Local hundred-meter minute-level humiture rainfall real-time analysis system and construction method thereof

By designing a local 100-meter minute-level air temperature, humidity, and precipitation real-time analysis system, the existing system code optimization problems are solved and the difficulty of analyzing area transformation is achieved, dynamic area analysis and high-resolution output are achieved, and local live products are improved. Technical maturity and analysis accuracy.

CN119066355APending Publication Date: 2024-12-03STATE QIXIANG INFORMATION CENT
View PDF 0 Cites 3 Cited by

Patent Information

Application Number
CN202410989062.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-23
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

The existing local temperature, humidity, wind and precipitation real-time analysis system for 100 meters has problems such as insufficient code optimization and difficulty in changing the analysis area, which is difficult to meet the dynamic meteorological guarantee needs of major activities.

Method used

A local temperature, humidity and precipitation real-time analysis system of 100 meters in minute-level air temperature, humidity and precipitation real-time analysis subsystem and local 100 meters in precipitation real-time analysis subsystem were designed, using 'downscale + STMAS' and ' PDF+OI+downscale fusion analysis method is the system deployed in the 'Pai-Shuguang' high-performance computing cluster to realize dynamic area analysis and output a 100m/10min resolution live analysis product.

Benefits of technology

The dynamic changes in the analysis area and high-resolution output are realized, the technical maturity and refined analysis capabilities of local live products are improved, radar information is effectively introduced to improve the quality of analysis products, and the spatial structure and magnitude accuracy of precipitation are improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119066355A_ABST
    Figure CN119066355A_ABST
Patent Text Reader

Abstract

The invention discloses a local hundred-meter minute-level temperature, humidity, wind and rainfall live analysis system and a construction method thereof. The local hundred-meter minute-level temperature, humidity, wind and rainfall live analysis system comprises a local hundred-meter temperature, humidity and wind live analysis subsystem and a local hundred-meter rainfall live analysis subsystem, the construction method of the live analysis system comprises a construction method of a local hundred-meter temperature, humidity and wind live analysis subsystem and a construction method of a local hundred-meter rainfall live analysis subsystem. In the local hundred-meter temperature, humidity and wind live analysis subsystem, the STMAS technology is utilized to fuse national 1km and 1h live analysis products, and a machine learning method is adopted to establish a site error inversion model to carry out time-airborne downscale fusion live analysis technology research; the construction of the local 100m / 10min rainfall condition analysis subsystem is based on ground observation, and a PDF matching method deviation correction, OI fusion analysis and downscaling technology is applied to 10 min radar rainfall, so that a local 100m / 10min rainfall analysis technology is developed and formed.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of weather forecasting technology, and more specifically to a local 100-meter minute-level temperature, humidity, wind and precipitation real-time analysis system and a construction method thereof. Background Art

[0002] The China Meteorological Administration clearly proposed in the "14th Five-Year Plan for Meteorological Forecast Business Development" to develop local regional real-time services with a resolution of 100 meters and 5 minutes (Qifa

[2021] No. 103); the dynamic regional local 100-meter real-time analysis system is an important part of the construction of local regional real-time services, and its importance is constantly increasing.

[0003] According to the work arrangements of the China Meteorological Administration (China Meteorological Office

[2020] No. 41, Meteorological Office

[2020] No. 45), the National Meteorological Information Center built a real-time system for 100-meter live analysis of the 2022 Beijing Winter Olympics and a real-time system for 100-meter live analysis of the 2021 Shaanxi Universiade, generating a series of 100-meter live analysis products in real time, and distributing them to the Winter Olympics Meteorological Center and the Shaanxi Provincial Meteorological Department, providing accurate and stable live analysis product support for the meteorological support services for the Winter Olympics and the National Games; however, the relevant systems still have shortcomings such as insufficient code optimization and difficulty in changing analysis areas. Therefore, in April 2022, the China Meteorological Administration issued the task of "establishing a regional dynamic local 100-meter live analysis system to serve major events such as the Winter Olympics and the Asian Games" to the National Meteorological Information Center (Meteorological Office

[2022] No. 48). Summary of the invention

[0004] Therefore, the technical problem to be solved by the present invention is to provide a local 100-meter real-time analysis system and a construction method thereof.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] The local 100-meter minute-level temperature, humidity, wind and precipitation real-time analysis system includes the local 100-meter temperature, humidity and wind real-time analysis subsystem and the local 100-meter precipitation real-time analysis subsystem;

[0007] The local 100-meter temperature, humidity and wind real-time analysis subsystem includes a static parameter localization module, a ground minute temperature, humidity and wind observation data acquisition and preprocessing module, an ART-1km real-time analysis product preprocessing analysis module, a local 100-meter temperature, humidity and wind fusion analysis module, a local 100-meter temperature, humidity and wind analysis result evaluation module, a local 100-meter temperature, humidity and wind fusion analysis result visualization module and a local 100-meter temperature, humidity and wind fusion analysis result distribution module;

[0008] The local 100-meter precipitation real-time analysis subsystem includes a static parameter localization module, a ground minute precipitation observation data acquisition module, a ground minute precipitation observation preprocessing module, a kilometer-level weather radar QPE product preprocessing module, a kilometer-level ground minute precipitation OI difference module, a kilometer-level weather radar QPE product deviation correction module, a 100-meter-level X-band radar QPE product preprocessing module, a local 100-meter precipitation fusion analysis module, a local 100-meter precipitation analysis result evaluation module, a local 100-meter precipitation fusion analysis result visualization module and a local 100-meter precipitation fusion analysis result distribution module.

[0009] The above-mentioned local 100-meter minute-level temperature, humidity, wind and precipitation real-time analysis system is deployed in the "Pai-Shuguang" high-performance computing cluster, and the dynamic transformation of the analysis area is realized according to the parameter configuration;

[0010] The system input data include: ground minute temperature, humidity, wind speed, precipitation observation data observed by ground observation stations; kilometer-level weather radar QPE products; hundred-meter-level X-band radar QPE products; ART-1km real-time analysis products;

[0011] Fusion analysis method: “downscaling + STMAS” and “PDF + OI + downscaling” fusion analysis method;

[0012] System output products: 100-meter-level real-time analysis products with a resolution of 100m / 10min for precipitation, 2m for temperature, 2m for relative humidity, 10m for U wind speed, and 10m for V wind speed. The product format is netcdf.

[0013] The operation process is as follows: according to the needs, the ground minute temperature, humidity, wind speed observation data and temperature, humidity, wind speed and ART-1km real-time analysis products are configured into the local 100-meter temperature, humidity and wind real-time analysis subsystem, and the ground minute precipitation observation data, kilometer-level weather radar QPE products and 100-meter-level X-band radar QPE products are configured into the local 100-meter precipitation real-time analysis subsystem. Then, according to the configuration parameters, static parameter localization modules are made in the local 100-meter temperature, humidity and wind real-time analysis subsystem and the local 100-meter precipitation real-time analysis subsystem respectively.

[0014] In the local 100-meter temperature, humidity and wind real-time analysis subsystem: provide the ground minute temperature, humidity and wind speed observation data to the ground minute temperature, humidity and wind observation data acquisition and preprocessing module, configure the ART-1km real-time analysis product to the real-time analysis product preprocessing analysis module, and then configure the preprocessed data to the local 100-meter temperature, humidity and wind fusion analysis module to realize the dynamic regional local 100-meter fusion real-time analysis, use the local 100-meter temperature, humidity and wind analysis result evaluation module to evaluate, and use the local 100-meter temperature, humidity and wind fusion analysis result visualization module to visualize, output product pictures, and distribute them to application units through the local 100-meter temperature, humidity and wind fusion analysis result distribution module;

[0015] In the local 100-meter precipitation real-time analysis subsystem: the ground minute precipitation observation data is configured to the ground minute precipitation observation data acquisition module and the ground minute precipitation observation preprocessing module, the kilometer-level weather radar QPE products and the 100-meter-level X-band radar QPE products are configured to the kilometer-level weather radar QPE product preprocessing module and the 100-meter-level X-band radar QPE product preprocessing module respectively, the kilometer-level ground minute precipitation OI difference module, the kilometer-level weather radar QPE product deviation correction module and the local 100-meter precipitation fusion analysis module fuse the preprocessed data for real-time analysis, the local 100-meter precipitation analysis result evaluation module is used for evaluation, the local 100-meter precipitation fusion analysis result visualization module is used for visualization, the product pictures are output, and the local 100-meter precipitation fusion analysis result distribution module is used for distribution to the application units.

[0016] A method for constructing a local 100-meter minute-level temperature, humidity, wind, and precipitation real-time analysis system, including a method for constructing a local 100-meter temperature, humidity, wind real-time analysis subsystem and a method for constructing a local 100-meter precipitation real-time analysis subsystem;

[0017] The construction method of the local 100-meter temperature, humidity and wind real-time analysis subsystem includes:

[0018] (1) Acquisition and preprocessing of ground minute temperature, humidity and wind observation data;

[0019] (2) Pretreatment methods for real-time analysis products;

[0020] (3) Temperature, humidity and wind fusion analysis method based on machine learning;

[0021] (4) Evaluation method of local 100-meter temperature, humidity and wind analysis results;

[0022] (5) Complete the visualization of the local 100-meter temperature, humidity and wind fusion analysis results, and generate the visualization results of the local 100-meter temperature, humidity and wind fusion analysis results;

[0023] The local 100-meter precipitation real-time analysis subsystem method includes:

[0024] (A) Acquisition of ground-level minute precipitation observation data;

[0025] (B) Preprocessing of surface minute precipitation observations;

[0026] (C) Kilometer-level surface minute precipitation OI interpolation module;

[0027] (D) Kilometer-level weather radar QPE product preprocessing;

[0028] (E) Preprocessing of 100-meter-class X-band radar QPE products;

[0029] (F) Bias correction of kilometer-level weather radar QPE products;

[0030] (G) Fusion method of local 100-meter radar and ground-based precipitation observations;

[0031] (H) Evaluation method of local 100-meter precipitation analysis results.

[0032] In the method for constructing the local 100-meter minute-level temperature, humidity, wind and precipitation real-time analysis system, in step (1), the minute temperature, humidity and wind observation data observed on the ground are 100-meter / 10-minute temperature, humidity and wind observation data, which are derived from the minute temperature observation data, minute relative humidity observation data and minute wind speed and direction observation data of the automatic weather station; the preprocessing method is: through the coordinate conversion, variable conversion and quality control processing process, standard quality control observation data is generated;

[0033] In step (2), the real-time analysis products: the national 1km and 1h real-time analysis products released by the National Meteorological Information Center and the digital elevation model products resampled to 100m resolution based on satellite observations are used;

[0034] The preprocessing method using spatiotemporal multi-scale analysis technology and height correction algorithm is as follows: the spatiotemporal resolution interpolation of the national 1km and 1h real-time analysis products is performed to obtain background field data, and the digital elevation model data with sub-100-meter resolution is used to fuse the background field and observation data using a multi-grid variational algorithm, thereby correcting the background field height, producing background field data with a 100-meter resolution at the entire point, and obtaining 100m / 1h real-time fusion data.

[0035] The method for constructing the local 100-meter minute-level temperature, humidity, wind and precipitation real-time analysis system, in step (3), uses the minute-level temperature, relative humidity, wind speed and wind direction observation data of the ground automatic meteorological observation station, 100m / 1h real-time fusion data and slope and aspect terrain information as the main data sources, adopts machine learning methods to establish a station error inversion model, and conducts research on the fusion real-time analysis technology of temperature, relative humidity and wind speed and wind direction on the scale of 100 meters / 10 minutes;

[0036] Machine learning methods include gradient boosting decision tree GBDT, random forest RF, Adaboost, K nearest neighbor classification algorithm KNN, support vector machine SVM, ER-FR model,

[0037] Model correction for machine learning methods is divided into two parts:

[0038] The first part is the training phase: using the preprocessed historical data as training samples, we conduct feature selection, model parameter tuning, and performance optimization training for different machine learning models to obtain the optimal training model.

[0039] The second part is the testing phase: using the optimal training model, feature extraction and parameter optimization are performed to obtain the optimal error inversion model, and the background field correction of temperature and humidity is carried out. The data results after correction of each model are evaluated, the performance characteristics of different models are analyzed, the correction weights of each model are calculated, and the correction results of different models are integrated according to the weights;

[0040] In step (4), temperature, humidity and wind are evaluated using correlation coefficient, mean error and root mean square error.

[0041] In the method for constructing the local 100-meter minute-level temperature, humidity, wind and precipitation real-time analysis system, in the local 100-meter precipitation real-time analysis subsystem method, in step (A): the ground minute precipitation observation data adopts the minute-by-minute precipitation data observed by the automatic weather station of the China Meteorological Administration obtained from the big data cloud platform, AWS, including the minute-by-minute precipitation data set at the encrypted observation automatic station;

[0042] In step (B), the preprocessing includes: according to the station ID numbers collected in the analysis area, at 00 / 10 / 20 / 30 / 40 / 50 minutes of each hour, statistically calculating the accumulated precipitation of the same station, that is, the same station ID number, in the past 10 minutes as the 10-minute observed precipitation at 00 / 10 / 20 / 30 / 40 / 50 minutes of the station; selecting only data with quality control codes of 0 / 1 / 3 / 4 / 9 to participate in the statistical calculation; before entering the analysis, the accumulated 10-minute precipitation data is also subjected to preliminary quality control, by checking the boundary value of precipitation elements to remove gross errors, and eliminating stations with long-term observations of "zero value rigid values";

[0043] In step (C), the precipitation gridding is performed on the minute-level precipitation observation data of the ground observation stations, and the optimal interpolation OI gridding analysis technology of the climate background field ratio is adopted, including real-time updated analysis station table and distance index file of analysis grid points and stations, analysis station selection strategy adaptive to station network density to automatically optimize the analysis radius, regional limitation of the maximum analysis radius to avoid analyzing the "big circle" precipitation structure due to too sparse stations, and increase the processing of the situation where the satellite background field in a large area in the north in winter is 0, so as to better retain the structure of the ground observation precipitation grid analysis.

[0044] In the method for constructing the local 100-meter minute-level temperature, humidity, wind and precipitation real-time analysis system, in step (C), the kilometer-level weather radar QPE product data is a 10-minute fast radar QPE product developed by the National Meteorological Information Center for the real-time and high-efficiency application needs of precipitation; the kilometer-level weather radar QPE product is generated by inverting the radar base data or volume scan data of the lower four layers obtained from the NAS disk of the big data cloud platform, and the specific preprocessing method is: first, after eliminating isolated points, radial interference, ground interference, and wave echo quality control, the mixed scanning rate is calculated, and the precipitation rate estimate is calculated according to the ZI relationship, converted into the precipitation in the past 10 minutes, and a single-station precipitation product is generated every 3 minutes; secondly, the single-station precipitation product is gridded in longitude and latitude, and is started regularly at intervals of 10 minutes, and the data of each station in the previous 10 minutes and the most recent time are used to perform the maximum value networking puzzle to generate a 10-minute precipitation product for the Chinese region;

[0045] In step (D), the 100-meter X-band radar QPE product is based on 11 X-band radar base data collected in Beijing and Hebei Province. It adopts the same preprocessing method as the kilometer-level weather radar QPE product. When the 100-meter X-band radar QPE product can detect the precipitation system, it is used to provide high-resolution precipitation structure information to achieve downscaling from kilometer level to 100-meter level.

[0046] In the method for constructing the local 100-meter minute-level temperature, humidity, wind and precipitation real-time analysis system, in step (F), the PDF matching method is used to correct the deviation of the radar QPE;

[0047] The principle of PDF matching method is: assuming that the probability density distribution of precipitation is relatively stable within a certain time and space range, the value of radar precipitation under the same cumulative probability is adjusted based on ground observations, so that the probability density distribution of radar precipitation is consistent with ground observations, so as to achieve the purpose of correcting the systematic deviation of radar precipitation;

[0048] The specific method is:

[0049] First, the consistency between radar QPE and surface-observed precipitation on the 10-min time scale is examined;

[0050] Secondly, the spatiotemporal matching window of the PDF sample is adjusted by comprehensively considering the characteristics of the occurrence and disappearance time of 10-minute precipitation, the detection range of a single radar, the stability of radar bias changes, and the changes in station network density in winter and summer.

[0051] In the method for constructing the local 100-meter minute-level temperature, humidity, wind and precipitation real-time analysis system, in step (G),

[0052] Firstly, the radar QPE precipitation after bias correction is used as the initial estimation field, i.e., the background field, and the surface minute precipitation observation data is fused using the OI method to obtain the meltwater precipitation field.

[0053] Specifically, as shown in formula (1):

[0054]

[0055] That is, the analysis value A of the grid k is the initial estimate F of the point k Add the deviation between the observed value of the point and the initial estimate, and the deviation is determined by n known initial estimates F within a certain analysis range. i With the observed value O i The bias-weighted estimate of is obtained;

[0056] OI method to solve the weight W i The linear equation of is shown in formula (2):

[0057]

[0058] in, Indicates the initial estimated field error correlation between point i and point j, represents the observation error co-correlation, λ i is the standard deviation of the observation error at point i and the initial estimate error standard deviation The ratio, λ j is the standard deviation of the observation error at point j and the initial estimate error standard deviation ratio; It indicates the co-correlation of the initial estimated field errors between the k-point and the j-point of the analysis grid;

[0059] The error ratio between the adjusted observation and the initial estimate is used The method is used to determine the OI fusion parameters to solve the weights, When it is set to 0.4, the fusion effect is the best;

[0060] Secondly, taking the kilometer-level / 10-minute fused precipitation field as the background, the 100m radar precipitation information is estimated using the OPE product data based on the 100-meter X-band radar, thereby downscaling the kilometer-level / 10-minute fused data;

[0061] The specific downscaling formula is as follows:

[0062]

[0063] Among them, P 100m,mg is the 100 m fused precipitation after downscaling, P 1km,mg is the radar-ground fusion analysis field of 1 km, P 1km,Rad and P 100m,Rad They are kilometer-level weather radar QPE products and hundred-meter-level X-band radar QPE products.

[0064] In the method for constructing the local 100-meter minute-level temperature, humidity, wind and precipitation real-time analysis system, in step (H), the fusion effect evaluation adopts an independent inspection method, selects some stations with higher observation quality as inspection stations, retains their observation values ​​and does not participate in precipitation analysis, and takes them as "true values". The various grid precipitation products to be inspected are interpolated to the inspection stations through the neighboring interpolation, and the differences are compared with the "true values";

[0065] The evaluation indicators include the correlation coefficient Coefficient Correlation, CC, the root mean square error Root Mean Square Error, RMSE, and the relative bias Relative Bias, Rbias statistical error. The specific calculation formula is as follows:

[0066]

[0067] Among them, g i is the precipitation observed at the inspection station, i.e. the “true value” of precipitation; p i is the precipitation of each precipitation product interpolated to the inspection station; n is the number of statistical matching samples, The average value of precipitation interpolated to the inspection station for each inspected precipitation product; is the average precipitation observed at the inspection station.

[0068] The technical solution of the present invention achieves the following beneficial technical effects:

[0069] 1. According to the requirements of the China Meteorological Administration for meteorological support deployment for major events, the National Meteorological Information Center has built a "dynamic regional local 100-meter minute-level temperature, humidity, wind and precipitation real-time analysis system" based on the real-time 100-meter real-time analysis system for the Beijing Winter Olympics, Chengdu Universiade, Hangzhou Universiade, etc. It can analyze regional dynamic changes and output 5 types of 100-meter real-time analysis products such as 100m / 10min resolution precipitation, 2m temperature, 2m relative humidity, 10m U wind speed, and 10mV wind speed. The system is currently deployed on the "Pai-Shuguang" high-performance computing cluster.

[0070] 2. Perform spatiotemporal resolution interpolation on the national 1km and 1h real-time analysis products to obtain background field data, and use the sub-100-meter resolution digital elevation model data to fuse the background field and observation data using a multi-grid variational algorithm to achieve deviation correction of the background field, produce background field data at 100-meter resolution, and obtain 100m / 1h real-time fusion data; use the minute-level temperature, relative humidity, wind speed and direction observation data of the ground automatic meteorological observation station, 100m / 1h real-time fusion data, and slope and aspect terrain information as the main data sources, use machine learning methods to establish a station error inversion model, and conduct research on the refined spatiotemporal downscaling fusion real-time analysis technology of temperature, relative humidity, and wind speed and direction on the scale of 100 meters / 10 minutes. Improve the maturity of technology and provide technical support for the development of refined local real-time products.

[0071] 3. Regarding the error estimation of the initial estimate and the observed field, in previous applications, statistical methods were used to fit the error estimation equation. As the temporal and spatial resolution of the analysis continues to increase, it is difficult to obtain sufficiently fine observations to statistically calculate a stable and accurate error change curve. This scheme uses the error ratio of the observed and initial estimate fields to adjust the error ratio of the observed and initial estimate fields. The OI fusion parameters are determined by the method to solve the weight. Generally speaking, the error of radar data is greater than that of ground observation. After testing, when the error ratio between observation and initial estimation is set to 0.4, the fusion effect is relatively good.

[0072] 4. Based on ground observations, for 10-minute radar precipitation, PDF matching method deviation correction, OI fusion analysis and downscaling technology are applied to develop a precipitation analysis technology suitable for local 100m / 10 minutes.

[0073] 5. From the results of independent statistical tests in summer, the spatial structure of precipitation is more refined after the introduction of radar information fusion analysis, and the overall high radar precipitation has been corrected to a certain extent. The value is closer to the ground observation. Overall, the correlation coefficient and root mean square error index are significantly improved compared with the single-source ground analysis products and radar products, indicating that the fusion analysis scheme can effectively introduce radar information to improve the quality of analysis products. However, the radar's ability to detect snowfall in the focus area in winter decreases, the accuracy of radar QPE products is greatly reduced, and the performance is unstable, which also has a significant impact on the accuracy of fused precipitation. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 The organizational structure of the "Dynamic Regional Local Hundred-meter Minute Level Temperature, Humidity, Wind and Precipitation Real-time Analysis System"

[0075] Figure 2 This is a schematic diagram of the operation process of the "Dynamic Regional Local Hundred-meter Minute Level Temperature, Humidity, Wind and Precipitation Real-time Analysis System";

[0076] Figure 3 This is a statistical map of ground observation sites in the Winter Olympics mountain competition area;

[0077] Figure 4a This is the spatial distribution map of the background field of the Winter Olympics mountain competition area - temperature; Figure 4b This is the spatial distribution map of the background field of the Winter Olympics mountain competition area - humidity; Figure 4c This is the spatial distribution map of the background field of the Winter Olympics mountain competition area - U wind; Figure 4d This is the spatial distribution map of the background field of the Winter Olympics mountain competition area - V wind;

[0078] Figure 5 This is the spatial distribution map of the 100m resolution digital elevation model of the Winter Olympics mountain competition area;

[0079] Figure 6 It is a real-time product technical solution for temperature and humidity at 100 meters / 10 minutes;

[0080] Figure 7 It is a technical route for downscaling temperature and humidity at 100m / 10min;

[0081] Figure 8 This is a flow chart of the model training and testing scheme;

[0082] Figure 9a This is a comparison chart of the root mean square error before and after downscaling correction of the GBDT model; Figure 9b This is a comparison chart of the root mean square error before and after downscaling correction of the RF model; Fig.9c This is a comparison chart of the root mean square error before and after downscaling correction of the Adaboost model; (the red line is before correction, and the blue line is after correction);

[0083] Fig.10a is the evaluation result of the temperature correlation coefficient, Fig.10b is the average temperature error evaluation result, Fig.10c is the RMS error evaluation result of the temperature; Fig.10d is the relative humidity correlation coefficient evaluation result, Fig.10e is the average error evaluation result of relative humidity; Fig.10f is the relative humidity root mean square error evaluation result;

[0084] Fig.11 It is the temperature spatial resolution map with different resolutions of 2m;

[0085] Fig.12a is the spatial distribution diagram of the U wind root mean square error; Figure 12b is the spatial distribution diagram of V wind root mean square error;

[0086] Fig.13a is the spatial distribution diagram of U wind correlation coefficient; Fig.13b is the spatial distribution map of V wind correlation coefficient;

[0087] Fig.14 The relationship between the correlation coefficient of the precipitation time series observed at neighboring stations on a 10-minute scale and the station distance for different cases

[0088] Fig.15a Example of spatial distribution of 1km / 10-minute ground automatic station observations in the Winter Olympics area. Fig.15b This is an example of spatial distribution of precipitation (unit: mm / 10min) for grid analysis; (07:00 UTC, April 30, 2021);

[0089] Fig.16 X-band radar coverage for Beijing and surrounding areas;

[0090] Fig.17 This is a schematic diagram of the 100-meter / minute precipitation fusion technology solution;

[0091] Fig.18a An example of the spatial distribution of precipitation analyzed from ground observations. Fig.18b This is an example of the spatial distribution of precipitation from radar QPE products; Fig.18c This is an example of spatial distribution of precipitation after PDF correction; Fig.18d This is an example of the spatial distribution of precipitation differences (unit: mm / 10min) after PDF bias correction (07:00 UTC, April 30, 2021);

[0092] Fig.19 It is the deviation time series (00:00 on July 1, 2021 to 23:00 on July 31, 2021) of the hourly 10-minute average precipitation rate (gauge, gray column) and radar QPE deviation before and after correction (NQPE and NQPE_ctr, red and green curves) in the Beijing-Tianjin-Hebei region;

[0093] Fig. 20 It is the deviation time series of the hourly 10-minute average precipitation rate (gauge, gray column) and radar QPE deviation before and after correction (NQPE and NQPE_ctr, red and green curves) in the Beijing-Tianjin-Hebei region (00:00 on February 27, 2021-23:00 on February 28);

[0094] Fig.21 This is an example of the spatial distribution of ground and radar fused precipitation (unit: mm / 10min) (07:00 UTC on April 30, 2021);

[0095] Fig. 22The spatial distribution of 10-minute precipitation of different products in the Winter Olympics area (10:10 UTC on June 30, 2021, unit: mm / 10min): a is ground station observation (Gauge), b is 100m ground analysis (GGA_100m), c is 1km radar QPE (Radar QPE), d is 1km radar-ground fusion (Merged), e is 100m radar QPE (X-band QPE) and f is 100m downscale fusion analysis (Downscale)

[0096] Fig.23 The 1-hour precipitation distribution in the Winter Olympics area at 11:00 UTC on June 30, 2021 (unit: mm): a is Gauge, ground observation; b is CMPAS_NRT, ART_1km near-real-time precipitation product of the National Meteorological Information Center; c is DS_100M, downscaling analysis product;

[0097] Fig.24a The spatial distribution of the inspection stations (red dots) and analysis stations (blue dots) in the Winter Olympics area during summer (right); Figure 24b Spatial distribution of inspection stations (red dots) and analysis stations (blue dots) in the Winter Olympics area during winter (left);

[0098] Fig.25 It is the time series of precipitation every 10 minutes observed by three ground inspection stations;

[0099] Fig.26a is the relative deviation of various 100 m / 10 min precipitation products under different precipitation intensities in July 2021; Figure 26b is the root mean square error of various 100 m / 10 min precipitation products under different precipitation intensities in July 2021;

[0100] Fig. 27 This is a scatter plot of the precipitation observed at the inspection stations in the core competition areas of the Winter Olympics and the precipitation values ​​analyzed by different products. DETAILED DESCRIPTION

[0101] Example 1: Local 100-meter minute-level temperature, humidity, wind and precipitation real-time analysis system

[0102] 1. System Overview

[0103] The "Dynamic Regional Local Hundred-meter Minute-level Temperature, Humidity, Wind and Precipitation Real-time Analysis System" is deployed in the "Pai-Shuguang" high-performance computing cluster, and the dynamic transformation of the analysis area is realized according to the parameter configuration; the system input data includes ground minute temperature, humidity, wind speed, precipitation observation data, kilometer-level and hundred-meter-level radar QPE products, and ART-1km real-time analysis products. Using the "downscaling + STMAS" and "PDF+OI+downscaling" fusion analysis methods, the system output products are 5 types of hundred-meter-level real-time analysis products such as 100m / 10min resolution precipitation, 2m temperature, 2m relative humidity, 10m U wind speed, 10m V wind speed, and the product format is netcdf.

[0104] Table 1 System overview

[0105]

[0106]

[0107] 2. System Organization Structure

[0108] The “Dynamic Regional Local 100-meter Minute-level Temperature, Humidity, Wind and Precipitation Real-time Analysis System” includes: (1) Dynamic Regional Local 100-meter Temperature, Humidity and Wind Real-time Analysis Subsystem; (2) Dynamic Regional Local 100-meter Precipitation Real-time Analysis Subsystem. The specific organizational structure of the system is as follows: Figure 1 .

[0109] The local 100-meter temperature, humidity and wind real-time analysis subsystem includes a static parameter localization module, a ground minute temperature, humidity and wind observation data acquisition and preprocessing module, a real-time analysis product preprocessing analysis module, a local 100-meter temperature, humidity and wind fusion analysis module, a local 100-meter temperature, humidity and wind analysis result evaluation module, a local 100-meter temperature, humidity and wind fusion analysis result visualization module and a local 100-meter temperature, humidity and wind fusion analysis result distribution module;

[0110] The local 100-meter precipitation real-time analysis subsystem includes a static parameter localization module, a ground minute precipitation observation data acquisition module, a ground minute precipitation observation preprocessing module, a kilometer-level weather radar QPE product preprocessing module, a kilometer-level ground minute precipitation OI difference module, a kilometer-level weather radar QPE product deviation correction module, a 100-meter X-band radar QPE product preprocessing module, a local 100-meter precipitation fusion analysis module, a local 100-meter precipitation analysis result evaluation module, a local 100-meter precipitation fusion analysis result visualization module and a local 100-meter temperature, humidity and wind fusion analysis result distribution module.

[0111] 3. System operation process

[0112] The "Dynamic Regional Local 100-meter Minute-level Temperature, Humidity, Wind and Precipitation Real-time Analysis System" first configures the analysis area and other parameters according to the needs, and then produces static data of the analysis area according to the configuration parameters, and provides it to the relevant pre-processing and fusion analysis function modules to realize the dynamic regional local 100-meter fusion real-time analysis. The fusion results are distributed to the application units through CTS, and visualized and output product pictures at the same time. Figure 2 .

[0113] 4. Test environment

[0114] The dynamic regional local 100-meter minute-level temperature, humidity, wind and precipitation real-time analysis system V1.0 is deployed on the Pai-Shuguang high-performance computer and uses the Linux operating system.

[0115] Table 2 Dynamic regional local 100-meter minute-level temperature, humidity, wind and precipitation real-time analysis system V1.0 test environment

[0116]

[0117] Embodiment 2: Method for constructing a local 100-meter minute-level temperature, humidity, wind and precipitation real-time analysis system

[0118] It includes a construction method for a local 100-meter temperature, humidity and wind real-time analysis subsystem and a construction method for a local 100-meter precipitation real-time analysis subsystem.

[0119] 1. Construction method of local 100-meter temperature, humidity and wind real-time analysis subsystem.

[0120] (1) Acquisition and preprocessing of ground minute temperature, humidity and wind observation data;

[0121] The data sources of the real-time fusion product of temperature, humidity and wind speed at 100 meters / 10 minutes include: Ground station observation data:

[0122] Minute-by-minute temperature observation data from about 150 automatic weather stations in the Winter Olympics mountain competition area ( Figure 3 ), minute-by-minute relative humidity observation data from about 100 automatic weather stations in the Winter Olympics mountain competition area, and minute-by-minute wind speed and direction observation data from about 125 automatic weather stations in the Winter Olympics mountain competition area.

[0123] Preprocessing method: Generate standard quality control observation data through coordinate transformation, variable transformation and quality control processing.

[0124] (2) Pretreatment methods for ART-1km real-time analysis products;

[0125] Integrated background field data: Winter Olympics mountain competition area temperature ( Figure 4a ), relative humidity( Figure 4b )、UV wind( Figure 4c and Figure 4dThe background field of the fusion product is produced by using the national 1km, 1h real-time analysis product (ART-1km) released by the National Meteorological Information Center after time-space interpolation and matching.

[0126] Digital elevation model products based on satellite observations resampled to 100-meter resolution ( Figure 5 ).

[0127] The preprocessing method using spatiotemporal multi-scale analysis technology and height correction algorithm is as follows: temperature, humidity, wind speed, and the basic technical idea adopted in the research and development is to perform spatiotemporal resolution interpolation on the national 1km and 1h real-time analysis products at 24 hours a day to obtain background field data, and use the digital elevation model data with a sub-100-meter resolution to fuse the background field and the observation data using a multi-grid variational algorithm, so as to perform height correction on the background field, produce background field data with a hundred-meter resolution hourly point, and obtain 100m / 1h real-time fusion data, such as Figure 6 shown.

[0128] Taking 00 o'clock as an example, the data development process of 00:00, 10 minutes, 20 minutes, 30 minutes, 40 minutes and 50 minutes is introduced: at 00:10, the data after spatiotemporal resampling and terrain correction of the hourly product at 00:00 is used as the background field, and then the real-time data of 00:10 in the Winter Olympics mountain competition area, including national business observation station data, regional automatic station observation data, non-national assessment station automatic observation data, and Winter Olympics encrypted observation station automatic observation data, are retrieved from the big data cloud platform. Finally, the multi-grid variational algorithm, which is more mature in the current live business, is used to correct the background field. The scene and observation data are fused to generate a live product for 00:10 minutes; at 00:20, the data at 00:10 is used as the background scene, and then the real-time data of 00:20 in the Winter Olympics mountain competition area, including national business observation station data, regional automatic station observation data, non-national assessment station automatic observation data, and Winter Olympics encrypted observation station automatic observation data, are retrieved from the big data cloud platform, and then the background field and observation data are fused using the multi-grid variational algorithm to generate a live product for 00:20 minutes; taking this as an example, the products for 00:30, 40, and 50 minutes are generated in a loop.

[0129] (3) Temperature, humidity and wind fusion analysis method based on machine learning;

[0130] Taking the minute-level observation data of temperature, relative humidity, wind speed and direction from the ground automatic meteorological observation station, the 100m / 1h real-time fusion data and the slope and aspect terrain information as the main data sources, the machine learning method is used to establish the station error inversion model, and the fusion real-time analysis technology of temperature, relative humidity and wind speed and direction on the scale of 100m / 10min is studied; Figure 7 shown.

[0131] 1. Data preparation for machine learning methods:

[0132] Observation data of ground automatic stations in the Olympic region. According to the Technical Plan for the Smart Winter Olympics 2022 Weather Forecast Demonstration Plan, in the Beijing-Tianjin-Hebei region covering the Winter Olympics, there are 4,210 stations with basic meteorological element observation data, 179 stations with ground meteorological standard value data, and 25 stations with Winter Olympics automatic station observation data, covering 113.2°E-120.2°E, 35.9°N-42.7°N. Currently, this data has been obtained in real time from the National Meteorological Information Center.

[0133] Table 3 Information of ground observation stations in the Winter Olympics area

[0134]

[0135] High-resolution digital elevation model data. ASTER GDEM data is jointly developed by Japan's METI and the United States NASA and distributed to the public free of charge. On August 5, 2019, NASA and METI jointly released the ASTER GDEM V3 version. This study uses the ASTER GDEM V3 version of DEM data to maximize the preparation of regional terrain data.

[0136] The high-resolution land cover type data uses the land cover type dataset developed by Gong Peng and others. This product covers multi-season samples uniformly covering the world from Landsat 8 images in 2014 and 2015, interpreted by experts. Among them, the training set contains about 340,000 samples of different sizes (from 30m×30m to 500m×500m), covering about 93,000 sample points around the world; the validation set contains about 140,000 samples from different seasons, covering more than 38,000 sample points. The sample library was applied to the Sentinel-2 images acquired in 2017, and a global 10-meter land cover map was obtained based on the random forest classifier.

[0137] 1km / 1h real-time analysis product of temperature and relative humidity. When the 10-minute real-time fusion analysis product was first developed, there was a lack of 10-minute / 100m background field data of temperature and relative humidity. Therefore, the 1km / 1h real-time fusion product developed by the National Meteorological Information Center was used as the initial background field data, resampled to 100m spatial resolution, and then calculated from the hourly resolution data according to the time weight to obtain data every 10 minutes. This study produced a total of 1008 samples for 7 days as the initial minute data, and matched them with the corresponding time-time ground automatic station observation data collected.

[0138] 2. Training and testing plan

[0139] Model correction of machine learning methods is divided into two parts, such as Figure 8 shown.

[0140] The first part is the training phase: using the preprocessed historical data as training samples, we conduct feature selection, model parameter tuning, and performance optimization training for different machine learning models to obtain the optimal training model.

[0141] The second part is the testing phase: using the optimal training model, feature extraction and parameter optimization are performed to obtain the optimal error inversion model, and the background field correction of temperature and humidity is carried out. The data results after correction of each model are evaluated, the performance characteristics of different models are analyzed, the correction weights of each model are calculated, and the correction results of different models are integrated according to the weights;

[0142] Using Gradient Boosted Decision Tree (GBDT) ( Figure 9a ), Random Forest (RF) ( Figure 9b )、Adaboost( Fig.9c ) model, training and testing were carried out. The training time was from 00:00 on June 25, 2020 to 23:00 on June 30, 2020, and the prediction time was from 00:00 on July 1, 2020 to 23:00 on July 1, 2020. The specific results are as follows Figure 9a-9c , the red line is before downscaling correction, and the blue line is after downscaling correction.

[0143] (4) Evaluation method of local 100-meter temperature, humidity and wind analysis results;

[0144] In general, the correlation coefficient of temperature assessment at the sub-100-meter level (100m / 10min) is above 0.99, the average error is around 0.2℃, the root mean square error is mainly 0.3-0.5℃, and sometimes 0.5-0.8℃. The correlation coefficient of relative humidity is around 0.98, the average error is around 1%, and the root mean square error is mainly 1%-3%. Figure 10a-10f .

[0145] Taking temperature as an example, the spatial distribution of real-time products with 5km resolution, 1km resolution, and 100m resolution at 00:00 (UTC) on September 16, 2020 is plotted, as shown in the figure below: Fig.11 As shown in the figure, by comparison, we can find that in the mountainous competition area of ​​the Winter Olympics, live products of different resolutions can better grasp the distribution of hot and cold temperatures. The temperature in the central and western regions is relatively low, mainly below 10°C, while there is a local 10-12°C variation area in the central and northern regions, and the temperature in the southern region is higher than other regions. In the comparison of different resolutions, as the resolution increases, the spatial details are more obvious, and the extreme values ​​that are smoothed in the case of coarse resolution are gradually refined as the resolution increases.

[0146] A preliminary quality assessment was conducted on the 10min / 100m U wind and V wind real-time fusion products produced in the Winter Olympics area (the assessment period was from August 27 to September 1, 2021), and the root mean square error and correlation coefficient were calculated, as shown in the figure below. The root mean square error of most stations is below 0.4, and the error of some encrypted stations is between 0.8-1.2m / s. The correlation coefficient of most stations is above 0.8, and the correlation coefficient of some encrypted stations is between 0.6-0.8. Fig.12a , Figure 12b , Fig.13a , Fig.13b shown.

[0147] (5) Complete the visualization of the local 100-meter temperature, humidity and wind fusion analysis results, and generate the visualization of the local 100-meter temperature, humidity and wind fusion analysis results.

[0148] 2. Construction method of local 100-meter precipitation real-time analysis subsystem

[0149] (A) Acquisition of ground-level minute precipitation observation data;

[0150] The ground observation data used in this technical solution is the minute-by-minute precipitation data observed by the China Meteorological Administration's automatic weather station (AWS) obtained from the big data cloud platform, including the minute-by-minute precipitation data of encrypted automatic observation stations set up in the Winter Olympics competition area.

[0151] (B) Preprocessing of surface minute precipitation observations;

[0152] Preprocessing method:

[0153] According to the station ID numbers collected in the analysis area, the accumulated precipitation of the same station (i.e., the same station ID number) in the past 10 minutes is statistically calculated at 00 / 10 / 20 / 30 / 40 / 50 minutes every hour as the 10-minute observed precipitation of the station at 00 / 10 / 20 / 30 / 40 / 50 minutes. Only data with quality control codes of 0 / 1 / 3 / 4 / 9 (i.e., correct, suspicious, corrected, modified, and uncontrolled observation data, which is the same as the selection strategy of hourly ground precipitation observation data in the ART_1km business system) are selected for statistical calculation. When statistics are performed, the precipitation accumulated by 10 complete minute-by-minute observations in every 10 minutes is involved in the actual precipitation fusion analysis and verification. Before entering the analysis, the accumulated 10-minute precipitation data is also subjected to preliminary quality control, mainly by checking the boundary value of precipitation elements to remove gross errors. In addition, analysis of the minute data revealed that the observation data of some stations had long-term values ​​of zero. The minute precipitation data of all automatic stations in the Beijing-Tianjin-Hebei region over the past year were checked, and stations with long-term observations of "zero values" were blacklisted and excluded from fusion analysis and verification.

[0154] (C) Kilometer-level surface minute precipitation OI interpolation module;

[0155] The ground-based precipitation gridding adopts the optimal interpolation (OI) gridding analysis technology of the climate background field ratio. This method can indirectly consider the climate effect of terrain on precipitation increase by introducing the climate background field information corrected by terrain, and is more accurate than the general OI method.

[0156] The National Meteorological Information Center has focused on upgrading the modules of the above OI analysis technology. For example, it has split the different functions such as data preprocessing, analysis station search strategy, OI analysis framework, error estimation, linear equation solving, and post-analysis processing into multiple independent modules to form OI analysis software, which is convenient for selection and subsequent technical upgrades during research and development. The ground analysis technology of this solution is based on the modified OI analysis software, focusing on upgrading the OI analysis algorithm for precipitation problems in winter live products, including:

[0157] 1. Real-time updated analysis station table and analysis grid and station distance index file can introduce more station observation information into fusion analysis in real time;

[0158] 2. An adaptive analysis station selection strategy based on the density of the station network. The analysis radius is automatically optimized based on the situation that a large number of tipping bucket rain gauges in northern China stop measuring in winter, causing the station network density to become sparse.

[0159] 3. The maximum analysis radius is limited by region (Tibet, Qinghai, and Xinjiang) to avoid analyzing a "big circle" precipitation structure due to too sparse stations;

[0160] 4. In order to address the problem that satellites have a weak ability to reflect solid precipitation in winter, we have added processing for the situation where the satellite background field in a large area in the north in winter is 0, and better retain the structure of the ground observation precipitation grid analysis.

[0161] The optimized OI analysis algorithm and module were applied to the optimization and upgrade of the operational precipitation subsystem of the China regional multi-source fusion real-time analysis 1km resolution product (ART_1km, ground) at the end of August 2021.

[0162] When applying OI analysis at the 1km / 10min scale, parameters such as analysis radius and impact attenuation distance were adjusted according to the characteristics of minute precipitation, such as Fig.14 It shows that there are obvious differences in the spatial correlation of precipitation in three different precipitation processes in the Winter Olympics area in June 2021, but in general, the correlation coefficient of precipitation at neighboring stations within 50km gradually decreases to 0.2 with increasing distance. If the negative exponential attenuation relationship of e is used in the analysis, the attenuation distance can be set at about 20-40km.

[0163] On April 30, 2021, the Zhangjiakou Winter Olympics area experienced a significant snowfall. Fig.15a and Fig.15b An example of 1km / 10-minute precipitation based on the analysis of automatic station observations in the Winter Olympics competition area at 07:00 UTC on April 30, 2021 is given. The spatial distribution of precipitation obtained from the analysis is consistent with the corresponding values ​​observed by the ground automatic station, and the analysis structure is reasonable.

[0164] (D) Kilometer-level weather radar QPE product preprocessing;

[0165] The radar data uses the 10-minute fast radar QPE product developed by the National Meteorological Information Center for real-time and high-efficiency precipitation applications, including two spatial resolutions of 1km and 100m. The 1km radar QPE product is generated by inverting the radar base data or volume scan data of the lower four layers obtained from the NAS disk of the big data cloud platform.

[0166] The specific analysis algorithm is: first, after quality control such as isolated point elimination, radial interference, ground interference elimination, and wave echo, the mixed scanning rate is calculated, and the precipitation rate is estimated according to the ZI relationship, converted into the precipitation in the past 10 minutes, and a single-station precipitation product is generated every 3 minutes; secondly, the single-station precipitation product is gridded in longitude and latitude, and is started regularly at 10-minute intervals, and the data of each station in the previous 10 minutes and the most recent time are used to perform the maximum value network puzzle to generate a 10-minute precipitation product for the Chinese region. This product has a high timeliness and can be generated within 2 minutes of the punctual lag.

[0167] (E) Preprocessing of 100-meter-class X-band radar QPE products;

[0168] The 100m radar QPE product is mainly based on the basic data of 11 X-band radars collected in Beijing and Hebei (covering the area of Fig.16 The green box in the figure indicates the scope of the Beijing Winter Olympics competition area), which is generated by QPE inversion using the above-mentioned 1km similar algorithm. This product will be put into real-time operation from April 2021. Before applying the X-band 100-meter information downscaling, a preliminary check is made on the consistency of the 1km and 100m radar QPE products.

[0169] At present, the application of X-band radar QPE still adopts the idea of ​​downscaling by using the high-resolution precipitation structure information it provides when it can detect precipitation systems. This will be applied in step (G).

[0170] Technical ideas of step (F) and step (G):

[0171] The local 100m / 10-minute scale precipitation fusion real-time analysis technology uses the ground automatic station minute-level precipitation observation data and high-resolution (1km and 100m) radar QPE as the main data sources, and generates 100m / minute-level precipitation real-time analysis products in two stages in combination with the application timeliness requirements (such as Fig.17 shown):

[0172] In the first stage, to meet the needs of rapid application, the OI method is used to generate 100m / 10min rapid ground precipitation grid analysis products based on minute-level precipitation observation data from ground automatic stations (see step C);

[0173] In the second stage, radar detection data was introduced, and QPE products with two resolutions of 1km and 100m were generated based on the inversion of radar base data covering the Winter Olympics area. At the 1km scale, the systematic deviation of radar precipitation was corrected and fused with ground observations based on ground observations. Then, the 100m radar QPE data was used to spatially downscale the 1km fused product to generate a better quality 100m / 10minute ground-radar fusion precipitation analysis product.

[0174] (F) Bias correction of kilometer-level weather radar QPE products;

[0175] The PDF matching method is used to correct the deviation of radar QPE. The principle of the PDF matching method is: assuming that the probability density distribution of precipitation is relatively stable within a certain time and space range, the value of radar precipitation under the same cumulative probability (Cumulative Distribution Function, CDF) is adjusted based on ground observations, so that the probability density distribution of radar precipitation is consistent with ground observations, so as to achieve the purpose of correcting the systematic deviation of radar precipitation.

[0176] First, the consistency of radar QPE and ground-observed precipitation on a 10-minute time scale was examined. The difference between the first 10 minutes of accumulated precipitation observed by the ground automatic station and the radar QPE at the exact time (0 minute lag) during the precipitation process, the correlation coefficient, root mean square error, relative deviation and other indicators show that the first 10 minutes of accumulated precipitation observed on the ground and the instantaneous 10 minutes of precipitation estimated by the radar at the exact time are most consistent when the lag is 4 minutes. From the change range of each indicator within 0-9 minutes of the exact time lag, the time difference of cumulative time sampling has a relatively small impact on the deviation, which is only 0.2% overall, indicating that the consistency of the precipitation detected by the radar and the ground observation time matching on the 10-minute scale is good, and PDF sample matching can be performed.

[0177] Secondly, the spatiotemporal matching window of the PDF sample was adjusted by comprehensively considering the characteristics of the occurrence and disappearance time of 10-minute precipitation, the detection range of a single radar, the stability of radar bias changes, and the changes in the density of station networks in winter and summer. Figure 18 shows the spatial distribution of precipitation before and after the correction of the radar precipitation PDF bias and the comparison with ground observations. It can be seen that the precipitation estimated by radar and analyzed by ground observations in the central, western and southern regions in the figure are similar in spatial morphology, but the precipitation center is basically overestimated. After the bias correction, the intensity of the radar precipitation center is weakened and more consistent with ground observations. In some local areas at certain times, such as the precipitation area in the northwest as shown in Figure 18, the radar will miss some precipitation compared with the observation, and it is easy to make an "over-" correction (the precipitation in the precipitation area is generally strong). But in general, the PDF method can still effectively correct the large positive deviations of radar precipitation, such as Fig.19 As shown in the figure, compared with the time before the bias correction, the corrected radar precipitation bias series is closer to the zero value in most cases.

[0178] In winter, the AWS network becomes sparse, and the stability of the matching between radar and ground-observed precipitation samples is greatly reduced. The proportion of non-zero precipitation samples at the minute level is low, and it is difficult to obtain the corrected sampling conditions that meet the PDF stability in general weak precipitation processes. However, from the perspective of some processes with more obvious precipitation ( Fig. 20 ), the original radar precipitation products in winter seriously underestimate the precipitation, and the effect of bias correction is weaker than that in summer, but there is still a certain degree of improvement.

[0179] (G) Fusion method of local 100-meter radar and ground-based precipitation observations;

[0180] Firstly, the radar QPE precipitation after bias correction is used as the initial estimation field, i.e., the background field, and the surface minute precipitation observation data is fused using the OI method to obtain the meltwater precipitation field.

[0181] Specifically, as shown in formula (1):

[0182]

[0183] That is, the analysis value A of the grid k is the initial estimate F of the point k Add the deviation between the observed value of the point and the initial estimate, and the deviation is determined by n known initial estimates F within a certain analysis range. i With the observed value O i The bias-weighted estimate of is obtained;

[0184] OI method to solve the weight W i The linear equation of is shown in formula (2):

[0185]

[0186] in, Indicates the initial estimated field error correlation between point i and point j, represents the observation error co-correlation, λ i is the standard deviation of the observation error at point i and the initial estimate error standard deviation The ratio, λ j is the standard deviation of the observation error at point j and the initial estimate error standard deviation ratio; It indicates the co-correlation of the initial estimated field errors between the k-point and the j-point of the analysis grid;

[0187] As the temporal and spatial resolution of the analysis continues to increase, it is difficult to obtain sufficiently fine observations to statistically calculate a stable and accurate error change curve. This scheme uses the error ratio of the adjusted observations and the initial estimate to calculate the error curve. The OI fusion parameters are determined by the method to solve the weight. Generally speaking, the error of radar data is greater than that of ground observation. After testing, when the error ratio between observation and initial estimation is set to 0.4, the fusion effect is relatively good.

[0188] After the fusion of ground and radar precipitation, the precipitation area is reduced in size and has a more scattered structure, which is more like retaining the finer structure of the radar, but the value is closer to the ground analysis product (Figure 18).

[0189] Secondly, the 100-meter downscaling of precipitation is integrated.

[0190] Taking the kilometer-level / 10-minute fused precipitation field as the background, the 100m radar precipitation information is estimated using the OPE product data based on the 100-meter X-band radar, thereby downscaling the kilometer-level / 10-minute fused data;

[0191] The specific downscaling formula is as follows:

[0192]

[0193] Among them, P 100m,mg is the 100 m fused precipitation after downscaling, P 1km,mg is the radar-ground fusion analysis field of 1 km, P 1km,Rad and P 100m,Rad They are kilometer-level weather radar QPE products and hundred-meter-level X-band radar QPE products.

[0194] Because the X-band radar data covering the Winter Olympics area were only fully collected in June 2021, an example of precipitation in June is given to illustrate the effect of the downscaling analysis.

[0195] like Fig. 22As shown in the figure, at 10:10 UTC on June 30, 2021, ground observations and analysis and two radar QPE products in the southeast of the competition area all showed precipitation. The 1km fused product (Merged) retained the general structure of the 1km radar estimated precipitation (Radar QPE), but the precipitation intensity was significantly lower than that of the radar QPE, which corresponded better to the ground station observation (Gauge). The precipitation structure of the 100m ground rain gauge grid analysis product (GGA_100m) was very smooth, while the 100m downscaled fused product (Downscale) showed more detailed structures of heavy precipitation centers, which were very consistent with the strong center structure shown by the 100m radar estimated precipitation (X-band QPE). This shows that the precipitation product obtained by the final downscaling can indeed effectively introduce the information of the X-band radar, thereby reflecting the more detailed structural characteristics of the precipitation system, and in terms of overall spatial structure, it is consistent with the QPE product of the 1km weather radar, and the value is close to the ground AWS observation. The 10-minute downscaled precipitation is accumulated into 1-hour precipitation. Its spatial structure is very similar to the ART_1km near-real-time precipitation analysis product, and its value is also close to the hourly precipitation observed at the station ( Fig.23 ), indicating that the product has good consistency in time scale with the mature 1km / hourly precipitation product in business applications.

[0196] (H) Evaluation method of local 100-meter precipitation analysis results.

[0197] The fusion effect evaluation adopts the independent inspection method, selecting some stations with higher observation quality as inspection stations, retaining their observation values ​​and not participating in precipitation analysis, taking them as "true values", and interpolating various grid precipitation products to be inspected to the inspection stations through the nearest interpolation, and comparing the differences with the "true values";

[0198] The evaluation indicators include the correlation coefficient Coefficient Correlation, CC, the root mean square error Root Mean Square Error, RMSE, and the relative bias Relative Bias, Rbias statistical error. The specific calculation formula is as follows:

[0199]

[0200] Among them, g i is the precipitation observed at the inspection station, i.e. the “true value” of precipitation; p i is the precipitation of each precipitation product interpolated to the inspection station; n is the number of statistical matching samples, The average value of precipitation interpolated to the inspection station for each inspected precipitation product; is the average precipitation observed at the inspection station.

[0201] There are more than 7,000 effective observation stations in the Beijing-Tianjin-Hebei region for analysis in summer. Among them, 414 national automatic stations (benchmark climate stations, basic meteorological stations and general stations) were selected as test stations, and the rest were used as analysis stations for the analysis of 1km / 10-minute precipitation. In winter, due to the suspension of a large number of tipping bucket rain gauges in AWS in northern China, the station network in the Beijing-Tianjin-Hebei region became sparse. In order to ensure sufficient observations for analysis (about 1,700 stations), only 79 national benchmark climate stations and basic meteorological stations were selected as test stations. It can be found that the station network density in the central and southern parts of the Beijing-Tianjin-Hebei region is relatively high in summer, and the northern region is relatively sparse compared to the central and southern parts. In winter, the station network in other regions becomes very sparse except for Beijing and Tianjin.

[0202] For the 100m / 10min precipitation product of the Beijing Winter Olympics competition area, the density of the observation station network in the competition area and its surrounding areas is very different in summer and winter. The station network is relatively dense in summer, with more than 450 automatic stations for observation. Some (44) regional observation stations with station level code 99 in the competition area can be reserved as inspection stations ( Fig.24a The number of effective observation stations in the winter competition area has been reduced to 33, and three stations with good quality and complete data near the core competition areas of Zhangjiakou and Yanqing have been selected as inspection stations ( Figure 24b ).

[0203] Table 4

[0204]

[0205] The statistical methods and indicators are as described above. In summer, the statistical test results of the period with precipitation in multiple precipitation processes in mid-July (11-20) (Table 4) were selected as a reference for method testing. In winter, the corresponding period with precipitation in the Winter Olympics area from February 14 to 28, 2021 (Table 4) was selected, which included three snowfall weather processes during the 2021 Winter Olympics test matches ( Fig.25 ).

[0206] Winter Olympics 100m / 10min precipitation product inspection

[0207] Judging from the error statistical indicators of multiple precipitation processes in the competition area in mid-July 2021 (Table 5), the indicators of 1km fusion analysis downscaling are relatively good, with the highest correlation coefficient and the lowest relative deviation with ground inspection station observations, indicating that this scheme has a relatively good fusion analysis effect in summer. The accuracy of directly performing grid analysis of site observations at 100m is not better than the results of downscaling analysis at 1km, indicating that the current AWS station network density and the OI analysis method used in the competition area are more suitable for analysis at the 1km scale. The 100m radar QPE product itself has relatively low quality and is systematically low, but applying it to the downscaling of 1km analysis products can still improve the accuracy of the analysis results.

[0208] Table 5 Error statistics of 100m precipitation products in the Winter Olympics area for multiple precipitation processes in July 2021 (44 inspection stations) and February (3 inspection stations)

[0209]

[0210] From the error of various 100-meter products at different intensities ( Fig.26a and Figure 26b ), all products are low overall. Compared with other products, the low degree of the fused downscaled product is reduced at all levels. The improvement of RMSE is mainly reflected in heavy precipitation, while the error of weak precipitation increases.

[0211] Compared with summer, the fusion effect of local winter precipitation products is not as significant (Table 5). The correlation of the 100m rapid ground analysis product is higher, but the precipitation intensity is significantly lower. The relative deviation and root mean square error of the 1km fusion analysis downscaling product are the smallest. From the scatter plot, it can be seen that the improvement is more obvious for precipitation with larger intensity ( Fig. 27 ).

Claims

1. Local 100-meter minute-level temperature, humidity, wind and precipitation real-time analysis system, characterized by: It includes the local 100-meter temperature, humidity and wind real-time analysis subsystem and the local 100-meter precipitation real-time analysis subsystem; The local 100-meter temperature, humidity and wind real-time analysis subsystem includes a static parameter localization module, a ground minute temperature, humidity and wind observation data acquisition and preprocessing module, an ART-1km real-time analysis product preprocessing analysis module, a local 100-meter temperature, humidity and wind fusion analysis module, a local 100-meter temperature, humidity and wind analysis result evaluation module, a local 100-meter temperature, humidity and wind fusion analysis result visualization module and a local 100-meter temperature, humidity and wind fusion analysis result distribution module; The local 100-meter precipitation real-time analysis subsystem includes a static parameter localization module, a ground minute precipitation observation data acquisition module, a ground minute precipitation observation preprocessing module, a kilometer-level weather radar QPE product preprocessing module, a kilometer-level ground minute precipitation OI difference module, a kilometer-level weather radar QPE product deviation correction module, a 100-meter-level X-band radar QPE product preprocessing module, a local 100-meter precipitation fusion analysis module, a local 100-meter precipitation analysis result evaluation module, a local 100-meter precipitation fusion analysis result visualization module, and a local 100-meter precipitation fusion analysis result distribution module.

2. The local 100-meter minute-level temperature, humidity, wind, and precipitation real-time analysis system according to claim 1 is characterized in that: The system is deployed on the "Pai-Shuguang" high-performance computing cluster, and the analysis area is dynamically transformed according to parameter configuration; The system input data includes: ground minute temperature, humidity, wind speed, and precipitation observation data from ground observation stations; kilometer-level weather radar QPE products; hundred-meter-level X-band radar QPE products; ART-1km real-time analysis products; Fusion analysis methods: "downscaling + STMAS" and "PDF + OI + downscaling" fusion analysis methods; System output products: 100-meter-level real-time analysis products with a resolution of 100m / 10min for precipitation, 2m for temperature, 2m for relative humidity, 10m for U wind speed, and 10m for V wind speed. The product format is NetCDF. The operation process is as follows: according to the needs, the ground minute temperature, humidity, wind speed observation data and temperature, humidity, wind speed and ART-1km live analysis products are configured into the local 100-meter temperature, humidity and wind live analysis subsystem; the ground minute precipitation observation data, kilometer-level weather radar QPE products and 100-meter-level X-band radar QPE products are configured into the local 100-meter precipitation live analysis subsystem. Then, according to the configuration parameters, static parameter localization modules are created in the local 100-meter temperature, humidity and wind live analysis subsystem and the local 100-meter precipitation live analysis subsystem respectively; In the local 100-meter temperature, humidity, and wind real-time analysis subsystem: provide the ground minute temperature, humidity, and wind speed observation data to the ground minute temperature, humidity, and wind observation data acquisition and preprocessing module, configure the ART-1km real-time analysis product to the real-time analysis product preprocessing analysis module, and then configure the preprocessed data to the local 100-meter temperature, humidity, and wind fusion analysis module to realize dynamic regional local 100-meter fusion real-time analysis, use the local 100-meter temperature, humidity, and wind analysis result evaluation module to evaluate, and use the local 100-meter temperature, humidity, and wind fusion analysis result visualization module to visualize, output product pictures, and distribute to application units through the local 100-meter temperature, humidity, and wind fusion analysis result distribution module; In the local 100-meter precipitation real-time analysis subsystem: the ground minute precipitation observation data is configured to the ground minute precipitation observation data acquisition module and the ground minute precipitation observation preprocessing module; the kilometer-level weather radar QPE products and the 100-meter-level X-band radar QPE products are configured to the kilometer-level weather radar QPE product preprocessing module and the 100-meter-level X-band radar QPE product preprocessing module respectively; the kilometer-level ground minute precipitation OI difference module, the kilometer-level weather radar QPE product deviation correction module and the local 100-meter precipitation fusion analysis module fuse the preprocessed data for real-time analysis, and the local 100-meter precipitation analysis result evaluation module is used for evaluation. The local 100-meter precipitation fusion analysis result visualization module is used for visualization, and product images are output. The data are distributed to application units through the local 100-meter precipitation fusion analysis result distribution module.

3. The method for constructing a local 100-meter minute-level temperature, humidity, wind, and precipitation real-time analysis system according to any one of claims 1-2, characterized in that: It includes the construction method of the local 100-meter temperature, humidity and wind real-time analysis subsystem and the construction method of the local 100-meter precipitation real-time analysis subsystem; The construction method of the local 100-meter temperature, humidity and wind real-time analysis subsystem includes: (1) Acquisition and preprocessing of ground minute temperature, humidity and wind observation data; (2) Pretreatment methods for real-time analysis products; (3) Temperature, humidity and wind fusion analysis method based on machine learning; (4) Evaluation method of local 100-meter temperature, humidity and wind analysis results; (5) Complete the visualization of the local 100-meter temperature, humidity and wind fusion analysis results, and generate the visualization results of the local 100-meter temperature, humidity and wind fusion analysis results; The local 100-meter precipitation real-time analysis subsystem method includes: (A) Acquisition of ground-level minute precipitation observation data; (B) Preprocessing of surface minute precipitation observations; (C) Kilometer-level surface minute precipitation OI interpolation module; (D) Kilometer-level weather radar QPE product preprocessing; (E) Preprocessing of 100-meter-class X-band radar QPE products; (F) Bias correction of kilometer-level weather radar QPE products; (G) Fusion method of local 100-meter radar and ground-based precipitation observations; (H) Evaluation method of local 100-meter precipitation analysis results; (I) Complete the visualization of the 100-meter precipitation analysis results and generate the visualization results of the 100-meter precipitation analysis results.

4. The method for constructing a local 100-meter minute-level temperature, humidity, wind, and precipitation real-time analysis system according to claim 3 is characterized in that: In step (1), the minute temperature, humidity and wind observation data observed on the ground are 100-meter / 10-minute temperature, humidity and wind observation data, which are derived from the minute temperature observation data, minute relative humidity observation data and minute wind speed and direction observation data of the automatic weather station; the preprocessing method is: through coordinate conversion, variable conversion and quality control processing, standard quality control observation data is generated; In step (2), the live analysis products are: the national 1km and 1h live analysis products released by the National Meteorological Information Center and the digital elevation model products resampled to 100m resolution based on satellite observations; The preprocessing method using spatiotemporal multi-scale analysis technology and height correction algorithm is as follows: the spatiotemporal resolution interpolation of the national 1km and 1h real-time analysis products is performed to obtain background field data, and the digital elevation model data with sub-100-meter resolution is used to fuse the background field and observation data using a multi-grid variational algorithm, thereby performing height correction on the background field, producing background field data with a 100-meter resolution at the entire point, and obtaining 100m / 1h real-time fusion data.

5. The method for constructing a local 100-meter minute-level temperature, humidity, wind, and precipitation real-time analysis system according to claim 4 is characterized in that: In step (3), the observation data of minute-level temperature, relative humidity, wind speed and direction from the ground automatic meteorological observation station, the 100m / 1h real-time fusion data, and the slope and aspect terrain information are used as the main data sources. A machine learning method is used to establish a station error inversion model, and research on the fusion real-time analysis technology of temperature, relative humidity and wind speed and direction at the 100m / 10min scale is carried out. Machine learning methods include gradient boosting decision tree GBDT, random forest RF, Adaboost, K nearest neighbor classification algorithm KNN, support vector machine SVM, ER-FR model, Model correction for machine learning methods is divided into two parts: The first part is the training phase: using preprocessed historical data as training samples, we conduct feature selection, model parameter tuning, and performance optimization training for different machine learning models to obtain the optimal training model. The second part is the testing phase: using the optimal training model, we perform feature extraction and parameter optimization to obtain the optimal error inversion model, and then carry out background corrections for temperature and humidity. We then evaluate the corrected data results for each model, analyze the performance characteristics of different models, calculate the correction weights for each model, and integrate the correction results of different models according to the weights. In step (4), temperature, humidity and wind are evaluated using correlation coefficient, mean error and root mean square error.

6. The method for constructing a local 100-meter minute-level temperature, humidity, wind, and precipitation real-time analysis system according to claim 3 is characterized in that: In the local 100-meter precipitation real-time analysis subsystem method, in step (A): the ground minute precipitation observation data uses the minute-by-minute precipitation data observed by the China Meteorological Administration's automatic weather station obtained from the big data cloud platform, AWS, including minute-by-minute precipitation data set at the encrypted observation automatic station; In step (B), preprocessing includes: based on the station ID numbers collected within the analysis area, statistically calculating the cumulative precipitation for the same station (i.e., the same station ID number) at 00:00 / 10:00 / 20:00 / 30:00 / 40:00 / 50:00 every hour, as the 10-minute observed precipitation at 00:00 / 10:00 / 20:00 / 30:00 / 40:00 / 50:00 at that station; selecting only data with quality control codes 0 / 1 / 3 / 4 / 9 for statistical calculation; and performing preliminary quality control on the accumulated 10-minute precipitation data before analysis. This includes checking precipitation element boundary values ​​to remove gross errors and eliminating stations with long-term "zero value" observations. In step (C), minute-level precipitation observation data from ground observation stations are gridded using the optimal interpolation (OI) gridding analysis technology based on the climate background field ratio. This includes a real-time updated analysis station table and a distance index file between analysis grid points and stations. An analysis station selection strategy that adapts to the station network density and automatically optimizes the analysis radius. Regionally limiting the maximum analysis radius avoids analyzing a "large circle" precipitation structure due to overly sparse stations. Furthermore, processing is added to handle cases where the satellite background field is zero over a large area in northern winter. This better preserves the structure of the ground-based precipitation grid analysis.

7. The method for constructing a local 100-meter minute-level temperature, humidity, wind, and precipitation real-time analysis system according to claim 3 is characterized in that: In step (C), the kilometer-level weather radar QPE product data is a 10-minute rapid radar QPE product developed by the National Meteorological Information Center for real-time, high-efficiency precipitation applications. The kilometer-level weather radar QPE product is generated by inverting the radar base data or volume scan data of the lower four layers obtained from the NAS disk of the big data cloud platform. The specific preprocessing method is as follows: first, after eliminating isolated points, radial interference, ground interference, and wave echo quality control, a mixed scan rate is calculated, and a precipitation rate estimate is calculated based on the ZI relationship. This is converted into the precipitation in the last 10 minutes, and a single-station precipitation product is generated every 3 minutes. Second, the single-station precipitation product is gridded in longitude and latitude and is regularly activated at 10-minute intervals. The data of each station within the previous 10 minutes and the most recent time are used to perform maximum network mosaic to generate a 10-minute precipitation product for the Chinese region. In step (D), the 100-meter X-band radar QPE product is based on data from 11 X-band radars collected in Beijing and Hebei Province. It uses the same preprocessing method as the kilometer-level weather radar QPE product. When the 100-meter X-band radar QPE product can detect precipitation systems, it is used to provide high-resolution precipitation structure information, thereby achieving downscaling from the kilometer level to the 100-meter level.

8. The method for constructing a local 100-meter minute-level temperature, humidity, wind, and precipitation real-time analysis system according to claim 3 is characterized in that: In step (F), the PDF matching method is used to correct the deviation of radar QPE; The principle of the PDF matching method is to assume that the probability density distribution of precipitation is relatively stable within a certain spatial and temporal range, and then adjust the radar precipitation value under the same cumulative probability based on ground observations so that the probability density distribution of radar precipitation is consistent with that of ground observations, thereby correcting the systematic deviation of radar precipitation. The specific method is: First, the consistency between radar QPE and surface-observed precipitation on the 10-min time scale is examined; Secondly, the spatiotemporal matching window of the PDF sample is adjusted by comprehensively considering the characteristics of the occurrence and disappearance time of 10-minute precipitation, the detection range of a single radar, the stability of radar bias changes, and the changes in station network density in winter and summer.

9. The method for constructing a local 100-meter minute-level real-time analysis system for temperature, humidity, wind, and precipitation according to claim 8, characterized in that: In step (G), First, the bias-corrected radar QPE precipitation is used as the initial estimate field, i.e., the background field, and the meltwater precipitation field is obtained by fusing the ground minute precipitation observation data using the OI method. Specifically as shown in formula (1): That is, the analysis value A of the grid k is the initial estimate F of the point k Add the deviation between the observed value of the point and the initial estimate, and the deviation is determined by n known initial estimates F within a certain analysis range. i With the observation value O i The bias-weighted estimate of is obtained; OI method to solve the weight W i The linear equation is shown in formula (2): in, Indicates the co-correlation of the initial estimated field errors at point i and point j, represents the co-correlation of observation errors, λ i is the standard deviation of the observation error at point i and the initial estimate error standard deviation The ratio, λ j is the standard deviation of the observation error at point j and the initial estimate error standard deviation ratio; It indicates the co-correlation of the initial estimated field errors between the analysis grid points k and j; The error ratio between the adjusted observation and the initial estimate is used The method is used to determine the OI fusion parameters to solve the weights. When it is set to 0.4, the fusion effect is the best; Secondly, using the kilometer-level / 10-minute fused precipitation field as the background, the 100-meter radar precipitation information is estimated using the OPE product data based on the 100-meter X-band radar, thereby downscaling the kilometer-level / 10-minute fused data. The specific downscaling formula is as follows: Among them, P 100m,mg is the 100m fused precipitation after downscaling, P 1km,mg is the radar-ground fusion analysis field of 1 km, P 1km,Rad and P 100m,Rad They are kilometer-level weather radar QPE products and hundred-meter-level X-band radar QPE products.

10. The method for constructing a local 100-meter minute-level real-time analysis system for temperature, humidity, wind, and precipitation according to claim 3, characterized in that: In step (H), the fusion effect is evaluated using an independent verification method. Some stations with high observation quality are selected as verification stations. Their observations are retained and excluded from precipitation analysis. These observations are used as "true values". The various gridded precipitation products to be tested are interpolated to the verification stations through neighboring interpolation and compared with the "true values". The evaluation indicators include the correlation coefficient (CC), the root mean square error (RMSE), and the relative bias (Rbias). The specific calculation formulas are as follows: Among them, g i is the precipitation observed at the inspection station, i.e. the "true value" of precipitation; p i is the precipitation amount at the inspection station interpolated from each inspected precipitation product; n is the number of statistical matching samples, The average value of precipitation interpolated to the testing station for each tested precipitation product; is the average precipitation observed at the inspection station.

Citation Information

Cited By

  • Spatial downscaling method and system for obtaining national 1-kilometer daily-scale meteorological grid live analysis data of more than 30 years

    CN120429640A

  • Radar rainfall deviation correction algorithm based on combination of probability density matching and climate state ratio method

    CN121069390A

  • A radar precipitation bias correction algorithm based on the combination of probability density matching and climatological ratio method

    CN121069390B