An all-weather assimilation method and device for satellite microwave temperature and humidity sounder based on retrieval of temperature and humidity profiles in cloud regions

Abstract

The present invention provides an all-weather assimilation method and device for a satellite microwave temperature and humidity sounder based on the inversion of cloud area temperature and humidity profiles. The method includes the following steps: obtaining satellite data, conventional observation data, and forecast field data; outputting the satellite data determined to be non-clear sky, and inversely obtaining the atmospheric temperature and humidity profiles based on the observed brightness temperature; converting the atmospheric temperature and humidity profiles into the prebufr format in the form of radiosonde data to obtain the non-clear sky area profile information in the prebufr format; performing bias correction on the satellite data determined to be clear sky to obtain the corrected microwave brightness temperature data for the clear sky area; performing quality control on the corrected microwave brightness temperature data for the clear sky area to eliminate the observation data with low quality; putting the non-clear sky area profile information converted into the prebufr format and the microwave brightness temperature data for the clear sky area into the WRFDA assimilation system for data assimilation, and outputting the assimilated analysis field. The present invention can improve the precipitation forecast effect of numerical models.

Description

Technical Field

[0001] The present invention relates to an all-weather assimilation method for satellite microwave temperature and humidity sounders based on the inversion of cloud region temperature and humidity profiles, and belongs to the technical field of satellite data assimilation. Background Art

[0002] Data assimilation is used to provide accurate and reasonable initial conditions for numerical models and has received increasing attention. Data assimilation should make full use of existing information to determine a most likely accurate atmospheric motion state. The sources of assimilated observation data mainly include conventional observations and non-conventional observation data such as radar and satellite data. Among them, the observation time and location of conventional observation data are fixed and mostly distributed in densely populated areas, etc., with low spatio-temporal resolution and uneven distribution. There are very few conventional observations in remote areas such as over the ocean. Satellite observations have the characteristics of all-weather, full coverage, and consistent data. With the continuous increase in the number of launched meteorological satellites, satellite data has made great contributions to improving the model initial field.

[0003] After years of development, the theory of satellite data assimilation has made significant progress, and the problem of satellite radiance data assimilation under clear sky conditions has been basically solved. However, the assimilation research of satellite data in cloud regions is still not perfect, which hinders the development of the theory of satellite data assimilation. From the perspective of the application of satellite data assimilation, due to the lack of the theory of satellite data assimilation in cloud regions, a large amount of satellite data affected by clouds and precipitation is discarded without being used. However, the atmospheric information contained in cloud regions is closely related to the occurrence and development of various weather systems such as typhoons, southwest vortices, and mei-yu fronts, which makes the assimilation application of cloud region data particularly important. Therefore, the research on satellite data assimilation in cloud regions is of great significance for improving the basic theory of satellite data assimilation and the forecasting level. Microwave radiation has a certain penetration ability for clouds and precipitation, and has stronger detection ability for cloud regions than infrared radiation, making the assimilation application of this part of data a hot research topic in the field of satellite data assimilation.

[0004] In the past decade or so, some scholars have carried out technical research and business applications in this field and achieved some progress. In 2010, Geer and Bauer of ECMWF proposed the symmetric cloud amount, using the cloud amount calculated by the polarization difference of the microwave imager at 37 GHz as a statistical factor, obtaining the full-field error distribution of clear sky and cloudy conditions under full-sky assimilation, further solving the problem of non-Gaussian distribution of errors in cloud and rain areas, and realizing the all-weather direct assimilation of SSM / I and AMSR-E data of microwave imagers. Yang Chun (2017) adjusted the wet control variables of the model based on the WRFDA system and added relevant data interfaces such as the symmetric observation error model using cloud amount as a prediction factor for full-sky assimilation, realizing the direct assimilation of AMSR2 full-sky emissivity. Xian Zhipeng (2019) connected the scattering module of RTTOV to the WRFDA system and established an observation error model for MWHS2 based on the symmetric error model proposed by Geer and Bauer, thus realizing the all-weather assimilation of MWHS2.

[0005] The above methods all belong to the direct assimilation method, which requires considering the scattering part in the observation operator, providing observation errors that satisfy the Gaussian distribution for the cloud area, and taking wet physical parameters as control variables for minimizing iterative calculations during assimilation. However, due to some bottleneck problems, it is still restricted, such as: more perfect wet physical parameterization schemes, faster radiative transfer models with higher accuracy for Mie scattering, correction of satellite data biases in cloud and rain areas, etc. The direct assimilation technology in the cloud area is not yet mature, the utilization of observation data in the cloud area is not sufficient, and the initial information of temperature, humidity, wind, clouds, and precipitation in the cloud area of numerical models is not accurately described. Summary of the Invention

[0006] The purpose of the present invention is to address the problems existing in the all-weather assimilation of existing satellite microwave observation data, and propose a method and device for all-weather assimilation of satellite microwave temperature and humidity meters based on the inversion of temperature and humidity profiles in the cloud area.

[0007] To achieve the above object, the present invention is implemented by the following technical solutions:

[0008] In the first aspect, the present invention provides a method for all-weather assimilation of satellite microwave temperature and humidity meters based on the inversion of temperature and humidity profiles in the cloud area, the method comprising the following steps:

[0009] Obtain satellite data, conventional observation data, and forecast field data; interpolate the ground and atmospheric profile information of the forecast field to the position where the satellite observations are located; the satellite data includes microwave thermometer and microwave humidity meter data;

[0010] Based on the cloud amount data in the satellite data, determine whether the observation point is a clear day or a non-clear day;

[0011] Output the satellite data determined to be non - sunny days, and retrieve the atmospheric temperature and humidity profiles based on the observed brightness temperature; convert the atmospheric temperature and humidity profiles into the prebufr format in the form of sounding data to obtain the non - clear - sky area profile information in the prebufr format;

[0012] Perform bias correction on the satellite data determined to be sunny days to obtain the microwave brightness temperature data of the corrected clear - sky area;

[0013] Perform quality control on the microwave brightness temperature data of the corrected clear - sky area, and reject the observed data with low quality;

[0014] Put the non - clear - sky area profile information converted into the prebufr format and the microwave brightness temperature data of the clear - sky area into the WRFDA assimilation system for data assimilation, and output the assimilated analysis field.

[0015] Furthermore, the method for performing quality control on the satellite data to obtain the corrected satellite data includes the following steps:

[0016] Read the longitude, latitude and time of the observed pixels in the satellite data, reject the observations outside the assimilation time window and outside the model area range, and perform preliminary quality control;

[0017] Check whether the observed brightness temperature of all satellite pixels exceeds the maximum and minimum thresholds, and reject the obvious outliers;

[0018] Judge the surface type according to the land - sea mask in the satellite data, and reject all observations of mixed surfaces;

[0019] Reject the observed data of the first 5 and last 5 scan angles of each scan line of the satellite data;

[0020] In the rainfall judgment, select the total cloud water as the detection index. When the total cloud water at a certain observation point is greater than 0.2, it is considered that the observation is affected by rainfall, and reject this observation point;

[0021] Reject the observed data in the satellite data where the observation residual (Obs - Background, observed brightness temperature minus simulated brightness temperature) is greater than 15K; K is the unit of brightness temperature, that is, Kelvin;

[0022] Reject the observed data in the satellite data where the observation residual is greater than three times the error standard deviation.

[0023] Furthermore, the method for judging whether the corrected satellite data belongs to a sunny day or a non - sunny day includes:

[0024] Perform field - of - view angle matching based on the cloud amount product of the Medium - Resolution Spectral Imager (MERSI). If the average value of the cloud amount product within the instantaneous field of view of the microwave vertical sounder is greater than 70%, then judge that this view point is a cloudy day.

[0025] Further, the method for updating the deviation correction coefficient includes the following steps:

[0026] Step A: The variational deviation correction predictor coefficient at the initial time t0 is from other similar microwave vertical sounders.

[0027] Step B: Run the WRFDA system at the initial time t0 to obtain the updated deviation correction coefficient.

[0028] Step C: Use the updated deviation correction coefficient as the correction coefficient for the next time t1.

[0029] Step D: Run the WRFDA system at time t1 to obtain the updated deviation correction coefficient.

[0030] Step E: Use the updated deviation correction coefficient as the correction coefficient for the next time t2.

[0031] Step F: Repeat the process of Steps D - E at the next assimilation time to cyclically update the deviation correction coefficient.

[0032] Further, the method for deviation correction includes the following steps:

[0033] Select the model thickness, surface temperature, cloud liquid water content, and scan position as correction factors; multiply the predictor by the corresponding correction coefficient to obtain the deviation correction amount; subtract the deviation correction amount from the original brightness temperature to obtain the corrected brightness temperature.

[0034] Further, the method for retrieving the atmospheric temperature and humidity profiles includes the following steps:

[0035] Obtain the satellite observation information in ASCII format output after the first assimilation of the system. The satellite observation information includes the longitude, latitude, brightness temperature, cloud amount, observation angle, sea - land code, and surface type information of each observation point; input the brightness temperature, latitude, and observation angle into the atmospheric temperature and humidity profile inversion model to obtain the atmospheric temperature and humidity profiles.

[0036] Further, the construction method of the atmospheric temperature and humidity profile inversion model includes the following steps:

[0037] Obtain satellite data and re - analysis data, and perform spatio - temporal matching; classify the profile samples into clear - sky and cloudy regions based on cloud products to establish a sample data set;

[0038] Based on the BP artificial neural network algorithm, develop the atmospheric temperature and humidity profile inversion model for ocean and land areas and under clear - sky and cloudy - sky conditions;

[0039] Carry out the atmospheric temperature and humidity profile inversion experiment for ocean and land areas and under clear - sky and cloudy - sky conditions, and verify the inversion accuracy of the model.

[0040] In a second aspect, the present invention provides a satellite microwave thermometer all-weather assimilation device based on cloud area temperature and humidity profile inversion, comprising:

[0041] Acquisition module: used to acquire satellite data, conventional observation data and forecast field data; interpolate the ground and atmospheric profile information of the forecast field to the location of satellite observation; the satellite data includes microwave thermometer and microwave hygrometer data;

[0042] Sunny or rainy judgment module: used to judge whether the observation point is sunny or cloudy based on the cloud cover data in the satellite data;

[0043] Conversion module: used to output satellite data identified as non-clear sky, and obtain atmospheric temperature and humidity profiles based on observed brightness temperature inversion; convert atmospheric temperature and humidity profiles into prebufr format in the form of sounding data, and obtain non-clear sky area profile information in prebufr format;

[0044] Bias correction module: used to correct the bias of satellite data identified as clear sky and obtain the corrected microwave brightness temperature data of clear sky area;

[0045] Output module: used to perform quality control on the corrected microwave brightness temperature data of the clear sky area and eliminate the observation data with low quality; put the non-clear sky area profile information converted into prebufr format and the microwave brightness temperature data of the clear sky area into the WRFDA assimilation system for data assimilation, and output the assimilated analysis field.

[0046] In a third aspect, the present invention provides a satellite microwave thermometer all-weather assimilation device based on cloud area temperature and humidity profile inversion, including a processor and a storage medium;

[0047] The storage medium is used to store instructions;

[0048] The processor is used to operate according to the instructions to execute the steps of the method described in the first aspect.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] 1. In the present invention, the satellite data identified as non-clear sky are output, and the atmospheric temperature and humidity profile is inverted according to the observed brightness temperature; the atmospheric temperature and humidity profile is converted into the prebufr format in the form of sounding data, and the non-clear sky area profile information converted into the prebufr format and the microwave brightness temperature data of the clear sky area are put into the WRFDA assimilation system for data assimilation, and the assimilated analysis field is output, and the atmospheric temperature and humidity profile is inverted and converted into the prebufr format to make the assimilation process more efficient and accurate.

[0051] 2. In the present invention, before constructing the atmospheric temperature and humidity profile inversion model, it is necessary to classify the training samples and select the cloud and sky samples. In addition, when performing all-weather assimilation, it is also necessary to distinguish the observations in the cloud area. Therefore, cloud detection is carried out based on the MERSI cloud amount product of the same platform (FY-3D): The field of view angles of MERSI (the slave instrument) and the microwave vertical sounder MWTS2 / MWHS2 (the master instrument) are matched. If the observation vector of the slave instrument is less than 1 / 2 of the instantaneous field of view angle of the master instrument, it is considered that the observation point of the slave instrument falls within the instantaneous field of view of the master instrument. If the average value of the cloud product within the instantaneous field of view is greater than 70 (the maximum is 100), it is considered that the observation is a cloud and sky condition. This method can improve the accuracy of sample classification.

[0052] 3. In the present invention, the bias correction coefficient is updated through cyclic assimilation. Data assimilation assumes that the observation error satisfies a normal distribution with a mean of zero. However, due to multi-source errors such as instrument sensitivity, sensor response characteristics, calibration, and radiation transfer models, there will be a systematic bias between the satellite observed brightness temperature and the simulated brightness temperature based on the model background field profile. Therefore, before assimilating satellite data, it is necessary to correct the bias of the observed brightness temperature to make the observation-background difference tend to be unbiased. In order to provide an accurate coefficient for the bias correction factor at the initial moment of the assimilation experiment, cyclic update is carried out through one month of assimilation, so that the coefficient has sufficient startup time to adjust the MWTS2 and MWHS2 data. The corrected observation-background difference is closer to a distribution with a mean of zero, the standard deviation decreases, and the systematic bias is largely eliminated, meeting the assimilation assumption.

[0053] 4. In the present invention, all-weather assimilation is achieved by assimilating the retrieved atmospheric temperature and humidity profiles: The observation points determined to be in the cloud area are selected and the atmospheric temperature and humidity profiles are retrieved. The atmospheric temperature and humidity profiles are converted into the PREPBUFR format in the form of conventional data. The temperature and humidity profile information in the cloud area and the brightness temperature data in the clear sky area converted into the PREPBUFR format enter the assimilation system together to achieve all-weather assimilation of FY-3D MWTS2 / MWHS2 data. This can improve the utilization rate of cloud area data, improve the initial information such as the temperature field and humidity field in the model, especially in the cloud area, and thus improve the forecasting effect of the numerical weather prediction model.

[0054] 5. The present invention is based on machine learning and solves the problem of retrieving the atmospheric temperature and humidity profiles in cloud areas by constructing a BP artificial neural network: The neural network takes a series of inputs, weights them through connection weights and sends them to the hidden layer. After summarizing all the inputs, each neuron in the hidden layer generates a certain response output through a transfer function and outputs it to the output layer through the connection weights of the next layer. After summarizing all the inputs, various neurons in the output layer generate a response output. Compared with traditional mathematical statistics methods, the BP artificial neural network can solve non-linear problems, and also has much lower requirements for the sample size. It does not require the samples to be independent or follow a normal distribution, and also has strong fault tolerance, that is, the discrimination accuracy of the network system is generally not affected by the noise in the samples. After training, the network can simulate the radiation transfer process and invert the input brightness temperature data of the microwave vertical sounder into high-precision atmospheric temperature and humidity profiles. Description of the Drawings

[0055] Figure 1 is the technical roadmap of satellite data assimilation for all-weather atmospheric vertical sounders;

[0056] Figure 2 is the structural diagram of the WRF model system;

[0057] Figure 3 is the flowchart of the WRFDA assimilation system;

[0058] Figure 4 is the process of data quality control and bias correction for microwave vertical sounder data;

[0059] Figure 5 is the schematic diagram of the variation of the bias prediction factors of each channel of MWTS2 with the number of cycle updates;

[0060] Figure 6 is the schematic diagram of the variation of the bias prediction factors of each channel of MWHS2 with the number of cycle updates;

[0061] Figure 7 is the schematic diagram of the average bias of each channel before and after using bias correction;

[0062] Figure 8 is the schematic diagram of the standard deviation of the bias of each channel before and after using bias correction;

[0063] Figure 9 is the schematic diagram of the technical roadmap for establishing the sample library of atmospheric temperature and humidity profile inversion;

[0064] Figure 10 is the schematic diagram of the principle of BP artificial neural network training;

[0065] Figure 11 is the schematic diagram of the structure of the BP artificial neural network;

[0066] Figure 12It is the flowchart of the atmospheric temperature and humidity profile inversion experiment;

[0067] Figure 13 It is the schematic diagram of the root mean square error (RMSE) of the inversion in the ocean area; (from left to right: clear-sky temperature, cloudy-sky temperature, clear-sky humidity, cloudy-sky humidity)

[0068] Figure 14 It is the schematic diagram of the root mean square error (RMSE) of the inversion in the land area; (from left to right: clear-sky temperature, cloudy-sky temperature, clear-sky humidity, cloudy-sky humidity)

[0069] Figure 15 It is the flowchart of the all-weather assimilation system for the microwave vertical sounder data;

[0070] Figure 16 It is the probability distribution diagrams of O - B and O - A before and after the bias correction of the 11th channel of MWHS2;

[0071] Figure 17 It is the schematic diagram of the simulated and observed brightness temperatures before and after the assimilation of the 11th channel of MWHS2;

[0072] Figure 18 It is the schematic diagram of the average deviation of temperature and geopotential height; (left: temperature, right: geopotential height)

[0073] Figure 19 It is the schematic diagram of the TS score;

[0074] Figure 20 It is the schematic diagram of the ETS score;

[0075] Figure 21 It is the schematic diagram of the TS score at different time periods; (a: 0 - 6h, b: 6 - 12h, c: 12 - 18h, d: 18 - 24h). Specific implementation manners

[0076] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0077] Embodiment 1:

[0078] This embodiment provides an all-weather assimilation method for the data of the Microwave Temperature Sounder-2 (MWTS2) and the Microwave Humidity Sounder-2 (MWHS2) on the Fengyun-3D (FY-3D) satellite, and constructs an all-weather assimilation system for the FY-3D MWTS2 and MWHS2 data based on this method. The system is developed and extended on the basis of the Weather Research and Forecasting Data Assimilation (WRFDA) system. The specific functions include the quality control and bias correction of MWTS2 and MWHS2, the inversion of atmospheric temperature and humidity profiles based on the artificial neural network algorithm, the PREPBUFR format conversion, and the all-weather assimilation of MWTS2 and MWHS2. Among them, the assimilation part adopts direct assimilation in the clear sky area and inversion assimilation in the cloud area. The technical route of all-weather satellite data assimilation of the atmospheric vertical sounder is as follows Figure 1 As shown, the detailed technical solutions of each part are described in detail below.

[0079] It should be noted that the method of this embodiment is adapted to the microwave vertical sounders and their data on most satellites. Only the data of the Fengyun-3D satellite are used for illustration, and the protection scope of this application should not be limited thereby.

[0080] 1 Data and Model

[0081] 1.1 FY-3D Microwave Temperature and Humidity Sounder Data

[0082] The vertical microwave sounders used in the present invention are MWTS2 and MWHS2 on FY-3D. Compared with MWTS on FY3-A / B satellites, MWTS2 has increased to 13 sounding channels, the sounding scan period is 8 / 3 seconds, each scan line observes 90 Earth fields of view, and the sub-satellite point resolution is 33 km. The sounding frequencies in the range of 50-60 GHz are used to detect surface information and the atmospheric temperature distribution state from the ground to the 3 hPa altitude. Table 1.1.1 gives the instrument channel information of MWTS2.

[0083] Table 1.1.1 MWTS2 Channel Parameter Design

[0084]

[0085]

[0086] Compared with MWHS on FY3 - A / B satellites, the MWHS2 detector has increased to 15 detection channels. The detection and scanning period is 8 / 3 seconds. Each scanning line observes 98 Earth fields of view, and the sub - satellite point resolution is 15 km. MWHS2 also has detection frequency points at 183.31 GHz and 118.75 GHz. Among them, the detection frequency point at 183.31 GHz is the same as the main detection frequency point of the original MWHS, but the number of detection channels has increased. There are 5 detection channels in the sub - divided channel settings for vertical detection of atmospheric humidity. The detection frequency point at 118.75 GHz is located in the oxygen absorption line, and it is the first microwave detector in the world to use this detection frequency point operationally. There are 8 sub - divided detection channels for high - spatial detection of atmospheric temperature. The auxiliary detection frequency points are located at 89 GHz and 150 GHz in the atmospheric window area for background microwave radiation detection and precipitation detection. Since MWHS2 has both temperature and humidity detection channels, it is also known as the microwave temperature - humidity detector MWTHS. Table 1.1.2 gives the instrument channel information of MWHS2.

[0087] Table 1.1.2 MWHS2 Channel Parameter Design

[0088]

[0089] In this invention, FY - 3D satellite observation data is downloaded from the China Meteorological Satellite Center. Among them, the observation data of MWTS2 and MWHS2 are Level 1 data encapsulated in HDF5 format. Each MWTS2 observation file contains about 100 minutes of observation information, with a total of about 1186 scanning lines. Each MWHS2 observation file contains about 100 minutes of observation information, with a total of about 2280 scanning lines. FY - 3D passes over the Chinese region at 06:00 and 18:00 every day.

[0090] 1.2 WRF Model and WRFDA Assimilation System

[0091] In this invention, the WRF4.1 version and its assimilation system are used to directly assimilate the MWTS2 and MWHS2 observation data of FY - 3D satellite. The observation operator of the satellite observation data used for direct assimilation is the RTTOV v12 fast radiative transfer model.

[0092] WRF is widely used in the field of mesoscale numerical weather prediction. The WRF model is a fully compressible non - hydrostatic model, and it is a model system that simultaneously includes data assimilation, atmospheric simulation, and numerical weather prediction, which can improve the simulation and prediction of mesoscale weather. The WRF model system framework mainly includes three parts: pre - processing, the WRF basic software library, and post - processing. Among them, the WRF basic software library is the main part, including dynamic solution schemes, initialization modules, data assimilation modules, and physical processes, etc. The specific system structure flow chart is as Figure 2 shown.

[0093] WRFDA is the assimilation system of the WRF model. The WRFDA variational assimilation system uses incremental assimilation technology and adopts the conjugate gradient method for minimization operations. In the WRFDA system, the functions of three-dimensional variational assimilation (3Dvar) and four-dimensional variational assimilation (4Dvar) have been implemented. The WRFDA assimilation process first conducts assimilation analysis on the Arakawa A grid, interpolates the analysis increments to the Arakawa C grid for calculation, and adds them to the background field to obtain the assimilated analysis field. Its assimilation process is as Figure 3 shown.

[0094] Figure 3 The input and output data in the WRFDA system are described in the circle, where x b is the background field, which can be generated by the real program or use the forecast field of the WRF model as the background field; y o is the input observation data, including conventional observation data and unconventional observation data represented by satellite observation data. The readable data formats include ASCII format or PRERBUFR format files. For observation data that cannot be directly read, preprocessing of the observation data is required; B o is the background deviation file, R is the observation deviation, and the correctness of the background and observation deviation files directly affects the accuracy of the assimilation results; x a is the analysis field, which is the result obtained by the WRFDA assimilation system assimilating the background field and the analysis field; x lbx is the side boundary. After the WRFDA obtains the assimilated analysis field, the side boundary can be updated. The analysis field x a can be used as the initial field of the model forecast, and can be added to the updated side boundary field and connected to the WRF model for model forecasting.

[0095] 2 Quality control and bias correction

[0096] Due to the influence of non-meteorological factors such as the observation position and observation technology of the satellite-borne atmospheric vertical sounder, the quality of the observation data will decline. Therefore, before using these data, we first need to conduct quality control on the observation data, eliminate the wrong data through objective analysis and processing, and retain the reasonable observation data. In addition, due to the systematic bias in the brightness temperature of each channel of the microwave vertical sounder, it is also necessary to correct the bias of the observed brightness temperature before assimilating the satellite radiation data. The satellite data quality control and bias correction process is as Figure 4 shown.

[0097] 2.1 Quality control

[0098] The quality control of the satellite-borne microwave vertical sounder data mainly has the following steps:

[0099] (1) Coarse inspection

[0100] Read the longitude, latitude and time of the observed pixels in the satellite data, reject the observations outside the assimilation time window and outside the model area, and perform preliminary quality control.

[0101] (2) Extreme value detection

[0102] Check whether the observed brightness temperature of all satellite pixels exceeds the maximum and minimum thresholds, and reject obvious outliers.

[0103] (3) Land surface type detection

[0104] Judge the land surface type according to the sea-land mask in the satellite data, and reject the observations of all mixed land surfaces.

[0105] (4) Limb detection

[0106] The scanning method of the microwave vertical sounder on the satellite results in the angle dependence of its brightness temperature data, that is, the limb characteristic. Especially when the observation angle is large, the limb effect is more significant. Therefore, reject the observation data of the first 5 and last 5 scanning angles of each scan line.

[0107] (5) Water droplet detection

[0108] In the judgment of cloud and rain areas, the total cloud water (cloudliquidwaterpath, CLWP) is selected as the detection index. When the total cloud water is greater than 0.2, it is considered that the observation is affected by cloud and rain, and reject the observation point.

[0109] (6) Cloud detection

[0110] Perform field of view angle matching based on the cloud amount product of the Medium Resolution Spectral Imager (MERSI) on the same platform. If the average cloud amount in the instantaneous field of view of the microwave vertical sounder is greater than 70%, then judge that the view point is a cloudy sky.

[0111] (7) Absolute deviation detection

[0112] Reject the observations where O - B is greater than 15K.

[0113] (8) Relative deviation detection

[0114] Reject the observations greater than three times the standard deviation.

[0115] 2.2 Bias correction

[0116] 2.2.1 Bias correction method

[0117] Bias correction is a key link in satellite assimilation applications. There will be certain biases between satellite observations and simulated radiation values based on model background field profiles. These biases contain multi-source errors such as instrument sensitivity, sensor response characteristics, calibration, and radiation transfer models. These errors are often comparable to the radiation changes corresponding to typical errors in numerical weather prediction temperature and humidity fields. Unless the biases are controlled and corrected below this level, it will be very difficult to apply satellite-measured radiation values to numerical weather prediction and obtain positive effects. Therefore, bias correction needs to be carried out before assimilating satellite radiance data to eliminate the systematic bias (O - B) and make it satisfy the normal distribution with a mean of zero. Currently, there are mainly two methods for bias correction of satellite radiance data used in research and operations: offline bias correction and variational adaptive bias correction. The common point of these two methods is to correct the observations based on the difference between the observed values and the background field simulated values. The present invention selects variational adaptive bias correction (VARBC) and conducts experiments on the adjustment of the scheme application.

[0118] The VARBC module of WRFDA needs to input the coefficients of the pre-statistical bias correction factors as initial values. The correction factor coefficients of all instruments are written in the VARBC.in file for the VARBC module to call and read. In order to provide accurate bias correction coefficients for the assimilation of satellite data at the initial time of the forecast, the present invention runs WRFDA to cyclically update the bias correction coefficients according to the cyclic update method.

[0119] FY-3D passes over the Chinese region at 06:00 and 18:00 every day. Therefore, the satellite observation data at 06:00 and 18:00 on each day from May 1st to 28th, 2019 are selected for assimilation calculation of the correction coefficients, totaling 56 time steps. The specific steps of the cyclic update process are as follows:

[0120] 1) At the initial time t0, the variational bias correction forecast factor coefficients come from the same type of microwave sounder.

[0121] 2) Run WRFDA at this initial time to obtain the updated bias correction coefficients;

[0122] 3) Use the coefficients obtained in step 2) as the initial coefficients for the next time t0 + 12;

[0123] 4) Run WRFDA at t0 + 12 to obtain the updated bias correction coefficients;

[0124] 5) Use the coefficients obtained in step 4) as the initial coefficients for the next time t0 + 24;

[0125] 6) Repeat the process of 4) - 5) at the next 54 assimilation times.

[0126] The latitude and longitude of the regional center are set to 23°N, 110°E. The model operation is set to a single-layer grid with a horizontal grid resolution of 9km, a grid number of 649*500, a vertical layer of 51 layers, and a model forecast integration step of 30s. The background field is interpolated from GFS data with a resolution of 0.25°. The thinning distance of MWTS2 and MWHS2 is 60km.

[0127] 2.2.2 Effect of Bias Correction

[0128] Figure 5 and Figure 6 Indicates the change of the bias correction coefficients of each assimilation channel of MWTS2 and MWHS2 with the cyclic assimilation time. During the cyclic update process, the global offset coefficients of MWTS2's 5 and 8 channels show an increasing trend, and the global offset coefficients of other channels do not change much, but there are still slow changes. The situation of MWHS2 is similar, and the global offset coefficient changes slowly. There are also certain changes in other bias correction coefficients, but their magnitude is much smaller, and the amplitude of the change of each bias correction coefficient of MWHS2 is greater than that of MWTS2. This may be because the change of water vapor in the atmosphere is more complicated, and the detection accuracy of water vapor channels is low, resulting in slow iterative convergence during minimization calculation. It can be seen that the change amplitude of most coefficients has been moderated after more than 20 days of update, but some coefficients still have large changes in the later stage of the update, and their change trend is not monotonic, and the cycle has not reached convergence. This phenomenon mainly occurs because the distribution of satellite data in the region at different times is different, resulting in large differences in the bias correction coefficients at different times.

[0129] Figure 7 The average deviations of each channel of MWTS2 and MWHS2 when using and not using the bias correction coefficients are shown in the data from the assimilation result diagnostic file at 18:00 on May 28, 2019. Before the bias correction, most of the channels of the two sensors had large systematic deviations. After the bias correction was performed using the updated bias correction coefficients, the deviations of each channel were greatly reduced compared with before. For MWTS2, the average deviations of basically all channels were close to 0; for MWHS2, the average deviations of channels 11 and 12 were close to 0, while the deviations of channels 13 and 15 were slightly less than 0. This may be due to the fact that the cloud detection scheme of MWHS2 is not strict enough, and the scattering effect of water droplets and ice crystals in the residual clouds leads to low observed brightness temperatures. The results show that the use of cyclically updated bias correction coefficients can achieve a good correction of the observational deviations of MWTS2 and MWHS2.

[0130] Figure 8It represents the standard deviation of the biases of each channel of MWTS2 and MWHS2 with and without using the bias correction coefficient. For MWTS2, the bias correction effectively reduces the standard deviation of the observation bias. Especially for channels 5, 6, and 7, the standard deviation of the bias of each channel of MWTS2 is reduced to less than 0.5 K. For MWHS2, the bias correction also reduces the standard deviation of the observation bias, but the reduction amplitude is smaller than that of MWTS2. The standard deviation of the bias of channels 11, 12, and 13 is reduced to less than 2 K, and the standard deviation of the bias of channel 15 is reduced from 3.6 K to 2.9 K. The results show that the cyclically updated bias correction coefficient can significantly reduce the standard deviation of the observation bias.

[0131] 3 Retrieval of Temperature and Humidity Profiles

[0132] 3.1 Establishment of the Retrieval Sample Library for Atmospheric Temperature and Humidity Profiles

[0133] The technical route for establishing the retrieval sample library for atmospheric temperature and humidity profiles is as Figure 9 shown.

[0134] 3.1.1 Temporal and Spatial Matching

[0135] Download ERA5 reanalysis data and MWTS2 and MWHS2 observation data, and perform temporal and spatial matching based on longitude, latitude, and time information. Select the profile information with a longitude and latitude difference of less than 0.1 degree and a time interval of less than 1 hour and the satellite observation information as sample pairs. The profile information comes from the surface and pressure level data of ERA5, and the satellite observation information comes from MWTS2 and MWHS2 data. The specific information included in each sample pair is shown in Table 3.1.1.

[0136] Table 3.1.1 Information of Sample Pairs

[0137]

[0138] 3.1.2 Cloud Detection

[0139] Since the present invention uses cloud area retrieval and assimilation, it is necessary to select cloud sky samples for the training and verification of the network model. At the same time, clear sky samples are also used for modeling as a control to compare the differences in the retrieval accuracy of the atmospheric temperature and humidity profiles by the microwave vertical sounder under clear sky and cloudy conditions.

[0140] Classify the selected sample pairs based on the MERSI cloud amount segment product of the same platform (FY-3D). Match the field of view angles of the microwave imager MERSI (slave instrument) and the microwave vertical sounder MWTS2 / MWHS2 (master instrument) of the same satellite. If the observation vector of the slave instrument is less than 1 / 2 of the instantaneous field of view angle of the master instrument, it is considered that the observation point of the slave instrument falls within the instantaneous field of view of the master instrument. The matching algorithm for the master and slave instruments is as follows:

[0141] First, convert the observation vector LOS from the satellite to the pixel into the local rectangular coordinate system (ENU):

[0142]

[0143] In the formula, Θ is the instrument zenith angle, Ф is the azimuth angle, and R is the distance from the instrument to the observed pixel.

[0144] After that, convert the LOS in the ENU coordinate system into the Earth-centered Earth-fixed coordinate system (ECEF):

[0145]

[0146] In the formula, λ represents the pixel longitude, and ψ represents the pixel latitude.

[0147] Calculate the observation vectors LOSm of the main instrument pixel and LOSf of the slave instrument pixel in the ECEF coordinate system respectively, and then calculate the cosine value of the included angle between the two vectors. If it is greater than the threshold, it means that the included angle is less than half of the main instrument's field of view The slave instrument pixel falls within the viewing angle of the main instrument, and the matching is successful (because high precision is required, the angle is not directly calculated using the inverse cosine).

[0148]

[0149] Finally, classify the observation samples of MWTS2 and MWHS2 based on the cloud amount segment products of the microwave imager within the instantaneous field of view and the surface types in the reanalysis data, which are divided into four categories: ocean clear sky, land clear sky, ocean cloudy sky, and land cloudy sky. Table 3.1.2 shows the categories and sample numbers of the sample library. Since the discrimination criterion for clear sky and cloudy sky is that the average value of the cloud product within the instantaneous field of view is greater than 70 (the maximum is 100), the number of cloudy sky samples is more than that of clear sky samples.

[0150] Table 3.1.2 Information of the Sample Library

[0151]

[0152] 3.2 Development of the Atmospheric Temperature and Humidity Profile Inversion Model

[0153] Combined with the radiation transfer theory and model, carry out the research on the inversion of the microwave vertical sounder data. Based on the BP artificial neural network method for inverting the atmospheric temperature and humidity profiles, develop an atmospheric temperature and humidity profile inversion model.

[0154] The algorithm flow of the BP artificial neural network is as Figure 10As shown in the figure. A neural network takes a series of inputs, weights them through connection weights and sends them to the hidden layer. After each neuron in the hidden layer sums up all the inputs, it generates a certain response output through a transfer function and outputs it to the output layer through the connection weights of the next layer. After various neurons in the output layer sum up all the inputs, they generate another response output. Then, its output is compared with the expected output. If the two tend to be the same or the difference is very small, it can be considered that this network has basically learned this problem. If the difference is relatively large or not satisfactory, the error between the network output and the expected output is sent back, and repeated training and learning are carried out by adjusting each connection weight. This cycle continues until it can generate an output result that approximates the true answer. Compared with traditional mathematical statistics methods, the BP artificial neural network can solve non-linear problems, and the requirement for the sample size can also be much less (if the sample can represent various types of characteristics of this problem), and it does not require the samples to be independent or follow a normal distribution. It also has strong fault tolerance, that is, the discrimination accuracy of the network system is generally not affected by the noise in the samples.

[0155] The three-layer network adopted in the present invention includes an input layer, a hidden layer and an output layer. The BP artificial neural network can be effectively used for the approximation of complex non-linear functions. A three-layer feedforward network can realize continuous function mapping with arbitrary precision. The BP artificial neural network model is shown in Figure 11 as shown.

[0156] The output of the neural network model can be described as:

[0157] O = f2(Yv + b2)

[0158] Y = f1(Xw + b1)

[0159] Among them, O represents the output of the network; x is the input from the input layer to the hidden layer and is also the input of the hidden layer; v is the connection weight coefficient matrix from the hidden layer to the output layer, which is a set of random numbers at the initial moment; w is the connection weight coefficient matrix from the input layer to the hidden layer, which is a set of random numbers at the initial moment; b1 and b2 are the bias value matrices of the hidden layer and the output layer units respectively; f is the non-linear action function of the neuron, which can be set according to each layer. The performance index of the BP algorithm for the multi-layer network is the mean square error, that is, MSE. For each input sample, the network output is compared with the target output, and the algorithm will adjust the network parameters to minimize the mean square error:

[0160] MSE = E[e 2 = E[(t - o) 2

[0161] In the formula, t is the expected output, o is the actual output, and e is the absolute error.

[0162] The parameter settings of the network in the present invention are as follows: ​

[0163] (1) Input layer setting: The brightness temperature observed by the satellite depends not only on the distribution of atmospheric elements but also on the observation deviation of the instrument itself. The observation deviation of the microwave vertical sounder depends on the scan angle and the latitude of the observation point. Therefore, the input layer of the network is the brightness temperature data of the microwave vertical sounder, the satellite zenith angle, the latitude of the observation point, and the position of the scan point.

[0164] (2) Output layer setting: The output layer is the temperature, specific humidity, and relative humidity profiles at 37 altitude levels.

[0165] (3) Setting of the number of hidden nodes: The performance of the neural network is easily affected by the setting of the number of nodes in its hidden layer. If the number of hidden nodes is too small, it will lead to insufficient information, thus affecting the inversion accuracy of the entire network. While if the number of hidden nodes is too large, it will result in too long training time and affect work efficiency. There is no good conclusion on how to determine the appropriate number of neurons in the hidden layer to maximize the generalization ability of the model. Currently, the approximate value range of the hidden nodes can be estimated by the following several formulas:

[0166]

[0167] h = log2 n

[0168]

[0169] Among them, h is the number of nodes in the hidden layer, m is the number of nodes in the input layer, n is the number of nodes in the output layer, where h is the number of hidden nodes, n is the number of input nodes, m is the number of output nodes, and a is a constant between 1 and 10.

[0170] (4) Transfer function setting: In the inversion network, the hyperbolic tangent S-shaped transfer function tansig is selected between the input layer and the hidden layer, and between the hidden layer and the output layer. This function can better express the nonlinear relationship between nodes and is not affected by the size of the input value. Its output value is bounded between -1 and 1.

[0171] (5) Training algorithm setting: Considering the large training samples of the network, the large number of network parameter settings, and the storage factor, the training algorithm of the network selects the Scaled conjugate gradient method trainscg. This algorithm is applicable to problems such as function fitting and pattern classification, converges relatively fast, has stable performance, and is especially suitable for the case of a relatively large network scale.

[0172] 3.3 Atmospheric temperature and humidity profile inversion experiment

[0173] Based on the developed model, an atmospheric temperature and humidity profile inversion experiment is carried out. The experimental process is as follows Figure 12As shown. In the study, satellite observation data were divided into land and ocean parts and tested separately. Taking the brightness temperature of land satellite observations as an example, the quality-controlled samples were put into the trained neural network to retrieve the atmospheric temperature profile and humidity profile, and their accuracy was evaluated.

[0174] To quantitatively analyze the advantages and disadvantages of the network, in addition to MSE, the correlation coefficient R, root mean square error RMSE, and mean error ME between the retrieved value and the actual value need to be considered. Their formulas are as follows:

[0175]

[0176] In the formula, n represents the number of sample profiles; x i is the value of the atmospheric temperature and humidity profile retrieved by the BP artificial neural network; y i is the actual value of the atmospheric temperature and humidity profile.

[0177] 3.3.1 Retrieval Results in the Ocean Region

[0178] Figure 13 are the retrieval results of temperature and humidity in the ocean region. It can be seen from Figure 13 that under clear-sky conditions, the maximum value of the retrieval RMSE of the atmospheric troposphere (100 - 1000 hPa) temperature is near 950 hPa, reaching 1.2 K, and then gradually decreases with the increase in altitude; in the atmospheric stratosphere (1 - 100 hPa), RMSE has a peak of 1.8 K at 70 hPa and shows an increasing trend near the top of the atmosphere. The retrieval RMSE of cloud region temperature is basically the same as that of the clear-sky region, gradually decreasing with the increase in altitude in the atmospheric troposphere and gradually increasing with the increase in altitude in the atmospheric stratosphere. This shows that cloud cover has little impact on the accuracy of temperature retrieval in the ocean region.

[0179] The retrieval RMSE of relative humidity first gradually increases with the increase in altitude, remains at about 10% in the range of 300 - 900 hPa, and then gradually increases with the increase in altitude again, reaching a maximum value of 19% at 100 hPa. This is because the peak values of the humidity detection channel weight functions of MWHS2 are all below 300 hPa and are not sensitive to the upper atmosphere. Above the atmospheric stratosphere, RMSE drops rapidly because the water vapor content in the atmosphere is already very low at this altitude. The trend of the retrieval RMSE of relative humidity in the cloud region is basically the same as that under clear-sky conditions, but it gradually increases from 10% to 17% with the increase in altitude in the range of 300 - 900 hPa and reaches a maximum value of 22% at 150 hPa. This shows that cloud cover has a certain impact on the accuracy of relative humidity retrieval in the ocean region.

[0180] Figure 14 are the retrieval results of temperature and humidity in the land region. It can be seen from Figure 14It can be seen that under clear-sky conditions, the maximum RMSE of the retrieved temperature in the atmospheric troposphere (100 - 1000 hPa) is located at the bottom layer of the atmosphere, reaching 2.5 K, and then gradually decreases with the increase in altitude. This is mainly due to the complex land topography and surface types, resulting in a large variation in emissivity. Starting from 400 hPa, it gradually increases with the increase in altitude and has a peak value of 1.7 K at 200 hPa. The trend of the retrieved RMSE of the cloud area temperature is basically the same as that of the clear-sky area. The difference is that the error increases within 850 - 1000 hPa, and there are two maximum values of 1.6 K in the upper atmosphere, located at 70 hPa and 200 hPa respectively. This shows that cloud cover has a certain impact on temperature retrieval in the bottom and upper layers of the atmosphere in the land area, but has little impact on the middle layer of the atmosphere.

[0181] The retrieved RMSE of relative humidity gradually increases with the increase in altitude in the troposphere and gradually decreases with the increase in altitude in the stratosphere. The maximum value is located at 200 hPa, reaching 19%. The trend of the retrieved RMSE of relative humidity in the cloud area is basically the same as that under clear-sky conditions, but the accuracy at each altitude layer is lower than that under clear-sky conditions, and the maximum value exceeds 23%. This shows that cloud cover has a certain impact on the accuracy of relative humidity retrieval in the land area.

[0182] Comparing the retrieved results of the ocean area and the land area, it can be obtained that the retrieval accuracy of temperature and humidity in the ocean area is better than that in the land area; cloud cover has little impact on the retrieval accuracy of temperature, but has a certain impact on specific humidity and relative humidity, and the impact degree of cloud cover in the land area is greater; the maximum values of the retrieved RMSE of temperature and specific humidity are located in the surface layer, while the maximum value of the retrieved RMSE of relative humidity is located near 200 hPa.

[0183] 3.4 Quality control of retrieved results

[0184] Considering the retrieval accuracy of the atmospheric temperature and humidity profiles, before converting the retrieved results into conventional data, it is necessary to perform quality control on the retrieved results to eliminate observations with low quality and retrieved results with poor accuracy. Table 3.4.1 gives the quality control scheme for the retrieved results. Similar to clear-sky assimilation, the retrieved results on both sides of each scan line are eliminated due to the limb effect, and the retrieved results of the mixed surface are eliminated due to the difference in emissivity between the ocean and the road surface. Finally, the retrieved results of the altitude layers with large root mean square errors are eliminated according to the retrieval accuracy of each altitude layer.

[0185] Table 3.4.1 Quality control of retrieved results

[0186]

[0187] 4 Construction of an all-weather assimilation system for microwave temperature and humidity sounders

[0188] 4.1 Expansion of the WRFDA system interface

[0189] The WRF model and WRFDA system developed by NCAR and other institutions have been able to assimilate data from a variety of satellite instruments, and the variational assimilation system is relatively mature. The system is easy to understand, portable, and highly interoperable. The modular design makes it possible to write other satellite data into the assimilation module, and it is relatively easy. Therefore, the present invention relies on the WRFDA system framework, and is modified on the basis of retaining the core computing programs such as the algorithm and main program of the system, and writes an assimilation program module specifically for MWTS2 and MWHS2, so that it can read MWTS2 and MWHS2 data, and use the assimilation system to assimilate satellite observation data, thereby changing the initial guess field and affecting the forecast results of the forecast module. In order to successfully apply the data to numerical weather forecasting, the present invention, based on the assimilation system, constructs a system suitable for data quality control, deviation correction and assimilation for the FY-3D polar-orbiting satellite microwave temperature and humidity sounder data.

[0190] The idea of ​​implementing the assimilation module is: first, add the definition of the variables required for assimilation data in the system, then read in the radiance data, and after quality control and deviation correction, input the observation increment into the minimization module to complete the calculation.

[0191] The main modified programs and added files are divided into the following parts:

[0192] Information initialization: Initialize the radiance data information in the satellite data assimilation initialization module da_setup_radiance_structures, and also initialize the RTTOV interface variables.

[0193] Data reading program writing: Add da_read_obs_hdf5mwts2.inc and da_read_obs_hdf5mwhs2.inc in the da_radiance directory. The key is to know the ID of the mwts2 and mwhs2 sensors in RTTOV mode. In the reading program, the data is preliminarily quality controlled, and the observation data that is not in the assimilation area and assimilation time range is eliminated.

[0194] Data quality control program: add da_qc_mwts2.inc and da_qc_mwhs2.inc in the da_radiance directory, and design a quality control program suitable for the data according to the characteristics of the observation data of each channel

[0195] Write sensor parameter files: Also add fy3-4-mwht2.info and fy3-4-mwhs2.info, which are files about sensor channel characteristics.

[0196] The current operating environment of the system is:

[0197] For the LINUX system and Intel compiler (version 18.0.0), it is necessary to install HDF5 (version 1.8.20) and NETCDF library (version 4.5.0) in advance.

[0198] The extended WRFDA system can realize the functions of reading, quality control, bias correction and assimilation for FY3D MWTS2 and MWHS2.

[0199] 4.2 bufr format conversion

[0200] The conventional observation data released by NCEP is in the PREPBUFR format, which includes sounding, surface, ship, buoy and other observation data. In order to make the inversion results enter the assimilation system, the present invention uses the BUFRLIB library to convert the retrieved atmospheric temperature and humidity profiles into the PREPBUFR format.

[0201] The retrieved atmospheric temperature and humidity profiles will be imported into txt in advance, and then prepbufr_append_sound.exe will be compiled and run to add the atmospheric temperature and humidity profiles to the existing PREPBUFR file in the form of conventional observations.

[0202] 4.3 All-sky assimilation process

[0203] The all-sky assimilation process of microwave temperature and humidity sounding data is as Figure 15 shown. The first-level data of MWTS2 and MWHS2 radiances enter the quality control module after data reading, and the assimilation system is started for the first time. In the quality control module, the MERSI cloud amount product matched with the observation point is used as the criterion for distinguishing clear sky and non-clear sky. The observation data that pass the quality control but are judged as non-clear sky are output, and the atmospheric temperature and humidity profiles are retrieved according to the observed brightness temperature. The atmospheric temperature and humidity profiles are converted into the prebufr format in the form of sounding data. Finally, the assimilation system is started for the second time, and the non-clear sky profile information converted into the prebufr format and the microwave brightness temperature data in the clear sky area enter the WRFDA assimilation system together.

[0204] As the second-generation polar-orbiting meteorological satellites in China, the newly launched FY-3D satellite is equipped with a new microwave thermometer and a new microwave humidity meter, which are more advanced than the microwave sounders carried on FY-3A / B. The number of detection channels has increased and the detection accuracy is higher. In order to break through the key points and difficulties of current satellite data operational assimilation, conduct all-weather data assimilation research on FY-3D microwave data, further explore the application potential of cloud area data, it is of great significance to improve the utilization rate of satellite observation data and the forecasting ability of disaster weather processes, reduce the casualties and economic losses caused by meteorological disasters, and has great application prospects.

[0205] At present, the mainstream method for assimilating all-weather satellite microwave observation data is direct assimilation. Directly assimilating brightness temperature makes a physically more continuous adjustment to the model dynamic field and avoids the problem of errors in inversion products. However, due to the relatively late start of related research, it is not yet systematic. Compared with direct assimilation, the advantage of indirect assimilation is that it can better control the non-linear response of the variational assimilation system to changes in the atmospheric state, obtain better quality control before finally entering variational assimilation, and can assist in defining background and observation errors in assimilation. Therefore, based on FY-3D microwave thermometer and microwave hygrometer data, this invention uses an artificial neural network algorithm to invert the three-dimensional temperature and humidity fields of the atmosphere in cloud and rain areas, conducts assimilation of inversion products, and realizes assimilation of all-weather microwave atmospheric vertical sounder data by integrating the direct assimilation method of clear-sky emissivity data, effectively improving the accuracy of numerical weather prediction.

[0206] 5 Verification of all-weather assimilation effect

[0207] Based on the developed data assimilation system, this invention conducts batch experiments on data assimilation in combination with the WRF model to evaluate the improvement of all-weather assimilation of FY-3D microwave thermometer and microwave hygrometer on numerical prediction results.

[0208] 5.1 Parameter design and assimilation scheme

[0209] The GFS forecast results of NCAR are used in the experiment to provide initial fields and boundary conditions. The time range of the experiment is from June 1st to 28th, 2019. Since the FY-3D satellite passes over the Chinese region at 06:00 and 18:00 every day, the assimilation time is set at 06:00 and 18:00 every day. The experiment first starts with a 6-hour integration from 00:00 and 12:00 every day as a cold start, uses the 6-hour integration result as the first guess field for assimilation, and then uses the assimilated analysis field as the initial field for a 24-hour forecast.

[0210] The horizontal resolution of the model grid is 9 km, the number of grid points is 649 * 500, and the regional center longitude and latitude are 23°N, 110°E. The vertical stratification is 51 layers, the model top pressure is 10 hPa, and the model forecast integration time step is 30 s. The physical parameterization scheme adopted in the research is shown in Table 5.1.1. The assimilation time window for both groups of experiments is one hour before and after, and CV7 is used as the background error. The sparsification distance of MWTS2 and MWHS2 is both 60 km.

[0211] Table 5.1.1 Parameterization scheme settings

[0212] Parameterization scheme type Parameterization scheme Microphysics Thompson Cumulus convection parameterization NewTiedtke Shortwave radiation RRTMG Longwave radiation RRTMG Surface layer RevisedMM5Monin-Obukhov Land surface process unifiedNoahland-surface Planetary boundary layer YSU

[0213] To quantitatively evaluate and compare the impacts of assimilating two microwave sounding instruments, MWTS2 and MWHTS2, on numerical weather prediction, three groups of experiments were designed, as shown in Table 5.1.2. CTRL is the control experiment, which assimilates only conventional observational data; CLEAR is the clear-sky assimilation experiment, which adds the clear-sky observational data of MWTS2 and MWHS2 on the basis of assimilating conventional observational data; ALLSKY is the all-sky assimilation experiment, which adds the clear-sky observational data and cloud retrieval data of MWTS2 and MWHS2 on the basis of assimilating conventional observational data.

[0214] Table 5.1.2 Experimental Scheme Settings

[0215]

[0216] 5.2 Experimental Results

[0217] 5.2.1 Case Analysis of Precipitation Forecast

[0218] Taking an assimilation experiment at 06:00 on June 3, 2019 as an example, the assimilation effect of the FY3D microwave vertical sounder was analyzed.

[0219] 5.2.1.1 Observation Error Distribution

[0220] Figure 16 They are the probability distributions of OMB and OMA before and after bias correction for the 11th channel of MWHS2. As shown in the figure, the corrected observation residuals are closer to the Gaussian distribution with a mean of zero, and the bias is greatly reduced. Figure 17 They are the scatter plots of OMB and OMA for the 11th channel of MWHS2. As shown in the figure, the pixel distribution is basically located on the main diagonal. After assimilation, the simulated brightness temperature of the analysis field is closer to the satellite observation than the background field. The root mean square error is reduced from 2.209 to 1.473, and the correlation coefficient is increased from 0.863 to 0.939.

[0221] The above results indicate that variational quality control can greatly reduce the observation residuals, and the data assimilation of the FY3D microwave vertical sounder can significantly improve the initial field of the WRF model.

[0222] 5.2.1.2 Cumulative Precipitation Distribution

[0223] The experiment gave the 24-hour cumulative precipitation from 06:00 on June 3, 2019 to 06:00 on June 4, 2019 for the three groups of experiments. The results of all three groups of experiments indicated that there were moderate to heavy rains from the eastern part of Fujian to the southern part of Guangzhou. However, compared with the CTRL experiment, both the CLEAR and ALLSKY experiments predicted precipitation above the 100 mm level, and the predicted area of the ALLSKY experiment was larger; the predicted results of the CLEAR and ALLSKY experiments had a smaller precipitation range of the 50 mm level in the central part of Guangzhou compared with the CTRL experiment.

[0224] 5.2.2 Verification of the Effect of Batch Tests

[0225] 5.2.2.1 Analysis of the Field Effect

[0226] Taking the FNL reanalysis data of NCEP as the standard, the statistical analysis of field errors is carried out. Figure 18 The variation of the mean deviation ME of temperature and geopotential height with altitude levels for the CTRL, CLEAR, and ALLSKY groups of tests from June 1 to 28, 2019 is given. From Figure 18 it can be seen that the trends of temperature ME in the CTRL and CLEAR tests are basically the same with altitude, while the ALLSKY test is significantly closer to 0 than the CTRL and CLEAR tests at almost all altitudes. The influence of assimilating MWTS2 and MWHS2 on geopotential height is similar to that of temperature. The above results show that the clear-sky assimilation of MWTS2 and MWHS2 data slightly improves the analysis field, mainly in the upper-middle atmosphere, while the influence of assimilating cloud area retrieval data on the analysis field is greater, making the error distribution more unbiased (ME close to 0).

[0227] 5.2.2.2 Precipitation Forecasting Effect

[0228] The forecast field is matched with precipitation stations, and taking the ground precipitation observation data as the true value, the numbers of TP, TN, FP, and FN of the 24-hour accumulated precipitation in the batch tests in June 2019 are counted, and the TS and ETS scores are calculated. Figure 19 and Figure 20 The comparison of the TS score and the ETS score is given; Tables 5.2.1 and 5.2.2 give the specific TS scores and ETS scores.

[0229] Compared with the CTRL test, the TS score of the CLEAR test has positive effects at the 25, 50, and 100 mm thresholds, with an increase of 1.57% at 25 mm, 3.56% at 50 mm, and 8.87% at 100 mm; the ETS score of the FY3D test also has positive effects at the 25, 50, and 100 mm thresholds, with an increase of 1.75% at 25 mm, 3.72% at 50 mm, and 8.19% at 100 mm. The above batch test results show that the clear-sky assimilation of FY3D microwave temperature and humidity sounder data significantly improves the precipitation forecast for heavy rain and above magnitudes.

[0230] Compared with the CLEAR experiment, the positive effect of the TS score in the ALLSKY experiment is further improved at the 50 and 100 mm thresholds, with a 5.93% increase at 50 mm and a 39.25% increase at 100 mm; the ETS score in the ALLSKY experiment is also further improved at the 50 and 100 mm thresholds, with a 6.37% increase at 50 mm and a 41.66% increase at 100 mm. The above batch test results show that the full-sky assimilation of FY3D microwave temperature and humidity profile data has an obvious improvement effect on precipitation forecasts of heavy rain and above.

[0231] Table 5.2.1 Precipitation TS Score

[0232]

[0233]

[0234] Table 5.2.2 Precipitation ETS Score

[0235]

[0236] To more objectively evaluate the impact of assimilating MWTS2 / MWHS2 data on the simulation effect of precipitation, the 24-hour period is divided into four time periods at 6-hour intervals, namely 0-6, 6-12, 12-18, and 18-24 hours. Figure 21 The TS scores statistically at 6-hour intervals are given, and the percentage increase in the TS scores of the CLEAR experiment and the ALLSKY experiment compared to the CTRL experiment is written above the bar chart.

[0237] By comparing the changes in the TS scores at 4 different thresholds over time periods, it can be seen that: in the 0-6h time period, the CLEAR experiment has an increase at the 1, 10, and 50 mm thresholds, and the ALLSKY experiment also has an increase at the 50 mm threshold and the increase amplitude is larger than that of the CLEAR experiment; in the 6-12h time period, the CLEAR experiment has a slight decrease, while the ALLSKY experiment has an increase at the 10 and 25 mm thresholds; in the 12-18h time period, the CLEAR experiment has an increase at the 1, 10, and 50 mm thresholds, and the ALLSKY experiment has an increase at the 1, 10, 25, and 50 mm thresholds and the increase amplitude is larger than that of the CLEAR experiment; in the 18-24h time period, the CLEAR experiment has an increase at the 1, 10, and 25 mm thresholds, and the ALLSKY experiment has an increase at the 1, 10, and 25 mm thresholds but the increase amplitude is slightly smaller than that of the CLEAR experiment.

[0238] Generally speaking, the positive effects of the CLEAR experiment and the ALLSKY experiment are mainly reflected in the time periods of 12 - 18h and 18 - 24h. The improvement of precipitation with magnitudes above 25 and 50 mm is particularly obvious in the 12 - 18h period. This may be due to the system spin - up. After the system starts, the precipitation distribution in the 12 - 24h period will be more accurate. In the 0 - 6h period, the overall score of the CLEAR experiment is higher than that of the ALLSKY experiment, while in the 6 - 12h and 12 - 18h periods, the overall score of the ALLSKY experiment is higher than that of the CLEAR experiment.

[0239] Example Two:

[0240] This example provides an all - weather assimilation device for satellite microwave temperature and humidity sensors based on the inversion of cloud - area temperature and humidity profiles, including:

[0241] An acquisition module: used to acquire satellite data, conventional observation data, and forecast field data; interpolate the surface and atmospheric profile information of the forecast field to the location of satellite observations; the satellite data includes microwave thermometer and microwave humidity sensor data;

[0242] A clear - sky / non - clear - sky determination module: used to determine whether the observation point is a clear - sky or non - clear - sky based on the cloud cover data in the satellite data;

[0243] A conversion module: used to output the satellite data determined to be non - clear - sky, and invert the atmospheric temperature and humidity profiles based on the observed brightness temperature; convert the atmospheric temperature and humidity profiles into the prebufr format in the form of radiosonde data to obtain the non - clear - sky area profile information in the prebufr format;

[0244] A bias correction module: used to correct the bias of the satellite data determined to be clear - sky to obtain the corrected microwave brightness temperature data in the clear - sky area;

[0245] An output module: used to perform quality control on the corrected microwave brightness temperature data in the clear - sky area, and eliminate the observation data with low quality; put the non - clear - sky area profile information converted into the prebufr format and the microwave brightness temperature data in the clear - sky area into the WRFDA assimilation system for data assimilation, and output the assimilated analysis field.

[0246] The device of this example can be used to implement the method described in Example One.

[0247] Example Three:

[0248] The present invention provides an all - weather assimilation device for satellite microwave temperature and humidity sensors based on the inversion of cloud - area temperature and humidity profiles, including a processor and a storage medium;

[0249] The storage medium is used to store instructions;

[0250] The processor is configured to operate according to the instructions to perform the steps of the method described in the first embodiment.

[0251] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of an all-hardware embodiment, an all-software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) that contain computer-usable program code.

[0252] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one or more of the flows Figure 1 or blocks or combinations of blocks.

[0253] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in Figure 1 one or more of the flows Figure 1 or blocks or combinations of blocks.

[0254] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are performed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one or more of the flows Figure 1 or blocks or combinations of blocks.

[0255] The above description is only a preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the technical principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. An all-weather assimilation method for satellite microwave temperature and humidity sounders based on the inversion of cloud region temperature and humidity profiles, characterized in that, The method includes the following steps: Obtain satellite data, conventional observation data, and forecast field data; interpolate the surface and atmospheric profile information of the forecast field to the position where the satellite observations are located; the satellite data includes microwave thermometer and microwave hygrometer data; Based on the cloud cover data in the satellite data, determine whether the observation point is a clear sky or not; Output the satellite data determined to be non-clear sky, and retrieve the atmospheric temperature and humidity profiles from the observed brightness temperature; convert the atmospheric temperature and humidity profiles into the prebufr format in the form of sounding data to obtain the non-clear sky area profile information in the prebufr format; Perform bias correction on the satellite data determined to be clear sky to obtain the microwave brightness temperature data of the corrected clear sky area; Perform quality control on the microwave brightness temperature data of the corrected clear sky area to remove the observation data with lower quality; Put the non-clear sky area profile information converted into the prebufr format and the microwave brightness temperature data of the clear sky area after removing the observation data with lower quality into the WRFDA assimilation system for data assimilation, and output the assimilated analysis field; The method for performing quality control on the microwave brightness temperature data of the corrected clear sky area to remove the observation data with lower quality includes the following steps: Read the longitude, latitude, and time of the observed pixels in the satellite data, reject the observations outside the assimilation time window and outside the model area range, and perform preliminary quality control; Check whether the observed brightness temperature of all satellite pixels exceeds the maximum and minimum thresholds, and reject the obvious outliers; Judge the surface type according to the land-sea mask in the satellite data, and reject all observations of mixed surfaces; Reject the observation data of the first 5 and last 5 scan angles of each scan line of the satellite data; In the rainfall judgment, select the total cloud water as the detection index. When the total cloud water of a certain observation point is greater than 0.2, it is considered that the observation is affected by rainfall, and the observation point is rejected; Reject the observation data with an observation residual greater than 15 Kelvin in the satellite data; Reject the observation data with an observation residual greater than three times the error standard deviation in the satellite data.

2. The all-weather assimilation method according to claim 1, wherein The method for judging whether the corrected satellite data belongs to clear sky or non-clear sky includes: Perform field of view angle matching based on the cloud cover product of the Medium Resolution Spectral Imager (MERSI), and judge whether each view point is a cloudy sky. If the average value of the cloud cover product within the instantaneous field of view of the microwave vertical sounder is greater than 70%, then judge that the view point is a cloudy sky.

3. The all-weather assimilation method according to claim 1, characterized in that, The method for performing bias correction includes the following steps: Select the model thickness, surface temperature, cloud liquid water content, and scan position as the correction factors; Multiply the correction factors by the corresponding correction coefficients to obtain the bias correction amount; Subtract the bias correction amount from the original brightness temperature to obtain the corrected brightness temperature; The method for obtaining the correction coefficients includes the following steps: Step A: The correction coefficients at the initial time t0 are from other similar microwave vertical sounders; Step B: Run the WRFDA system at the initial time t0 to obtain updated correction coefficients; Step C: Use the updated correction coefficients as the correction coefficients at the next time t1; Step D: Run the WRFDA system at time t1 to obtain updated correction coefficients; Step E: Use the updated correction coefficients as the correction coefficients at the next time t2; Step F: Repeat the process of steps DE at the next assimilation time to cyclically update the correction coefficients.

4. The all-weather assimilation method according to claim 1, characterized in that, The method for inverting the atmospheric temperature and humidity profile includes the following steps: Obtain satellite observation information in the form of ASCII output after the first assimilation of the system; the satellite observation information includes the latitude and longitude, brightness temperature, cloud cover, observation angle, sea and land code, and surface type information of each observation point; The brightness temperature, latitude and observation angle are input into the atmospheric temperature and humidity profile inversion model to obtain the atmospheric temperature and humidity profile.

5. The all-weather assimilation method according to claim 4, wherein The method for constructing the atmospheric temperature and humidity profile inversion model comprises the following steps: Obtain satellite data and reanalysis data for time and space matching; classify the clear sky and cloud areas of profile samples based on cloud products and establish a sample data set; Based on the BP artificial neural network algorithm, develop an inversion model for atmospheric temperature and humidity profiles in ocean and land areas, as well as under clear and cloudy conditions; Atmospheric temperature and humidity profile inversion experiments are carried out for ocean and land areas, as well as clear sky and cloudy sky conditions, and the model inversion accuracy is verified.

6. An all-weather assimilation device for satellite microwave temperature and humidity sounder based on inversion of cloud area temperature and humidity profiles, characterized in that, include: Acquisition module: used to acquire satellite data, conventional observation data and forecast field data; interpolate the ground and atmospheric profile information of the forecast field to the location of satellite observation; the satellite data includes microwave thermometer and microwave hygrometer data; Sunny or rainy judgment module: used to judge whether the observation point is sunny or cloudy based on the cloud cover data in the satellite data; Conversion module: used to output satellite data identified as non-clear sky, and obtain atmospheric temperature and humidity profiles based on observed brightness temperature inversion; convert atmospheric temperature and humidity profiles into prebufr format in the form of sounding data, and obtain non-clear sky area profile information in prebufr format; Bias correction module: used to correct the bias of satellite data identified as clear sky and obtain the corrected microwave brightness temperature data of clear sky area; Output module: used to perform quality control on the corrected microwave brightness temperature data of the clear sky area and remove the observation data with low quality; put the non-clear sky area profile information converted into prebufr format and the microwave brightness temperature data of the clear sky area after removing the observation data with low quality into the WRFDA assimilation system for data assimilation, and output the assimilated analysis field; The method of quality control of the corrected microwave brightness temperature data in the clear sky area and eliminating the observation data with low quality includes the following steps: Read the latitude, longitude and time of the observation pixels in the satellite data, reject the observations outside the assimilation time window and the model area, and perform preliminary quality control; Check whether the observed brightness temperature of all satellite pixels exceeds the maximum and minimum thresholds, and reject obvious outliers; Determine the surface type based on the land and sea masks in the satellite data and reject all observations of mixed surfaces; The observation data of 5 scanning angles before and after each scanning line of the satellite data are rejected; In rainfall judgment, the total amount of cloud water is selected as the detection index. When the total amount of cloud water at a certain observation point is greater than 0.2, it is considered that the observation is affected by rainfall and the observation point is rejected. Reject satellite data with observational residuals greater than 15 Kelvin; Observations in satellite data with residuals greater than three times the standard deviation of the error are rejected.

7. An all-weather assimilation device for a satellite microwave temperature and humidity sounder based on the inversion of cloud region temperature and humidity profiles, characterized in that including processor and storage medium; The storage medium is used to store instructions; The processor is configured to operate in accordance with the instructions to perform the steps of the method recited in claims 1-5.

Images