An ultra-rapid numerical weather prediction method

Abstract

This invention belongs to the field of weather forecasting technology and relates to an ultra-fast numerical weather prediction method. It includes regional numerical model compilation and setup, acquisition of global numerical weather prediction data products, processing and preprocessing of multi-source meteorological observation data, fusion and assimilation analysis of multi-source meteorological observation data, numerical model integral calculation, post-processing of numerical weather prediction products, rapid updating of numerical weather predictions, comparative verification of the assimilation and forecasting effects of different data, selection of observation data with shorter latency for assimilation application and numerical integral calculation, and establishment of an operational ultra-fast numerical weather prediction system. This invention establishes an ultra-fast numerical weather prediction system that, while ensuring forecast accuracy, further shortens the generation time of numerical weather predictions, thereby improving the timeliness of regional model forecast products and the ability to forecast and warn of severe weather.

Description

Technical Field

[0001] This invention belongs to the field of weather forecasting technology and relates to a method for ultra-fast numerical weather forecasting. Background Technology

[0002] With the development of numerical weather prediction technology and the improvement of data assimilation levels, the accuracy of numerical weather prediction has been increasing, playing an increasingly important role in weather forecasting operations. Currently, provincial weather forecasting operations can receive numerical weather predictions including global and regional numerical weather predictions, providing hourly forecasts of precipitation, temperature, wind, and other elements for at least three days twice daily. Meanwhile, to improve the forecasting capability for rapidly developing severe convective weather and other hazardous weather events, national, regional, and provincial meteorological centers have established rapid-update numerical weather prediction systems in recent years, capable of cyclically updating forecasts of future weather every 1-3 hours. The core idea of ​​these rapid-update numerical weather prediction systems is to rapidly assimilate multi-source observational data to generate the latest numerical model initial field, allowing the model to recalculate the integral and obtain the latest numerical forecast product. Therefore, scientists have long been committed to the quality control of multi-source meteorological data and the development of better data assimilation methods, aiming to better utilize various types of observational data and obtain more accurate model initial fields.

[0003] In current numerical weather prediction (NMR) operations, rapid update NMR systems based on multi-source observation data typically assimilate and analyze various types of effective observation data into the model's initial field. This data includes conventional automatic weather stations, radiosondes, aircraft reports, GPS / MET water vapor data, radar data, satellite data, and wind profilers. The timeliness of these data arrivals on the operational platform varies; some data have short lag times, while others have longer lag times. For example, due to the scarcity of upper-air observation data, although radiosonde data assimilation is crucial for NMR, the availability of nationwide radiosonde data in actual operations usually lags by more than one hour. This results in a one-hour delay in initiating data processing and assimilation for rapid update NMR models. Furthermore, if all types of observation data are to be assimilated and applied to the model, involving data preprocessing, data assimilation analysis, model integral forecasting, and model product post-processing, the resulting NMR products from the rapid update cycle typically arrive at the forecaster's desktop with a 3-5 hour delay after the entire process is completed. For national or regional center-guided products, data transmission time issues cause a longer delay in reaching forecasters' desktops. For example, models that start forecasting at 8:00 AM typically don't provide forecasts to forecasters until after 12:00 PM, or even later. Even rapidly updated numerical weather prediction models often fail to provide timely updates on sudden severe weather events. Currently, no ideal solution specifically addressing this problem has been reported. Summary of the Invention

[0004] This invention addresses the problem of long delays in traditional rapid-update numerical weather prediction systems. By combining the characteristics of numerical weather prediction, the time it takes for multi-source observation data to reach the operational platform, and the data assimilation effect, an ultra-fast-update numerical weather prediction method is designed. This method can further shorten the generation time of numerical weather predictions while ensuring the accuracy of the predictions, thereby improving the timeliness and availability of numerical weather prediction products.

[0005] To achieve the above objectives, the present invention is implemented using the following technical solution: A method for ultra-fast numerical weather forecast updates includes the following steps: (1) Regional numerical mode compilation settings A regional numerical model is selected. Based on a supercomputing cluster, each module of the numerical model is compiled and debugged. The latitude and longitude range and model grid of the regional model are set. Through the static data processing module, the terrain and underlying surface data that are associated with or are not associated with the regional numerical model are processed into the grid of the regional numerical model (this grid itself is three-dimensional, but the terrain and underlying surface data are only two-dimensional), and the terrain and underlying surface data that can be used by the regional model are produced.

[0006] (2) Obtaining global numerical weather prediction data products The system acquires global numerical weather prediction (GNMR) products from the China Meteorological Administration or foreign meteorological agencies (including data from the "Tianqing" big data cloud platform of the meteorological data center, the meteorological numerical forecast cloud sharing platform, and the Internet). The regional model background field processing module processes the GNMR data to the local regional numerical model's 3D grid range, generating the regional numerical model's 3D initial field (background field) and lateral boundaries. The initial field data can be used for data assimilation or directly for model integration calculations.

[0007] (3) Processing and preprocessing of multi-source meteorological observation data Relying on the "Tianqing" big data cloud platform of the meteorological data center, multi-source meteorological observation data are collected. The data from different sources and of different natures undergo format conversion and processing. The data processing includes multi-step checks on the observation data, quality control before assimilation, and assigning quality labels to each observation without altering the original data values, thereby improving data utilization and effectiveness. The processed data generates standardized format data that can be assimilated and applied to regional numerical models.

[0008] (4) Fusion and assimilation analysis of multi-source meteorological observation data Using data assimilation analysis technology, based on the generated three-dimensional initial field (background field), a certain data assimilation time window is set, and multi-source meteorological observation data within the time window are assimilated and analyzed to correct the three-dimensional initial field and update the corresponding side boundary, forming a three-dimensional fusion assimilation analysis field of the regional numerical model, which serves as the new initial field and side boundary of the regional numerical model.

[0009] (5) Numerical model integral calculation Based on the initial field and lateral boundary conditions of the three-dimensional fusion analysis of the regional numerical model updated in step (4), suitable microphysical parameterization schemes, cumulus convection parameterization schemes, boundary layer schemes, etc., for local severe weather forecasting are selected from the CMA-MESO model. The numerical model is integrated to predict the atmospheric state for a period of time in the future and obtain future weather forecast products, namely regional numerical weather forecast products, including three-dimensional atmospheric temperature, air pressure, wind direction and speed, relative humidity, precipitation and other meteorological elements and physical fields, etc. The output time interval of the numerical weather forecast products is set as needed.

[0010] (6) Post-processing of numerical weather forecast products Based on user needs, regional numerical weather prediction products are processed, including data format conversion, interpolation, and physical quantity diagnostic calculations, to generate meteorological element forecasts for any location within the model's forecast range. The products are then post-processed through methods such as plotting to generate the detailed forecast products required by the user.

[0011] (7) Conduct rapid updates to numerical weather forecasts Set the frequency of rapid updates for the regional numerical model (e.g., 1 hour, 3 hours). Based on the regional numerical model forecast field in step (5), use it as the new initial field and boundary conditions. Based on the latest observation data, repeat steps (3)-(5) to perform rapid updates for numerical weather forecasts. Continue in this manner until the next new global numerical weather forecast product is generated. Then return to step (2) and repeat the following steps (3)-(5) to perform rapid updates for numerical weather forecasts.

[0012] (8) Conduct comparative tests on the forecasting effects of different data assimilation methods. The delay time of different types of meteorological data was analyzed, and meteorological observation data with less delay was selected. Different data assimilation time windows and assimilation observation data for rapid update numerical models were determined. Steps (4)-(6) were repeated. Data assimilation and numerical integration were performed on these different types of "rapid" arrival data. The results of different rapid update numerical weather forecasts were tested and evaluated. The meteorological elements of the numerical models tested included precipitation, wind speed, and temperature. The testing methods included TS score, average error, average absolute error, and root mean square error. The forecast performance of different rapid update numerical models was evaluated.

[0013] (9) Establish an operational ultra-fast updated numerical weather forecasting system Following the evaluation results in step (8), and without compromising the forecasting effectiveness of the rapid update numerical model, the data assimilation time window and the assimilated observation data for the ultra-rapid update numerical model are determined, and operational application is carried out in step (4). Operational scripts are built according to steps (2)-(7). Based on the generation time of global numerical weather prediction products and the delay time of observation data, the scheduled start time of the operational scripts for the ultra-rapid update numerical model is determined, and an operational ultra-rapid update numerical weather prediction system is established.

[0014] Preferably, the initial field mentioned in step (2) refers to the state of the atmosphere at the initial moment, such as the three-dimensional atmospheric temperature, pressure, humidity and wind values.

[0015] Preferably, the multi-source meteorological observation data mentioned in step (3) usually includes observation data from ground automatic weather stations, radiosondes, aircraft reports, GPS / MET water vapor, wind profiler radar, weather radar, satellite, and other numerical forecast models.

[0016] As a preferred option, the quality control of observation data in step (3) usually includes data extreme value checks, inter-element consistency checks, vertical consistency checks of multi-layer observation data (temperature lapse rate checks, abnormal temperature inversion checks, static checks, vertical interpolation checks, wind shear checks), background field checks, horizontal consistency checks, temporal consistency checks, continuity checks, sparsification, deviation correction, etc.

[0017] As a preferred option, the data assimilation analysis technique mentioned in step (4) usually refers to using variational assimilation, hybrid assimilation and other methods to mathematically optimize the fusion of multi-source observation data (such as satellite, radar and ground station data) of different times and properties with the background field of numerical models, so as to generate a more accurate and physically more consistent three-dimensional atmospheric "analysis field" as a new initial field, making the initial field closer to the actual state of the atmosphere.

[0018] As a preferred option, the data assimilation analysis technique described in step (4) also includes other coupling techniques, artificial intelligence, and other technical methods.

[0019] As a preferred option, the data assimilation time window mentioned in step (4) refers to a specific time interval in numerical weather prediction where the system collects and integrates observation data in order to obtain the optimal initial field required for model integration. For example, 3 hours before and after the assimilation time. Observation data within this time range can be used by the numerical model for assimilation analysis to correct the model's initial field.

[0020] As a preferred option, the variables that are modified and assimilated in step (4) to correct the initial field typically include horizontal wind field, dimensionless air pressure, potential temperature, specific humidity, etc.

[0021] As a preferred option, the selection of suitable microphysical parameterization schemes, cumulus convection parameterization schemes, boundary layer schemes, etc. in step (5) refers to selecting suitable physical parameters for local severe weather forecasting from different microphysical parameterization schemes, cumulus convection parameterization schemes, boundary layer schemes, etc. provided by the CMA-MESO model, based on the characteristics of local severe weather, through sensitivity numerical experiments on the corresponding parameters, or in combination with relevant research results.

[0022] Preferably, the model integration calculation in step (5) refers to the computer calculating the state of the entire three-dimensional atmospheric field after the first time step according to the numerical model (equation) and the initial field, based on the set time step. Then, using this field as the new "initial condition", it is substituted into the equation to calculate the state of the three-dimensional atmosphere after the second time step. This process is repeated to predict the atmospheric state for a period of time in the future and obtain future weather forecasts.

[0023] Preferably, the future period mentioned in step (5) usually refers to the forecast for the next 3 days, and the forecast period can be extended or shortened as needed.

[0024] Preferably, the data delay time mentioned in step (8) refers to the difference between the actual time when the observed data arrives at the database and the observation time.

[0025] Preferably, the calculation formula for the TS score test in step (8) is as follows: .

[0026] In the formula, NA is the number of grid points that are correctly predicted. If the forecast predicts weather of a certain intensity or above (such as precipitation, temperature, wind speed, etc.), and the actual weather of that intensity or above (such as precipitation, temperature, wind speed, visibility, etc.) also occurs, it is recorded as a correct forecast.

[0027] NB represents the number of false alarm grid points. If a forecast indicates weather conditions of a certain intensity or higher (such as precipitation, temperature, wind speed, visibility, etc.), but the actual weather conditions of that intensity or higher (such as precipitation, temperature, wind speed, visibility, etc.) do not occur, it is recorded as a false alarm.

[0028] NC represents the number of missed data points. If a weather event of a certain intensity or above (such as precipitation, temperature, wind speed, etc.) is not predicted but actually occurs, it is recorded as a missed data point.

[0029] The definitions of NA, NB, and NC are shown in Table 1.

[0030] Table 1 Definitions of NA, NB, and NC

[0031] The average error is the average of the errors between the forecast and the actual situation. The verification formula is as follows: .

[0032] In the formula: ME – Forecast Mean Error; NF – Total number of forecasts; i—Forecast frequency indicator; Fi — the i-th forecast value; Oi — the i-th actual value.

[0033] The mean absolute error is the average of the absolute values ​​of the errors between the forecast and the actual situation. The verification formula is as follows: .

[0034] In the formula: MAE – Mean Absolute Error of Forecast; NF – Total number of forecasts; i—Forecast frequency indicator; Fi — the i-th forecast value; Oi — the i-th actual value.

[0035] The root mean square error (RMSE) is the square root of the ratio of the square of the error between the forecast and the actual result to the total number of forecasts. The verification formula is as follows: .

[0036] In the formula: RMSE – Root Mean Square Error of Forecast; NF – Total number of forecasts; i—Forecast frequency indicator; Fi — the i-th forecast value; Oi — the i-th actual value.

[0037] Compared with the prior art, the advantages and positive effects of the present invention are as follows: This invention designs an ultra-fast numerical weather prediction method. Based on a supercomputing cluster, a regional numerical model is compiled and debugged to acquire global numerical weather prediction products and multi-source meteorological observation data. A specific data assimilation method is used to perform assimilation analysis and numerical model integration calculations on the multi-source observation data to predict weather for a future period. Experiments are conducted on the assimilation effects of different observation data. Based on the different assimilation effects and latency times, a reasonable data assimilation time window is set. Observation data with shorter latency and better assimilation effects are selected for assimilation application, establishing an ultra-fast numerical weather prediction system. While ensuring forecast accuracy, this system further shortens the generation time of numerical weather prediction products, thereby improving the timeliness of regional numerical model weather prediction products and the ability to forecast and warn of severe weather. Attached Figure Description

[0038] Figure 1 Comparison of TS scores (%) for general precipitation (1-hour precipitation 1-20 mm) between Ultra-Rapid Update Numerical Weather Prediction (URUC) and Rapid Update Numerical Weather Prediction (RUC) from July 1 to 10, 2024.

[0039] Figure 2 The mean absolute error (m·s) of hourly surface 10-meter wind speed forecasts from the Ultra-Rapid Update Numerical Weather Prediction (URUC) and Rapid Update Numerical Weather Prediction (RUC) for July 1-10, 2024. -1 )contrast.

[0040] Figure 3 Comparison of the mean absolute error (°C) of hourly 2-meter surface temperature forecasts from Ultra-Rapid Update Numerical Weather Prediction (URUC) and Rapid Update Numerical Weather Prediction (RUC) for July 1-10, 2024. Detailed Implementation

[0041] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described below with reference to specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0042] Numerous specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways than those described herein, and therefore the invention is not limited to the specific embodiments disclosed in the following specification. Example 1

[0043] This embodiment is based on the China Meteorological Administration's Regional Numerical Weather Prediction System (CMA-MESO) and has locally built a 1km resolution hourly ultra-fast updating numerical weather prediction system (CMA-MESO-SD) for the Shandong region on the "Shanhe" cluster of the National Supercomputing Center in Jinan. The following steps were taken: (1) Compilation settings for regional numerical weather prediction system Based on the CMA-MESO regional numerical model and the "Shanhe" cluster at the National Supercomputing Center in Jinan, this study compiles the static data processing, data preprocessing, data assimilation, and model integration calculation modules inherent in the regional numerical model. A CMA-MESO-SD 1km numerical weather prediction system is built, centered on Shandong Province, with a horizontal range of approximately 1500×1000km. 2The model has 71 vertical layers. Based on the global 30-second terrain data provided by CMA-MESO and the MODIS satellite remote sensing ground static data, the static data of terrain and underlying surface used in the model is processed by the CMA-MESO model static data processing module to generate terrain and underlying surface data of the CMA-MESO-SD model regional grid.

[0044] (2) Obtaining global numerical weather prediction data products The CMA-GFS global numerical weather prediction products from the China Meteorological Administration and the US GFS global numerical weather prediction products were obtained from the "Tianqing" big data cloud platform of the Shandong Provincial Meteorological Data Center, with the two sets of data serving as backups for each other. The global numerical weather prediction product data were interpolated and processed into the three-dimensional grid range of the CMA-MESO-SD numerical model using the CMA-MESO background field processing module, generating the three-dimensional initial field (background field) and lateral boundary field of the CMA-MESO-SD numerical model.

[0045] (3) Processing and preprocessing of multi-source meteorological observation data Relying on the "Tianqing" big data cloud platform of the Shandong Provincial Meteorological Data Center, multi-source meteorological observation data from ground automatic stations, radiosondes, aircraft reports, GPS / MET water vapor, weather radar, satellites, and wind profiler radar were collected. These meteorological observation data underwent quality control before assimilation. Specifically, ground automatic station data selected key ground stations and underwent extreme value checks to remove outliers and perform data quality control; radiosonde and aircraft report data underwent bias correction quality control; radar combined reflectivity data underwent data smoothing, filtering, and missing data filling, non-meteorological echo quality control (including removal of ground object echoes, biological echoes, and electromagnetic interference echoes); radar radial wind data underwent range folding and velocity ambiguity removal to achieve quality control of radar reflectivity and radial wind data; and wind profiler radar data underwent vertical interpolation and sparsification quality control. After quality control processing of various observation data, a standardized data format suitable for assimilation and application in the CMA-MESO-SD numerical model was generated.

[0046] (4) Fusion and assimilation analysis of multi-source meteorological observation data A three-dimensional variational assimilation (3dvar) system was adopted, and the assimilation time window of the model data was set to half an hour. Based on the generated three-dimensional initial field (background field), the multi-source meteorological observation data within the assimilation window were assimilated and analyzed. The horizontal wind field (u, v), dimensionless air pressure π, potential temperature θ, and specific humidity q in the initial field (background field) were updated to form the three-dimensional fusion assimilation analysis field of CMA-MESO-SD numerical model, which served as the new initial field. At the same time, the corresponding lateral boundary conditions were updated.

[0047] Three-dimensional variational assimilation primarily considers how to interpolate observations from spatially irregular points to grid points under simple dynamic constraints. Its basic objective is to find an optimal analysis field x (representing atmospheric state variables such as temperature, pressure, and wind field) that achieves the best trade-off between the background field (short-term forecast) and the observations, while simultaneously satisfying atmospheric physical constraints. This problem is formulated as minimizing a cost function (or objective function) J(x). Its objective function is: .

[0048] in: x: State vector (the unknown quantity to be solved), representing the meteorological variables at all grid points in the region at a certain moment.

[0049] x b Background field (prior information), derived from global numerical weather prediction or short-term numerical weather prediction.

[0050] y: Observation vector (actual observation data).

[0051] B: Background error covariance matrix. It determines the reliability of background field information and how to propagate (smooth) observation information in space and between different variables. It is one of the most critical matrices in three-dimensional variational assimilation.

[0052] R: Observation error covariance matrix, which includes instrument error and representativeness error.

[0053] H: Observation operator. It maps the model state space to the observation space (e.g., interpolating temperature and pressure at grid points to observation locations, or calculating satellite radiance temperature). When H is nonlinear, the problem requires an iterative solution.

[0054] T denotes the transpose of a matrix.

[0055] (5) Numerical model integral calculation Based on the updated three-dimensional initial field and lateral boundary conditions, the WSM6 microphysical parameterization scheme and NMRF boundary layer scheme, which are suitable for heavy precipitation forecasting in Shandong, were selected. The cumulus convection parameterization scheme was not used. The CMA-MESO-SD model was integrated to provide numerical weather forecasts for the next 12 hours. The forecast products include three-dimensional atmospheric temperature, air pressure, wind direction and speed, relative humidity, precipitation, and other meteorological elements and physical quantities. The forecasts are output as model surface products (58 physical quantities) and isobaric surface products (32 physical quantities), respectively. The numerical forecast products are output at 1-hour intervals, and the time resolution of surface elements is set to high-frequency output with a time interval of 5 minutes.

[0056] (6) Post-processing of numerical weather forecast products Based on the needs of weather forecasting and service operations, the CMA-MESO-SD numerical weather prediction products underwent format conversion, element diagnostic calculations, and data processing. First, the model surface products (58 physical quantities) and isobaric surface products (32 physical quantities) were converted from binary to GRIB2 code for data compression, facilitating the transfer of historical data. Simultaneously, 1km grid meteorological element forecasts for the Shandong region were generated in both text and GRIB2 code formats. These forecasts included precipitation, convective precipitation, longwave radiation, shortwave radiation, water vapor, 2-meter temperature, surface 10-meter wind direction and speed, combined reflectivity, total cloud cover, upper cloud cover, middle cloud cover, lower cloud cover, gusts, visibility, and high-level surface forecast data products such as wind direction and speed at different altitudes, specific humidity, temperature, relative humidity, and visibility, facilitating intelligent grid forecasting operations and service applications. Finally, Python and GrADS software were used to perform plotting processing on the numerical weather prediction products, generating detailed forecast products required by forecasters, which were then published on the website.

[0057] (7) Conduct rapid updates to numerical weather forecasts The frequency of rapid updates for CMA-MESO-SD numerical weather forecasts is set to 1 hour. Based on the 1-hour forecast field of the numerical model in step (5), the new initial field is used. Steps (3)-(5) are repeated to perform rapid updates of numerical weather forecasts hourly. This process is repeated until the next new global numerical weather forecast product is generated. Then, step (2) is returned, and the subsequent steps (3)-(5) are repeated to perform rapid updates of numerical weather forecasts. Due to the long time it takes for radiosonde and GPS / MET water vapor data to reach the operational platform, the rapid update of the numerical model needs to start forecasting 1.5 hours later.

[0058] (8) Conduct comparative tests on the forecasting effects of different data assimilation methods. The delay times of different types of meteorological data were analyzed, and data with shorter delays were selected as the data to be assimilated for ultra-fast update numerical weather prediction. These included data from automatic weather stations, radial wind from weather radar, aircraft reports, and wind profiler radar. Different data assimilation time windows were set, and steps (4)-(6) were repeated. Data assimilation and numerical integration were performed on these different types of "rapidly" arriving data to conduct ultra-fast update numerical weather prediction. The results of different rapid update numerical weather predictions were then tested and evaluated. The meteorological elements tested included precipitation, wind speed, and temperature. The testing methods included TS score and mean absolute error. The forecast performance of different rapid update numerical models was evaluated. The test results (see Tables 2-4) show that when only data from automatic weather stations, radial wind from weather radar, aircraft reports, and wind profiler radar were assimilated for ultra-fast update prediction, the model's short-term forecast was still better than the rapid update prediction that assimilated all observation data.

[0059] (9) Establish an operational system for ultra-fast numerical weather forecasting. Following the evaluation results in step (8), without compromising the forecasting performance of the CMA-MESO-SD rapid update numerical model, data from ground automatic weather stations, radial wind from weather radar, aircraft reports, and wind profiler radar were selected as the data for CMA-MESO-SD ultra-rapid update numerical weather forecast assimilation. The data assimilation time window was set at 15 minutes, and operational application was carried out in step (4). Following steps (2)-(7), operational scripts were built, and an operational CMA-MESO-SD ultra-rapid update numerical weather forecast system was established. The ultra-rapid update numerical model started forecasting 20 minutes after the hour, which is more than one hour earlier than the traditional rapid update numerical model weather forecast product generation time.

[0060] The hourly precipitation, surface wind speed at 10 meters per second, and surface air temperature at 2 meters per second were verified for the CMA-MESO-SD rapid update numerical weather forecast (RUC) and the CMA-MESO-SD ultra-rapid update numerical weather forecast (URUC) from 08:00 to 18:00 daily from July 1 to 10, 2024. The results are shown in Tables 2-4.

[0061] Table 2. Comparison of TS scores (%) for general precipitation (1-hour precipitation 1-20 mm) from Ultra-Rapid Update Numerical Weather Prediction (URUC) and Rapid Update Numerical Weather Prediction (RUC) from July 1 to 10, 2024. Forecast lead time 2h 3h 4h 5h 6h 7h 8h 9h 10h RUC 46.4 53.7 45.8 38.3 38.3 43.0 47.5 51.4 51.9 URUC 62.6 56.3 50.7 48.3 48.7 52.7 52.2 53.0 54.2 Table 3. Mean absolute error (m·s) of hourly surface 10-meter wind speed forecasts from Ultra-Rapid Update Numerical Weather Prediction (URUC) and Rapid Update Numerical Weather Prediction (RUC) from July 1 to 10, 2024 -1 )contrast Forecast lead time 2h 3h 4h 5h 6h 7h 8h 9h 10h RUC 1.04 1.15 1.19 1.25 1.31 1.29 1.28 1.24 1.23 URUC 0.97 1.17 1.18 1.18 1.26 1.26 1.22 1.19 1.18 Table 4. Comparison of the mean absolute error (°C) of hourly 2-meter surface temperature forecasts from Ultra-Rapid Update Numerical Weather Prediction (URUC) and Rapid Update Numerical Weather Prediction (RUC) from July 1 to 10, 2024. Forecast lead time 2h 3h 4h 5h 6h 7h 8h 9h 10h RUC 1.47 1.63 1.75 1.91 1.84 1.62 1.49 1.37 1.21 URUC 0.88 1.12 1.19 1.25 1.15 1.04 0.99 0.96 0.97 To more intuitively represent the accuracy of each product, the data in Tables 2-4 are plotted as graphs. Figures 1-3 ), combining Tables 2-4 Figures 1-3 The results show that the CMA-MESO-SD Ultra-Fast Update Numerical Weather Prediction System (URUC) has a better accuracy (TS score) for hourly precipitation forecasts within 10 hours than the Rapid Update Numerical Weather Prediction System (RUC). The overall forecast errors for hourly 10-meter surface wind speed and 2-meter surface temperature are also smaller than those of RUC. This indicates that the method is reasonable and effective and has important significance for short-term forecasting of severe weather.

[0062] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments for application in other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for ultra-fast updating numerical weather forecasts, characterized in that, The steps are as follows: (1) Select a regional numerical model, compile and debug each module of the regional numerical model based on the supercomputing cluster, set the latitude and longitude range and model grid of the regional numerical model, and process the terrain and underlying surface data onto the grid of the regional numerical model through the static data processing module to obtain the terrain and underlying surface data applicable to the regional numerical model. (2) Obtain global numerical weather prediction products and process the data of global numerical weather prediction products onto the regional numerical model grid through the regional model background field processing module to generate the side boundary of the regional numerical model and the three-dimensional initial field for data assimilation and model integration. (3) Collect meteorological observation data from multiple sources, convert and process meteorological observation data from different sources and of different natures to generate standardized format data that can be assimilated and applied by regional numerical models; (4) Using data assimilation analysis technology, based on the three-dimensional initial field, a data assimilation time window is set, and data assimilation analysis is performed on the standard format data of multi-source meteorological observation data within the time window to correct the three-dimensional initial field and lateral boundary, forming a new initial field A and lateral boundary A; (5) Based on the new initial field A and lateral boundary A, the numerical model is integrated to predict the atmospheric state for a future set time period and obtain future weather forecast products. (6) Process future weather forecast products to generate meteorological element forecasts for any location within the model forecast range, and generate refined forecast products through post-processing of the products; (7) Set the frequency of rapid updates of the numerical model, determine the time period corresponding to the frequency, and repeat steps (3)-(5) based on the new initial field A and lateral boundary A in step (5). Continue to update the numerical weather forecast rapidly until the new time period is reached. Then return to step (2) and repeat steps (3)-(5) to perform rapid updates. (8) Analyze the delay time of different types of meteorological data, select meteorological observation data with less delay, determine different rapid update numerical model data assimilation time windows, repeat steps (4)-(6), perform data assimilation and numerical integration calculations on different types of data, and verify and evaluate the results of different rapid update numerical weather forecasts. (9) According to step (8), check the evaluation results, select the assimilation time window of the ultra-fast update numerical model data and the assimilated observation data, and carry out operational application in step (4); according to steps (2)-(7), build the operational script, determine the timed start time of the operational script, and establish the operational ultra-fast update numerical weather forecast system.

2. The ultra-fast numerical weather prediction method according to claim 1, characterized in that, In step (2), the global numerical weather prediction data product acquisition platform can be any one or more of the following: the meteorological data center Tianqing Big Data Cloud Platform, the meteorological numerical prediction cloud sharing platform, and the Internet; the initial field refers to the state of the atmosphere at the initial moment.

3. The ultra-fast numerical weather prediction method according to claim 1, characterized in that, The observation data processing in step (3) includes multi-step inspection of the observation data, quality control before assimilation, and quality labeling of each observation without changing the original value of the observation data; the multi-source meteorological observation data includes observation data from ground automatic stations, radiosondes, aircraft reports, GPS / MET water vapor, wind profiler radar, weather radar, and satellites.

4. The ultra-fast numerical weather prediction method according to claim 3, characterized in that, The multi-step inspection and quality control of the observation data before assimilation includes data extreme value inspection, inter-element consistency inspection, vertical consistency inspection of multi-layer observation data, background field inspection, horizontal consistency inspection, temporal consistency inspection, continuity inspection, sparsification, and bias correction.

5. The ultra-fast numerical weather prediction method according to claim 1, characterized in that, In step (4), the data assimilation analysis technique refers to the use of variational assimilation and hybrid assimilation methods to mathematically optimize the fusion of multi-source observation data of different times and properties with the background field of numerical models, so as to generate a new initial field A and lateral boundary A that are closer to the actual state of the atmosphere.

6. The ultra-fast numerical weather prediction method according to claim 1, characterized in that, The data assimilation analysis technology in step (4) also includes coupling technology and artificial intelligence; the data assimilation time window refers to the specific time interval in numerical weather prediction where the system collects and integrates observation data; the data assimilation analysis corrects the variables of the three-dimensional initial field, including horizontal wind field, dimensionless air pressure, potential temperature, and specific humidity.

7. The ultra-fast numerical weather prediction method according to claim 1, characterized in that, The future weather forecast products in step (5) include three-dimensional atmospheric temperature, air pressure, wind direction and speed, relative humidity, and precipitation. The integration calculation of the numerical model refers to the computer calculating the state of the entire three-dimensional atmospheric field after the first time step according to the numerical model equation and the initial field, and using it as the new initial condition. Then, it is substituted into the numerical model equation to calculate the state of the three-dimensional atmospheric field after the second time step, and used as the new initial condition. The initial conditions are repeatedly calculated and updated according to the set time step to predict the atmospheric state for a period of time in the future and obtain the future weather forecast.

Images