Weather forecasting method based on GNSS-R sea surface wind and meteorological large model
By integrating GNSS-R sea surface wind and European center forecast data, a two-dimensional lattice-based sea surface wind vector data is formed, and a variety of meteorological models are used to integrate forecasts and results, the problem of insufficient accuracy of sea surface wind vector data acquisition and extreme weather forecasts is solved, and more efficient and accurate weather forecasts are achieved.
Patent Information
- Application Number
- CN202411913754.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-24
- Publication Date
- 2025-04-29
AI Technical Summary
The existing technology lacks a fast and effective acquisition scheme for sea surface wind vector data, and the existing wind speed inversion model lacks forecast accuracy in extreme weather, resulting in a large deviation from the actual situation of the meteorological model output results.
By integrating GNSS-R sea surface wind and the 10m vector wind field of the European Center forecast data IFS, a regular two-dimensional lattice-tuned sea surface wind vector data is formed, and the data is input into the Pangu, FourCastNet, and GraphCast meteorological models for global weather forecasting, and finally the forecast results of the three meteorological models are integrated.
It realizes the rapid and effective acquisition of sea surface wind vector data, improves the accuracy of weather forecasts in extreme weather, reduces the deviation of meteorological model forecast results, and improves the real-time prediction capability of weather forecasts.
Smart Images
Figure CN119902308A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to weather forecasting technology in the field of meteorological science and technology, and in particular to a weather forecasting method based on GNSS-R sea surface wind and a large meteorological model. Background Art
[0002] The Earth's atmosphere is a dynamic, complex and ever-changing system. Especially in the context of global warming, the atmospheric system has become more unstable, and extreme weather events have shown an increasing and intensifying trend, with more complex and variable temporal and spatial distributions, which has greatly increased the difficulty of weather forecasting. At present, international weather forecasting services mostly use numerical models and large meteorological models, and the accuracy of numerical models and large meteorological models is heavily dependent on the quality and quantity of observational data. The ocean occupies more than 70% of the global surface area, but the ocean observation data is seriously insufficient, which seriously restricts the development of global weather forecasting.
[0003] In recent years, the global navigation satellite system reflectometry (GNSS-R) remote sensing technology has provided a new way to detect global sea surface winds. The US CYGNSS constellation, China's FY-3 series satellites, China's Tianmu-1 constellation and Yunyao constellation have all used this technology to observe global sea surface winds, greatly enriching ocean observation data. However, unlike observation methods such as buoys, microwave scatterometers, and synthetic aperture radars (SAR), GNSS-R remote sensing technology mainly relies on the time delay and Doppler effect of reflected signals to invert sea surface wind speeds, and these signals are less sensitive to wind direction. Therefore, the sea surface wind data observed by GNSS-R remote sensing technology is in the form of scalar wind speed, that is, only wind speed is provided without wind direction information, that is, sea surface wind data is not a vector. However, the current numerical models and large meteorological model forecasting technologies use vector wind field data, and the use of GNSS-R sea surface wind is limited in weather forecasting because it is scalar data.
[0004] To this end, Wang Feng, Li Jianqiang, Zhang Guodong, Zhang Qi, Yang Dongkai and others published an article entitled Maximum Likelihood Estimation of Sea Surface Wind Vector from Spaceborne Satellite Navigation Reflection Signal Observed from Sub-satellite Points[J](Journal of Electronics and Information Technology, 2024, 46(4):1418-1427.doi:10.11999 / JEIT230464), which proposed a GNSS-R-based sea surface wind vector inversion algorithm. Through multi-satellite observation and maximum likelihood estimation, wind direction information can be extracted from scalar data.
[0005] For another example, the article "GloWS-Net: A Deep Learning Framework for RetrievingGlobal Sea Surface Wind Speed Using Spaceborne GNSS-R Data" published by Bu Jinwei, Yu Kegen, Zuo Xiaoqing, Ni Jun, Li Yongfa, Huang Weimin, etc. proposed a GloWS-Net deep learning framework for retrieving global sea surface wind speed through spaceborne GNSS-R data. In previous studies, delay Doppler maps (DDMs) and empirical geophysical model functions (GMFs) were usually used to extract features to invert wind speed. However, the GMFs method faces challenges when multiple sea state parameters are used as model inputs. Therefore, GloWS-Net considers the fusion of auxiliary information, including wave significant wave height (SWH), sea surface rainfall, and wave direction, and constructs an end-to-end wind speed inversion model.
[0006] Although the above-mentioned existing technologies provide certain ideas, they still have the following problems: 1. There is no fast and effective solution for obtaining sea surface wind vector data; 2. Existing wind speed inversion models are unable to cope with forecasts under extreme weather conditions, especially when real-time data of high wind speeds exceeding 40m / s appears, the output results of the large meteorological model will have a large deviation from the actual situation. Summary of the invention
[0007] A brief summary of embodiments of the present invention is provided below to provide a basic understanding of certain aspects of the present invention. It should be understood that the following summary is not an exhaustive summary of the present invention. It is not intended to identify key or important parts of the present invention, nor is it intended to limit the scope of the present invention. Its purpose is merely to provide certain concepts in a simplified form as a prelude to a more detailed description discussed later.
[0008] In order to solve the above-mentioned combination problem, this solution proposes a weather forecasting method based on GNSS-R sea surface wind and meteorological large models. In view of the problem that GNSS-R sea surface wind only has wind speed but no wind direction, which is difficult to apply in the field of weather forecasting, this solution will integrate GNSS-R sea surface wind and the 10m vector wind field of the European Center Forecast Data IFS to form regular two-dimensional gridded sea surface wind vector data. At the same time, in response to forecasts under extreme weather conditions, this application uses Pangu, FourCastNet, and GraphCast meteorological large models for comprehensive processing, especially for training the Pangu meteorological large model for extreme data such as high wind speed in the training data set, and input the data into the Pangu, FourCastNet, and GraphCast meteorological large models for global forecasting, and integrate the forecast results of these three meteorological large models.
[0009] Specifically, the present invention provides a weather forecasting method based on GNSS-R sea surface wind and meteorological large model, which includes a data fusion process and a meteorological forecasting process combined with artificial intelligence;
[0010] The data fusion process is used to fuse the GNSS-R sea surface wind data with the IFS forecast data from the European Center to generate a two-dimensional gridded sea surface vector wind field. Specifically, it includes:
[0011] Historical data collection and threshold calculation steps: Collect historical GNSS-R sea surface wind and the 10m vector wind field (U 10 ,V 10 ), and perform data analysis and spatiotemporal matching on the two sets of data, calculate the average standard deviation between the two sets of wind speed data, and set the average standard deviation as the threshold;
[0012] Real-time data download and quality control steps: Download real-time GNSS-R sea surface wind and 10m vector wind field in IFS (U 10 ,V 10 ), perform data analysis, time-space matching, and calculate the standard deviation of real-time sea surface wind speed data. If the standard deviation is greater than the threshold, the GNSS-R sea surface wind data of the point is eliminated; quality control is achieved by eliminating data that does not meet the threshold;
[0013] Data fusion steps based on data assimilation method: Use data assimilation method to merge the quality-controlled GNSS-R sea surface wind and the 10m vector wind field (U 10 ,V 10 ) are fused. The data assimilation method is to achieve the best fit of the GNSS-R sea surface wind and the 10m vector wind field of IFS by minimizing the objective function. The objective function is defined as:
[0014]
[0015] Among them, x is the fused two-dimensional gridded sea surface wind speed data, x b is the 10m wind speed data of IFS, (U 10 ,V 10 ) is the 10m vector wind field of IFS, y o is the GNSS-R sea surface wind data, B is the background error covariance matrix, and R is the observation error covariance matrix;
[0016] The longitude and latitude of the fused two-dimensional gridded sea surface wind speed data are matched with the longitude and latitude of the 10m vector wind field in IFS, and the wind direction of the 10m vector wind field in IFS is used as the wind direction of the fused two-dimensional gridded sea surface wind speed data to form a regular two-dimensional gridded sea surface vector wind field.
[0017] The process of combining artificial intelligence to perform weather forecasting includes:
[0018] Meteorological big model forecasting steps: use the fused two-dimensional gridded sea surface vector wind field as input data into the meteorological big model, carry out global weather forecast for a predetermined number of days, obtain ground forecast data and multi-level high-altitude forecast data as global forecast results.
[0019] In the prior art, the use of AI-based meteorological big models alone, because of their strong nonlinear mapping capabilities, well reflects the strong nonlinear characteristics of the atmospheric system. However, in extreme weather conditions (such as strong winds with wind speeds exceeding 40m / s), due to the lack of extreme weather observation data, data-driven meteorological big models face significant interference in the forecast process. Therefore, as a feasible solution, this application adopts three meteorological big models: PanGu meteorological big model, FourCastNet meteorological big model and GraphCast meteorological big model, and reduces the errors that may be caused by a single model by integrating multiple models.
[0020] The weather forecasting process combined with artificial intelligence also includes a forecast data integration processing step: a weather model forecasting step obtains three sets of global forecast results according to the PanGu weather model, the FourCastNet weather model and the GraphCast weather model; respectively calculates the errors of the three sets of global forecast results, assigns a large weight to the weather model with a relatively small forecast error (the PanGu weather model, the FourCastNet weather model or the GraphCast weather model), and assigns a small weight to the weather model with a relatively large forecast error, to obtain a global forecast result:
[0021] F(x)=w p f p (x)+wf f f (x)+w g f g (x);
[0022] Among them, w p 、w f and w g are the weights of the PanGu meteorological model, the FourCastNet meteorological model, and the GraphCast meteorological model, respectively, and w p +w f +w g =1,f p (x), f f (x), f g (x) are the forecast results of PanGu meteorological model, FourCastNet meteorological model and GraphCast meteorological model, respectively, and F(x) is the integrated global forecast result.
[0023] As an implementation scheme, the meteorological big model forecasting step specifically includes: using the fused two-dimensional gridded sea surface vector wind field as input data, inputting it into the PanGu meteorological big model, the FourCastNet meteorological big model and the GraphCast meteorological big model, respectively carrying out global weather forecasts for a certain period of time, and obtaining three sets of global forecast results, including ground forecast data such as 2m temperature, 2m humidity, 10m wind, rain and snow, and sea level pressure, as well as high-altitude forecast data such as temperature, humidity, wind, gravity potential, etc. at multiple levels of different air pressures (1000hpa, 925hpa, 850hpa, 700hpa, 600hpa...).
[0024] As an implementation scheme, the weather forecasting process combining artificial intelligence also includes a three-dimensional weather visualization step to visualize the global forecast results, including meteorological elements such as air pressure, temperature, humidity, rain, snow, and wind.
[0025] At the same time, it was found in the experiment that the Pangu meteorological model performed outstandingly in terms of accuracy and prediction of extreme weather events, the FourCastNet meteorological model had significant advantages in computational efficiency and large-scale ensemble forecasting, and GraphCast performed well in terms of detail texture and comprehensive evaluation. Therefore, in order to further reduce the occurrence of forecast errors caused by abnormal interference in the meteorological model, the meteorological models include the PanGu meteorological model, the FourCastNet meteorological model and the GraphCast meteorological model. When training the meteorological model, the data set is divided into a training data set and a test data set, and the training data set is divided into a normal data set and an abnormal data set (abnormal here refers to data that deviates from the theoretical area, such as data obtained under extreme conditions or a few The weather forecast abnormal data) is classified, and the normal data set, the abnormal data set and the entire training data set are used to train the meteorological model respectively. The model parameters are adjusted during the training process, and the test data set is used for testing and verification, so that the PanGu meteorological model is more suitable for the forecast of abnormal data, and the FourCastNet meteorological model and the GraphCast meteorological model are more suitable for the forecast of normal data; in the meteorological model forecasting step, the PanGu meteorological model, the FourCastNet meteorological model and the GraphCast meteorological model respectively carry out global weather forecasts for a predetermined number of days, and the obtained forecast data are fused (the data under different models can be set with corresponding weights according to the experimental results, and the total weight sum is 1) as the global forecast result to improve the prediction accuracy. The meteorological model composed of the meteorological models optimized and trained separately as above can improve its generalization ability and adaptability.
[0026] This application focuses on the application of GNSS-R sea surface wind in large-scale meteorological model forecasting. By integrating GNSS-R sea surface wind and the 10m vector wind field of the European Center Forecast Data IFS, regular two-dimensional gridded sea surface wind vector data is formed, and the fused sea surface wind speed data is input into the Pangu, FourCastNet, and GraphCast meteorological models for global forecasting.
[0027] In addition, the selection of true wind speed during extreme weather is crucial to the inversion performance. In order to improve the forecast accuracy under extreme conditions, this application specifically adjusts the parameters of the Pangu large model to make it more suitable for extreme weather forecasts, and ultimately improves the real-time prediction capability of weather forecasts by integrating it with the FourCastNet and GraphCast meteorological large models. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] The present invention may be better understood by referring to the following description in conjunction with the accompanying drawings, wherein the same or similar reference numerals are used throughout the drawings to represent the same or similar components. The accompanying drawings, together with the following detailed description, are incorporated in and form a part of this specification and are used to further illustrate the preferred embodiments of the present invention and to explain the principles and advantages of the present invention.
[0029] In the figure:
[0030] Figure 1 The figure is a flow chart of an algorithm according to an embodiment of the present invention. DETAILED DESCRIPTION
[0031] Embodiments of the present invention will be described below with reference to the accompanying drawings. Elements and features described in one drawing or one embodiment of the present invention may be combined with elements and features shown in one or more other drawings or embodiments. It should be noted that for the purpose of clarity, the representation and description of components and processes that are not related to the present invention and are known to those of ordinary skill in the art are omitted in the drawings and descriptions.
[0032] In order to solve the problem that GNSS-R sea surface wind only has wind speed but no wind direction, which makes it difficult to apply it in the field of weather forecasting, this solution will integrate GNSS-R sea surface wind and the 10m vector wind field of the European Center Forecast Data IFS to form regular two-dimensional gridded sea surface wind vector data, and input the data into the Pangu, FourCastNet, and GraphCast meteorological models for global forecasting, and integrate the forecast results of these three meteorological models.
[0033] See also Figure 1 The weather forecasting method based on the GNSS-R sea surface wind and meteorological large model in the embodiment of the present invention includes:
[0034] S1: Download historical data for one year, including GNSS-R sea surface wind data and 10m vector wind field (U 10 ,V 10 )Two sets of information.
[0035] S2: Data processing and threshold determination: Data analysis and time-space matching are performed on the GNSS-R sea surface wind and IFS 10m wind vector data files respectively, and the time average standard deviation between the two sets of data is calculated, and the time average standard deviation is used as the threshold.
[0036] S3: Download real-time data and perform quality control. Download the real-time GNSS-R sea surface wind and the 10m vector wind field data files in IFS, calculate the standard deviation between the real-time GNSS-R sea surface wind and the 10m vector wind field data in IFS, and if the standard deviation is greater than the threshold, remove the GNSS-R sea surface wind data of the point; if the standard deviation is less than or equal to the threshold, retain the GNSS-R sea surface wind data of the point.
[0037] S4: Sea surface wind speed data fusion.
[0038] ① Construct the background error covariance matrix. Calculate the time average of the 10m wind speed data of the IFS for one year of historical data, and then use the following formula to calculate the background error covariance matrix.
[0039]
[0040] in, is the long-term average of the 10m wind speed data of IFS, x b It is the real-time data of 10m wind speed of IFS.
[0041] ②Construct the observation error covariance matrix. The observation error covariance matrix is calculated in the same way.
[0042] ③ Solve the sea breeze fusion data. According to Bayesian theory, only when the objective function J(x) is the minimum, x is the sea surface wind speed fusion data.
[0043]
[0044] Mathematically, the solution x that satisfies min J(x) is The sea surface wind speed fusion data can be solved by minimizing the objective function.
[0045] S5: Sea surface wind direction matching. The longitude and latitude of the two-dimensional gridded sea surface wind speed data fused in step S4 are matched with the longitude and latitude of the 10m vector wind field in IFS, and the wind direction of the 10m vector wind field in IFS is used as the wind direction of the fused two-dimensional gridded sea surface wind speed data to form a regular two-dimensional gridded sea surface vector wind field.
[0046] S6: Meteorological large model forecast. The fused two-dimensional gridded sea surface vector wind field is used as input data and input into the PanGu, FourCastNet, and GraphCast meteorological large models to carry out global weather forecasts for 7 days respectively, and three sets of global forecast results are obtained, including ground forecast data such as 2m temperature, 2m humidity, 10m wind, precipitation, and sea level pressure, as well as high-altitude forecast data such as temperature, humidity, wind, and gravity potential at multiple levels (1000hpa, 925hpa, 850hpa, 700hpa, 600hpa...).
[0047] S7: Evaluation of the forecast results of the meteorological large model. Collect observation data from meteorological ground stations, sounding stations, buoys, etc., extract 2m temperature, 2m humidity, 10m wind, precipitation, sea level pressure, and temperature, humidity, wind of the pressure layer, and calculate the errors between the observation data and the forecast results of the meteorological large model.
[0048] S8: Meteorological large model forecast data integration. The three sets of meteorological large model forecast data are integrated into one set of optimal forecast data results using the following formula, and large weights are assigned to meteorological large model members with relatively small forecast errors, and small weights are assigned to meteorological large model members with relatively large forecast errors.
[0049] F(x)=w p f p (x)+w f f f (x)+w g f g (x);
[0050] Among them, w p 、w f 、w g are the weights of the PanGu, FourCastNet, and GraphCast meteorological models, respectively, and w p +w f +w g =1,f p (x), f f (x), f g (x) are the forecast results of PanGu, FourCastNet and GraphCast meteorological models, respectively, and F(x) is the integrated forecast result.
[0051] S9: 3D meteorological visualization. Visualize the forecast results of the optimal model in S8, including meteorological elements such as air pressure, temperature, humidity, rain, snow, and wind.
[0052] As another specific embodiment, in order to further reduce the occurrence of forecast errors caused by abnormal interference of the meteorological large model, the step of training the meteorological large model is also included: during training, the data set is divided into a training data set and a test data set, and the training data set is divided into a normal data set and an abnormal data set (abnormal here refers to data that deviates from the theoretical forecast, such as data obtained under extreme conditions or a few abnormal weather forecast abnormal data), and the normal data set, the abnormal data set and the entire training data set are used to train the meteorological model respectively, and the model parameters are adjusted during the training process, and the test data set is used for testing and verification, so that the PanGu meteorological large model is more suitable for the forecast of abnormal data, and the FourCastNet meteorological large model and the GraphCast meteorological large model are more suitable for the forecast of normal data; in the meteorological large model forecasting step, the PanGu meteorological large model, the FourCastNet meteorological large model and the GraphCast meteorological large model respectively carry out global weather forecasts for a predetermined number of days, and the obtained forecast data are fused. The data under different models can be set with corresponding weights according to the experimental results, and the total weight sum is 1. The fused data is used as the global forecast result to improve the prediction accuracy. The large meteorological model composed of the above-mentioned separately optimized and trained meteorological models can improve its generalization ability and adaptability. In addition, generative adversarial networks (GANs) and reinforcement learning techniques can be introduced to enhance the model's ability to handle complex climate dynamics.
[0053] Compared with the prior art, this application has the following advantages:
[0054] 1. Vector and scalar data fusion: To address the problem that GNSS-R sea surface wind data only has wind speed but no wind direction, this application uses data assimilation technology to fuse GNSS-R sea surface wind and the 10m vector wind field of the European Center Forecast Data IFS to form regular two-dimensional gridded sea surface wind vector data.
[0055] 2. Forecasts of multiple large meteorological models: In view of the application of GNSS-R sea surface wind data in the field of meteorological forecasting, this application inputs the two-dimensional gridded sea surface wind vector data containing GNSS-R sea surface wind information into multiple large meteorological models (PanGu, FourCastNet, GraphCast), and conducts global weather forecasts for a certain period of time to obtain multiple sets of global forecast results.
[0056] 3. Integration of multiple large meteorological models: Integrate the global forecast results of multiple large meteorological models, and adjust the parameters of the PanGu meteorological model to obtain a forecast model that is more suitable for extreme weather. Rationally use the effective forecast information of different models, reduce forecast deviations and thus improve forecast accuracy.
[0057] It should be emphasized that the term “include / comprises” when used herein refers to the presence of features, elements, steps or components, but does not exclude the presence or addition of one or more other features, elements, steps or components.
[0058] In addition, the method of the present invention is not limited to being executed in the time sequence described in the specification, and may also be executed in other time sequences, in parallel or independently. Therefore, the execution order of the method described in this specification does not limit the technical scope of the present invention.
[0059] Although the present invention has been disclosed above by describing specific embodiments of the present invention, it should be understood that all the above embodiments and examples are exemplary rather than restrictive. Those skilled in the art may design various modifications, improvements or equivalents of the present invention within the spirit and scope of the appended claims. These modifications, improvements or equivalents should also be considered to be included in the protection scope of the present invention.
Claims
1. A weather forecast method based on GNSS-R sea surface wind and meteorological large model, characterized by: Including data fusion process and weather forecasting process combined with artificial intelligence; The data fusion process is used to fuse the GNSS-R sea surface wind data with the IFS forecast data from the European Center to generate a two-dimensional gridded sea surface vector wind field. Specifically, it includes: Historical data collection and threshold calculation steps: Collect historical GNSS-R sea surface wind and the 10m vector wind field (U 10 ,V 10 ), and perform data analysis and spatiotemporal matching on the two sets of data, calculate the average standard deviation between the two sets of wind speed data, and set the average standard deviation as the threshold; Real-time data download and quality control steps: Download real-time GNSS-R sea surface wind and 10m vector wind field in IFS (U 10 ,V 10 ), perform data analysis, time-space matching, and calculate the standard deviation of real-time sea surface wind speed data. If the standard deviation is greater than the threshold, the GNSS-R sea surface wind data of the point is eliminated; quality control is achieved by eliminating data that does not meet the threshold; Data fusion steps based on data assimilation method: Use data assimilation method to merge the quality-controlled GNSS-R sea surface wind and the 10m vector wind field (U 10 ,V 10 ) are fused. The data assimilation method is to achieve the best fit of the GNSS-R sea surface wind and the 10m vector wind field of IFS by minimizing the objective function. The objective function is defined as: Among them, x is the fused two-dimensional gridded sea surface wind speed data, x b is the 10m wind speed data of IFS, (U 10 ,V 10 ) is the 10m vector wind field of IFS, y o is the GNSS-R sea surface wind data, B is the background error covariance matrix, and R is the observation error covariance matrix; The longitude and latitude of the fused two-dimensional gridded sea surface wind speed data are matched with the longitude and latitude of the 10m vector wind field in IFS, and the wind direction of the 10m vector wind field in IFS is used as the wind direction of the fused two-dimensional gridded sea surface wind speed data to form a regular two-dimensional gridded sea surface vector wind field; The process of combining artificial intelligence to perform weather forecasting includes: Meteorological big model forecasting steps: use the fused two-dimensional gridded sea surface vector wind field as input data into the meteorological big model, carry out global weather forecast for a predetermined number of days, obtain ground forecast data and multi-level high-altitude forecast data as global forecast results.
2. The weather forecasting method based on the GNSS-R sea surface wind and meteorological large model according to claim 1, characterized in that: The meteorological big models include the PanGu meteorological big model, the FourCastNet meteorological big model and the GraphCast meteorological big model.
3. The weather forecasting method based on the GNSS-R sea surface wind and meteorological large model according to claim 2 is characterized in that: The weather forecasting process combined with artificial intelligence also includes a forecast data integration processing step: a weather model forecasting step obtains three sets of global forecast results based on the PanGu weather model, the FourCastNet weather model and the GraphCast weather model; calculates the errors of the three sets of global forecast results respectively, assigns a large weight to the weather model with a relatively small forecast error, and assigns a small weight to the weather model with a relatively large forecast error, to obtain a global forecast result: F(x)=w p f p (x)+w f f f (x)+w g f g (x); Among them, w p 、w f and w g are the weights of PanGu meteorological model, FourCastNet meteorological model and GraphCast meteorological model respectively, and w p +w f +w g =1,f p (x), f f (x), f g (x) are the forecast results of PanGu meteorological model, FourCastNet meteorological model and GraphCast meteorological model, respectively, and F(x) is the integrated global forecast result.
4. The weather forecasting method based on the GNSS-R sea surface wind and meteorological large model according to claim 2 is characterized in that: The meteorological big model forecasting step specifically includes: using the fused two-dimensional gridded sea surface vector wind field as input data, inputting it into the PanGu meteorological big model, the FourCastNet meteorological big model and the GraphCast meteorological big model, respectively carrying out global weather forecasts for a certain period of time, and obtaining three sets of global forecast results.
5. The weather forecasting method based on GNSS-R sea surface wind and meteorological large model according to claim 1, characterized in that: The process of combining artificial intelligence with weather forecasting also includes a three-dimensional weather visualization step to visualize global forecast results.
6. The weather forecasting method based on the GNSS-R sea surface wind and meteorological large model according to claim 2, characterized in that: When training the big meteorological model, the data set is divided into a training data set and a test data set. The training data set is classified into a normal data set and an abnormal data set. The normal data set, the abnormal data set and the entire training data set are used to train the meteorological model respectively. The model parameters are adjusted during the training process and tested and verified with the test data set, so that the PanGu meteorological model, the FourCastNet meteorological model and the GraphCast meteorological model are more suitable for accurate weather forecasting.