[0091] For wind field correction analysis, the horizontal wind field at the lower level (e.g., 850 hPa) at the same forecast time is extracted, and the wind vector difference field is calculated. The difference in the U components of the wind field is obtained. Difference between V and V components .

[0092] Based on this, the two-dimensional divergence of the wind vector difference field is calculated using the following formula: In the formula, Divergence represents the intensity of horizontal divergence or convergence of the wind field. This represents the rate of change of the difference in the U components along the x-direction. This represents the rate of change of the V component difference along the y-direction, and the partial derivatives are calculated on the model grid using the central difference method.

[0093] Simultaneously calculate the vertical vorticity using the following formula: In the formula, Vertical relative vorticity represents the intensity of wind field rotation. This represents the rate of change of the V component difference along the x-direction. This represents the rate of change of the difference in the U component along the y-direction.

[0094] By analyzing the spatial distribution and magnitude of the divergence and vorticity fields, the effects of assimilation on the low-level convergence and divergence characteristics and the intensity of the cyclonic circulation are clarified. For example, the following conclusions can be drawn: After assimilating FY-4B data, a significant convergence enhancement region appears on the eastern side of the target low vortex, with the center of negative divergence reaching [value missing]. At the same time, the cyclonic vorticity at the center of the low vortex increases. This indicates that assimilation effectively enhanced low-level convergence and cyclonic rotation in the low-vortex system.

[0095] Based on the above diagnostic results, a physical link was established between increased water vapor flux, lower-level wind field correction, and increased convective effective potential energy. By diagnosing the spatial distribution of water vapor flux differences, it was confirmed whether FY-4B data assimilation introduced more abundant water vapor transport in the lower strata of the target region. Analysis of the lower-level wind field correction and its divergence characteristics revealed that enhanced convergence regions favored water vapor accumulation, leading to increased lower-level specific humidity. This increased lower-level specific humidity directly resulted in higher humidity in the air parcels during uplift, enhanced release of latent heat of condensation, and a greater and longer-lasting difference in virtual temperature between the air parcels and the ambient virtual temperature during humid adiabatic processes, ultimately manifesting as an increase in convective effective potential energy. This resulted in the output of a "Physical Mechanism Diagnostic Conclusion," illustrating the improvement mechanism in graphical and textual form. This embodiment further outputs "Technical Optimization Suggestions," proposing targeted technical improvement directions based on the diagnostic conclusions. For example, for the eastern slope of the plateau, it was suggested to strengthen the optimization of surface emissivity parameters under dry and warm conditions during the bias correction process. This embodiment combines traditional skill assessment with physical diagnosis of key weather processes, which can quantitatively evaluate the effect of forecast improvement and reveal the reasons for the improvement from the perspective of dynamic and thermodynamic mechanisms. It provides an objective basis for the iterative optimization of numerical forecasting methods for complex terrain areas based on FY4B, and further enhances the scientificity and practicality of the forecasting method.

[0096] Please see Figure 2 This embodiment also provides a numerical prediction system for complex terrain areas based on FY4B, including: The data acquisition module is used to acquire the real-time observation dataset of FY4B satellite, the auxiliary dataset of background field in complex terrain areas, and the short-term forecast background field; The channel filtering module is used to input the full-channel brightness temperature dataset and the background field auxiliary dataset of complex terrain areas from the real-time observation dataset of the FY4B satellite into the preset radiative transfer model, so that the radiative transfer model can perform channel filtering based on the full-channel brightness temperature dataset and the background field auxiliary dataset of complex terrain areas to obtain a channel subset brightness temperature dataset. The dynamic deviation correction module is used to perform dynamic deviation correction based on the channel subset brightness temperature dataset, the background field auxiliary dataset of complex terrain area, the short-term forecast background field, and the satellite scan angle and cloud mask data in the FY4B satellite real-time observation dataset, to obtain the corrected brightness temperature dataset. The quality control module is used to perform cloud-topography quality control based on the corrected brightness temperature dataset, the preset digital elevation model, and the cloud image data and satellite observation geometric information in the FY4B satellite real-time observation dataset, so as to obtain an effective brightness temperature observation dataset. The assimilation analysis module is used to construct an assimilation three-dimensional analysis field based on the effective brightness temperature observation dataset and a preset high-frequency cyclic assimilation algorithm; The numerical prediction module is used to perform high-resolution prediction based on the assimilated three-dimensional analysis field to obtain numerical prediction results for complex terrain areas.

[0097] In this embodiment, the FY4B-based numerical weather prediction system for complex terrain regions acquires real-time FY4B satellite observation datasets, auxiliary datasets for the background field of complex terrain regions, and short-term forecast background fields through a data acquisition module. A channel filtering module uses a preset radiative transfer model to filter the full-channel brightness temperature dataset, obtaining a subset brightness temperature dataset. A dynamic bias correction module combines multiple data types with satellite scan angle and cloud mask data to perform dynamic bias correction, obtaining a corrected brightness temperature dataset. A quality control module performs cloud-topography quality control based on a digital elevation model, cloud image data, and satellite observation geometry information, obtaining an effective brightness temperature observation dataset. An assimilation analysis module constructs an assimilation three-dimensional analysis field based on the effective brightness temperature observation dataset and a preset high-frequency cyclic assimilation algorithm. The numerical weather prediction module performs high-resolution forecasts based on this analysis field, obtaining numerical weather prediction results for complex terrain regions. The various modules of the system are interconnected and operate collaboratively. Through full-process data quality control and efficient assimilation processing, the system improves the utilization rate of satellite data, enhances the quality of the initial field, and comprehensively improves the accuracy and reliability of numerical weather prediction for complex terrain regions.

[0098] It should be noted that the system described in this invention can be deployed on hardware platforms such as high-performance computing clusters, cloud computing platforms, and meteorological business servers. It supports multiple operating modes such as real-time forecasting, batch experiments, and retrospective forecasting. Furthermore, the system's output results can be connected to meteorological business forecasting platforms, visualization platforms, etc., to achieve rapid release and display of forecast results. Any hardware deployment and operating mode that can realize the system's functions is applicable to this invention.

[0099] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A numerical prediction method for complex terrain areas based on FY4B, characterized in that, Includes the following steps: Acquire real-time observation datasets from FY4B satellites, auxiliary datasets of background fields in complex terrain areas, and short-term forecast background fields; The full-channel brightness temperature dataset and the background field auxiliary dataset of complex terrain areas in the real-time observation dataset of the FY4B satellite are input into the preset radiative transfer model so that the radiative transfer model can perform channel filtering based on the full-channel brightness temperature dataset and the background field auxiliary dataset of complex terrain areas to obtain a channel subset brightness temperature dataset. Based on the aforementioned channel subset brightness temperature dataset, the complex terrain area background field auxiliary dataset, the short-term forecast background field, and the satellite scan angle and cloud mask data in the FY4B satellite real-time observation dataset, dynamic bias correction is performed to obtain the corrected brightness temperature dataset. Based on the corrected brightness temperature dataset, the preset digital elevation model, and the cloud image data and satellite observation geometric information in the FY4B satellite real-time observation dataset, cloud-topography quality control is performed to obtain an effective brightness temperature observation dataset. A three-dimensional assimilation analysis field is constructed based on the effective brightness temperature observation dataset and the preset high-frequency cyclic assimilation algorithm. High-resolution forecasts are performed based on the assimilated three-dimensional analysis field to obtain numerical forecast results for complex terrain areas.

2. The numerical prediction method for complex terrain areas based on FY4B according to claim 1, characterized in that, The process involves inputting the full-channel brightness temperature dataset and the auxiliary dataset of the complex terrain background field from the FY4B satellite real-time observation dataset into a preset radiative transfer model. This allows the radiative transfer model to perform channel filtering based on the full-channel brightness temperature dataset and the auxiliary dataset of the complex terrain background field, resulting in a channel subset brightness temperature dataset, including: In the preset radiative transfer model, the following steps are performed: Based on the background field auxiliary dataset of the complex terrain region, typical atmospheric profile data and land surface type data of the complex terrain region are obtained; For each channel brightness temperature data in the full-channel brightness temperature dataset, the weighted peak height of that channel is obtained based on the brightness temperature data of that channel and the typical atmospheric profile data of complex terrain areas. Peak values ​​are filtered based on the peak height of several channel weights and the corresponding brightness temperature data of several channels to obtain the first channel subset; Based on the full-channel brightness temperature dataset, typical atmospheric profile data of complex terrain areas, and surface type data, a surface emissivity sensitivity analysis was performed to obtain the second channel subset. The brightness temperature dataset of the channel subset is obtained by taking the intersection of the first channel subset and the second channel subset.

3. The numerical prediction method for complex terrain areas based on FY4B according to claim 2, characterized in that, The surface emissivity sensitivity analysis, based on the full-channel brightness temperature dataset, typical atmospheric profile data of complex terrain areas, and surface type data, yields a second channel subset, including: For each channel brightness temperature data in the full-channel brightness temperature dataset, the basic emissivity value is determined based on the channel brightness temperature data and the surface type data. The first simulated brightness temperature is obtained based on the brightness temperature data of this channel and the basic emissivity value; The perturbation emission rate value is obtained based on the base emission rate value and the preset perturbation amplitude; The second simulated brightness temperature is obtained based on the brightness temperature data of this channel and the perturbation emissivity value; Based on the first simulated brightness temperature and the second simulated brightness temperature, obtain the absolute value of the brightness temperature change corresponding to the brightness temperature data of this channel; Based on a preset change threshold, several channel brightness temperature data and corresponding absolute values ​​of brightness temperature change, the channels are filtered to obtain the second channel subset.

4. The numerical prediction method for complex terrain areas based on FY4B according to claim 1, characterized in that, Based on the aforementioned channel subset brightness temperature dataset, the auxiliary dataset of background field in complex terrain areas, the short-term forecast background field, and the satellite scan angle and cloud mask data in the FY4B satellite real-time observation dataset, dynamic bias correction is performed to obtain the corrected brightness temperature dataset, including: Land surface type data are obtained based on the background field auxiliary dataset of the complex terrain area; Based on the channel subset brightness temperature dataset, surface type data, and satellite scan angle simulation of theoretical brightness temperature, a theoretical brightness temperature dataset is obtained; The observation bias is calculated based on the theoretical brightness temperature dataset and the channel subset brightness temperature dataset to obtain the original observation bias dataset. A multiple linear regression model is constructed based on the original observation bias dataset, satellite scan angle, land surface type data, and cloud mask data. A systematic bias estimate is obtained based on the multiple linear regression model and the original observation bias dataset. Brightness temperature correction is performed based on the original observation bias dataset and the systematic bias estimate to obtain the corrected brightness temperature dataset.

5. The numerical prediction method for complex terrain areas based on FY4B according to claim 1, characterized in that, The cloud-topography quality control is performed based on the corrected brightness temperature dataset, the preset digital elevation model, and the cloud image data and satellite observation geometric information in the FY4B satellite real-time observation dataset to obtain an effective brightness temperature observation dataset, including: Thick cloud identification is performed based on the cloud map data and a preset multispectral threshold algorithm to obtain thick cloud detection results; Based on the cloud map data and the preset local binary pattern texture analysis algorithm, broken and shallow clouds are identified to obtain broken and shallow cloud detection results. Cloud contamination labeling results are obtained based on the thick cloud detection results and the broken, shallow cloud detection results. Based on the satellite observation geometric information and the preset digital elevation model, terrain occlusion is identified to obtain terrain occlusion marking results; Based on the cloud pollution labeling results and terrain occlusion labeling results, data is removed from the corrected brightness temperature dataset to obtain the effective brightness temperature observation dataset.

6. The numerical prediction method for complex terrain areas based on FY4B according to claim 5, characterized in that, The data removal process based on the cloud contamination labeling results and terrain occlusion labeling results in the corrected brightness temperature dataset yields the effective brightness temperature observation dataset, including: The satellite zenith angle is obtained based on the aforementioned satellite observation geometric information; Based on the satellite zenith angle and the preset zenith angle threshold, the modified brightness temperature dataset is initially purged to obtain the initial brightness temperature observation dataset. Based on the cloud pollution labeling results and terrain occlusion labeling results, the initial brightness temperature observation dataset is purged to obtain the effective brightness temperature observation dataset.

7. The numerical prediction method for complex terrain areas based on FY4B according to claim 1, characterized in that, Based on the effective brightness temperature observation dataset and the preset high-frequency cyclic assimilation algorithm, an assimilation three-dimensional analysis field is constructed, including: Get the default observation error covariance matrix; For each observation point in the effective brightness temperature observation dataset, cloud cover data, terrain complexity index, and satellite zenith angle are obtained based on the observation point. The observation error adjustment factor is obtained based on the cloud cover data, terrain complexity index, and satellite zenith angle. The observation error covariance matrix corresponding to the observation point is obtained based on the observation error adjustment factor and the default observation error covariance matrix. Assimilation analysis is performed based on the preset high-frequency cyclic assimilation algorithm, the effective brightness temperature observation dataset, and the observation error covariance matrix corresponding to several observation points to obtain the assimilation three-dimensional analysis field.

8. The numerical prediction method for complex terrain areas based on FY4B according to claim 1, characterized in that, The high-resolution forecasting based on the assimilated three-dimensional analysis field, yielding numerical forecasting results for complex terrain areas, includes: Acquire large-scale boundary condition data and conventional initial fields; High-resolution forecasts are performed based on the large-scale boundary condition data, the assimilated three-dimensional analysis field, and the preset regional numerical forecasting model to obtain high-precision forecast results. High-resolution forecasts are performed based on the large-scale boundary condition data, conventional initial fields, and preset regional numerical forecast models to obtain conventional forecast results. Numerical forecast results for the complex terrain region are obtained based on the high-precision forecast results and the conventional forecast results.

9. A numerical prediction method for complex terrain areas based on FY4B according to claim 1, characterized in that, Also includes: Acquire real-time observation data, and obtain high-precision forecast results and conventional forecast results based on the numerical forecast results of the complex terrain area; Based on the high-precision forecast results, conventional forecast results, and actual observation data, traditional skill scoring is performed to obtain high-precision forecast scoring results and conventional forecast scoring results; Based on the high-precision forecast scoring results and the conventional forecast scoring results, a significant difference analysis was conducted to obtain the key areas for forecast improvement and the key forecast lead time. Based on the forecast improvement key areas and key forecast lead times, high-precision three-dimensional meteorological element fields corresponding to the regions and lead times are extracted from the high-precision forecast results; Based on the forecast improvement key areas and key forecast lead times, the conventional three-dimensional meteorological element fields corresponding to the regions and lead times are extracted from the high-precision forecast results; Based on the high-precision three-dimensional meteorological element field and the conventional three-dimensional meteorological element field, multi-physical quantity collaborative diagnosis is performed to obtain physical mechanism diagnosis results.

10. A numerical prediction system for complex terrain areas based on FY4B, characterized in that, include: The data acquisition module is used to acquire the real-time observation dataset of FY4B satellite, the auxiliary dataset of background field in complex terrain areas, and the short-term forecast background field; The channel filtering module is used to input the full-channel brightness temperature dataset and the background field auxiliary dataset of complex terrain areas from the real-time observation dataset of the FY4B satellite into the preset radiative transfer model, so that the radiative transfer model can perform channel filtering based on the full-channel brightness temperature dataset and the background field auxiliary dataset of complex terrain areas to obtain a channel subset brightness temperature dataset. The dynamic deviation correction module is used to perform dynamic deviation correction based on the channel subset brightness temperature dataset, the background field auxiliary dataset of complex terrain area, the short-term forecast background field, and the satellite scan angle and cloud mask data in the FY4B satellite real-time observation dataset, to obtain the corrected brightness temperature dataset. The quality control module is used to perform cloud-topography quality control based on the corrected brightness temperature dataset, the preset digital elevation model, and the cloud image data and satellite observation geometric information in the FY4B satellite real-time observation dataset, so as to obtain an effective brightness temperature observation dataset. The assimilation analysis module is used to construct an assimilation three-dimensional analysis field based on the effective brightness temperature observation dataset and a preset high-frequency cyclic assimilation algorithm; The numerical prediction module is used to perform high-resolution prediction based on the assimilated three-dimensional analysis field to obtain numerical prediction results for complex terrain areas.

Images