A method for data assimilation of clear sky channel infrared hyperspectral data

Abstract

The application discloses a data assimilation method for clear-sky channel infrared hyperspectral data, and comprises the following steps: obtaining clear-sky channel infrared hyperspectral data; using a fast radiation transfer model to convert a variable from a model space to a linear or nonlinear relationship in an observation space; combining a principal component method to screen satellite infrared hyperspectral atmospheric vertical detection data channels; using the difference between an observed infrared observation brightness temperature and a background field simulation brightness temperature to find channels not affected by cloud pollution; eliminating statistical deviation between satellite instrument observation radiation values and radiation values simulated and calculated according to a background field profile; removing observation data with errors exceeding a preset threshold on a mixed ground surface type; and outputting data assimilation results of the clear-sky channel infrared hyperspectral data. The application improves the analysis field of each layer of the global atmosphere, effectively improves the numerical prediction level, significantly improves the error of typhoon prediction intensity, and is superior to a conventional observation data test in the path error before landing.

Description

Technical Field

[0001] The present invention belongs to the technical field of weather forecasting and numerical forecasting, and in particular relates to a data assimilation method for clear-sky channel infrared hyperspectral data. Background Art

[0002] Compared to conventional infrared radiance data (HIRS4), infrared hyperspectral data has over thousands of detection channels (HIRAS-II has 3041 channels), providing information on atmospheric temperature, humidity, and composition at higher vertical resolution. Because infrared radiance data is susceptible to cloud and rain contamination, existing infrared hyperspectral data often utilizes clear-sky field assimilation methods. However, cloud and rain areas are often sensitive areas that have a significant impact on forecasts. Therefore, building on the existing assimilation capabilities of international infrared hyperspectral data from IASI and AIRS, it is necessary to design a full-process autonomous numerical prediction method tailored to the characteristics of hyperspectral data from the domestic HIRAS-II satellite. This method, employing clear-sky channel assimilation, enables the direct assimilation and application of domestic infrared hyperspectral data. Summary of the Invention

[0003] In view of this, this application proposes a data assimilation method for clear-sky channel infrared hyperspectral data, comprising the following steps:

[0004] S1 Assimilation Preprocessing: Acquire clear-sky channel infrared hyperspectral data; Use the rapid radiative transfer model to transform variables from pattern space into linear or nonlinear relationships in observation space;

[0005] S2 channel selection: Combine the principal component method to screen satellite infrared hyperspectral atmospheric vertical detection data channels;

[0006] S3 Cloud Detection: Utilizes the difference between infrared observation brightness temperature and background field simulation brightness temperature to find channels that are not affected by cloud contamination;

[0007] S4 Bias Correction: Eliminate the statistical deviation between the radiation values ​​observed by satellite instruments and the radiation values ​​calculated based on the background field profile simulation;

[0008] S5 Data quality control: Eliminate observations with errors exceeding a preset threshold on mixed surface types;

[0009] S6 outputs the data assimilation results of the clear sky channel infrared hyperspectral data.

[0010] Furthermore, the rapid radiative transfer mode includes: interpolating the model profile to the observation position, then using the rapid radiative transfer mode forward operator to simulate the observation, obtaining the observation increment compared with the actual observation, and then converting the observation increment into an analysis increment through the rapid radiative transfer mode tangent and adjoint operators.

[0011] Furthermore, the rapid radiative transfer model is based on the RTTOV rapid radiative transfer model, adopts a high-precision spectral line model, introduces the infrared channel spectral response function, simulates radiative transfer based on the typical atmospheric profile distribution in different regions and seasons, designs a statistical regression model for atmospheric transmittance, generates a normalized rapid transmittance calculation lookup table for HIRAS-II, completes the rapid radiative transfer model required for the four-dimensional variational assimilation of satellite infrared spectral data, and completes the corresponding tangent line / adjoint model training. Specifically, the following steps are included:

[0012] Based on the atmospheric profile dataset and spectral library, the infrared hyperspectral transmittance dataset is generated using the line-by-line integration mode;

[0013] Using the established hyperspectral infrared representative atmospheric transmittance dataset, a satellite infrared hyperspectral clear sky fast forward operator was established based on the multivariate linear regression method.

[0014] Before the regression calculation, the transmittance of the channel is expressed as follows using the transmittance of the monochromatic absorbing gas:

[0015] ;

[0016] Where v is the channel center frequency, j is the mode layer, is the channel mixed gas transmittance from mode layer j to the external space with a center frequency of v, is the transmittance of the mixed gas containing water vapor in the range from mode layer j to the outer space for the channel with the center frequency v from mode layer j to the outer space, is the atmospheric transmittance of water vapor and ozone in the mixed gas from mode layer j to the outer space;

[0017] When the transmittance of any isobaric surface is less than the preset empirical constant, the transmittance is corrected using the water vapor content of the adjacent isobaric surface:

[0018] ;

[0019] Among them, constant constant=1.e-6, amount is the absorbed gas content, k is the number of pressure layers, is the mixed gas transmission rate containing water vapor in the kth layer, is the mixed gas transmission rate, is the water vapor content in the kth layer;

[0020] For a given atmospheric state and surface parameters, along the observation angle The propagation path of the fast radiation transfer mode simulates the channel transmittance from the corresponding pressure layer to the top of the atmosphere. Or the optical thickness of the channel from the corresponding pressure layer to the top of the atmosphere ,in , after convolution, the optical thickness of beam νi from pressure layer j to the top of the atmosphere is defined as:

[0021] ;

[0022] in, is the prediction factor that depends on the profile, M is the number of prediction factors, is the transmittance calculation coefficient; define the linear regression equation group:

[0023] ;

[0024] Used for regression calculation of transmittance coefficient .

[0025] Furthermore, the principal component method is used to screen the satellite infrared hyperspectral atmospheric vertical detection data channels. The steps are as follows:

[0026] Firstly, based on the channel selection method of the cumulative influence coefficient of the principal component, the importance of the original observation data channels of the infrared hyperspectral data is sorted offline. The original observation data of the infrared hyperspectral data is : ; , and then use the principal component method to map it to the new variable space : ; ,in:

[0027] , ;

[0028] in: : ; , The row vector is the original data The eigenvectors of the covariance matrix, u ij is the feature of the jth channel of the i-th pattern layer after mapping, n is the number of pattern layers, m is the number of channels, x ij is the original observation of the jth channel of the i-th mode layer, y ij is the observation of the jth channel of the i-th mode layer after mapping, X m is the original observation of channel m; if the coefficient satisfy , ; It also guarantees are unrelated, and yes The jth largest variance among all linear combinations of For the original data The jth principal component of

[0029] The principal component cumulative influence coefficients are calculated for the temperature, humidity, and window area detection channels respectively. Then, the combination of daytime and nighttime detection channels, as well as the influence of sunlight, are gradually considered to ultimately achieve the optimal confirmation of the hyperspectral channel.

[0030] The influence coefficients of the HIRAS channels of the four-dimensional variational assimilation system were ranked. Then, the channels were further screened based on the observation-simulation distribution characteristics of multiple HIRAS channels. Based on the channel ranking, the observation channels with observation-simulation mean > 0.5 and observation-simulation variance > 3.5 were further excluded. Finally, the HIRAS-II preferred channels suitable for the four-dimensional variational assimilation system were obtained.

[0031] Furthermore, for the temperature detection channel, after removing the channels affected by water vapor and those in the blacklist,

[0032] The calculation process of the principal component cumulative influence coefficient is based on the Jacobian matrix of the temperature detection channel. The process is as follows:

[0033] Step 1: Get the Jacobian matrix H(x) of the temperature detection channel. The variables of each channel are calculated according to the following formula: After the transformation, the standardized matrix is ​​obtained, which is recorded as: ;in are the elements of the normalized matrix, Indicates the number of mode layers, Indicates the number of channels, s ji is the variance of the j-th channel in layer i, is the mean of the i-th layer;

[0034] Step 2: Calculate the normalized matrix The covariance matrix of ;

[0035] Step 3: Calculation The eigenvalue of and eigenvectors, and the eigenvalues Arrange the values ​​from large to small;

[0036] Step 4: Perform principal component analysis and take the first p principal components; the principle of selecting the first p principal components is: given a preset percentage ε, select p to make , represents the percentage of the first p principal components in the total variance, that is, the total contribution rate of the p principal components, where represents the jth largest eigenvalue;

[0037] Step 5: Map it into a new vector, obtain the previous main components according to Step 4, and enter the analysis of cumulative influence coefficient;

[0038] Step 6: If the same channel is selected in different principal components, the influence coefficient s needs to be calculated, which is defined as the product of the eigenvector of the principal component and the variance contribution rate of each principal component, where the difference contribution rate is defined as the ratio of the corresponding eigenvalue of each principal component to the sum of the total eigenvalues, which is the variance contribution rate of the principal component; otherwise, the influence coefficient is calculated and finally the influence coefficients are sorted to select the channels whose influence on temperature is greater than the preset threshold. When the number of selected channels reaches the initially given number of channels, or the cumulative influence coefficient of a channel is less than the given threshold, the channel selection is stopped at this time and go to Step 4.

[0039] Furthermore, the bias correction includes at least one of scanning bias correction, air mass bias correction and variational bias correction;

[0040] The scanning deviation correction includes:

[0041] Introduces latitude-dependent scan bias corrections and o The Earth is divided into 18 latitude bands using meridians as a band; the average values ​​for each scan position and each latitude band are calculated and then used to calculate the scan correction factor ,Right now:

[0042] ;

[0043] in It is a latitude belt, is the average value of the scan positions, is the average value for the latitudinal band;

[0044] To avoid discontinuity in the scan bias correction between latitudinal bands, a smoothing technique is used to generate continuous correction coefficients in areas that span latitudinal bands. Once the scan correction coefficients for each scan position and each latitudinal band are calculated, a smoothing method is used to generate a smooth transition between latitudinal bands. The smoothed scan correction coefficient calculation method is:

[0045] ;

[0046] At this point, the scanning deviation correction coefficient is calculated ;

[0047] The following formula is used to eliminate the relative deviation caused by the scanning position and obtain the observation residual after scanning deviation correction:

[0048] ;

[0049] The air mass bias correction includes:

[0050] Using a set of bias predictors To relate to the air mass deviation, the following linear regression equation is used to calculate each channel Air mass deviation:

[0051] ;

[0052] here, Coefficient and It is calculated by least squares fitting using a large number of samples; the above formula is simplified using vectors and matrices as follows:

[0053] ;

[0054] in yes dimensional predictor vector, coefficient matrix It is given by the following formula:

[0055] ;

[0056] here represents the covariance, yes vector of

[0057] use Indicates channel Deviation:

[0058] ;

[0059] Y E is the observed radiance, y(x) is the radiance simulated by the forecast vector x;

[0060] Adjustments to the coefficients and offsets of the prediction factors are made to reduce bias in the radiosonde analysis;

[0061] The variational bias correction includes:

[0062] Variational bias correction estimates and updates the bias parameters within the NWP variational assimilation system and corrects the observed radiance. Specifically, this involves introducing a background term for the bias parameters to avoid fitting any local features of the data based on the bias. For three-dimensional variational assimilation, the new objective functional after applying the variational bias correction method is written as follows:

[0063] ;

[0064] in is the radiance deviation parameter vector, with dimension , Identifies the number of predictors, represents the number of sensors assimilated, Indicates the number of channels each sensor has, is the estimate of the coefficient of variation from the previous analysis, and denote the background error covariance of the atmospheric state vector and the bias parameter, respectively, x b It is a priori knowledge of the atmosphere, It contains the deviation factor X and the deviation coefficient A combinatorial polynomial of two parameters.

[0065] Furthermore, data quality control includes: eliminating observation data from high-latitude areas on mixed surface types, noise, pixel zenith angle, and observation error greater than a preset threshold; for cases where the pixel quality score QA is not equal to 100, and when the channel identification quality is equal to 1, the pixel or channel is eliminated.

[0066] Furthermore, the clear-sky channel cloud detection method, based on band division, sorts the deviations between the observed data and the radiance simulated under the clear-sky assumption according to the channel sensitivity to clouds, so that the cloud signal changes monotonically with the sorting; secondly, a digital filtering method is used to process the deviation data to reduce the atmospheric and instrument noise signals contained in the deviation data; finally, based on the sorted channel sequence, the channels are looped to find the channel where the cloud signal becomes significant for the first time; with this channel as the boundary, the channels sorted above this channel are retained, and the channels sorted below this channel are discarded.

[0067] Furthermore, the specific steps of the clear sky channel cloud detection method are as follows:

[0068] For each field of view point of the hyperspectral data, the observed brightness temperature of each channel is subtracted from the simulated brightness temperature to obtain the brightness temperature deviation vector;

[0069] According to the sensitivity of each channel relative to the cloud in each field of view point, a channel height is assigned to each channel; all channels in the infrared band are grouped, and the channels in each group of separated bands are sorted according to the channel height;

[0070] The sliding average method is used to eliminate the influence of instrument noise and model error in the brightness temperature deviation vector;

[0071] Starting from the channel that is most sensitive to clouds, the following channels are gradually searched. When the brightness temperature deviation is less than a given threshold and the absolute value of the brightness temperature deviation gradient is less than a certain threshold, the channel is identified as a clear sky channel, otherwise it is identified as a cloudy channel. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 Flowchart of the method of this application;

[0073] Figure 2 Flow chart of clear sky channel cloud detection algorithm. DETAILED DESCRIPTION

[0074] The application will be further described below with reference to the drawings, but the application is not limited in any way by the following description, any transformation or replacement based on the teaching of the application shall fall within the protection scope of the application.

[0075] The prerequisite of satellite data assimilation is to simulate or emulate the observation process, that is, a fast radiative transfer model must be designed and developed for each type of observation data. The fast radiative transfer model represents the linear or nonlinear relationship of the variable conversion from the model space to the observation space, which includes a mapping from the geophysical input vector (such as temperature, humidity and pressure variables) of the model space to the simulated observation instrument (such as radiance) of the observation space, in which the physical characteristics of the observation and the instrument properties need to be considered in detail. At present, there are two implementation approaches for satellite radiance assimilation. The first approach is to interpolate the model profile to the observation location, then use the fast radiative transfer model forward operator to simulate the observation, compare the simulated observation with the actual observation to obtain the observation increment, and then convert the observation increment to the analysis increment through the tangent linear and adjoint operators of the fast radiative transfer model. The second approach is to directly apply the radiative transfer model to the model grid location, then compare the output quantity with all observation data within a given configuration radius centered on the model grid to establish a relationship. This assimilation scheme is currently only used to directly assimilate microwave radiance data affected by rain or cloud. The radiative transfer operators used in the two assimilation schemes are both RTTOV radiative transfer models, but the forms are slightly different. The interpolation operator in space and the radiative transfer model together constitute the satellite radiance observation operator. According to the research target of the present application, the implementation of clear sky infrared hyperspectral data assimilation function is given priority, therefore, the satellite infrared hyperspectral data assimilation of the present application adopts the first approach.

[0076] REFERENCE Figure 1In one embodiment, the present application proposes a data assimilation method for clear-sky channel infrared hyperspectral data, comprising the following steps: S1 assimilation preprocessing: obtaining clear-sky channel infrared hyperspectral data; using a fast radiation transfer model to convert variables from a pattern space into a linear or nonlinear relationship in an observation space; S2 channel selection: combining the principal component method to screen satellite infrared hyperspectral atmospheric vertical detection data channels; S3 cloud detection: using the difference between the observed infrared observation brightness temperature and the background field simulated brightness temperature to find channels that are not affected by cloud contamination; S4 bias correction: eliminating the statistical deviation between the radiation value observed by the satellite instrument and the radiation value calculated based on the background field profile simulation; S5 data quality control: eliminating observation data on mixed surface types whose errors exceed a preset threshold; S6 outputting the data assimilation results of the clear-sky channel infrared hyperspectral data.

[0077] Designing an efficient and accurate observation operator is a prerequisite for the direct assimilation of domestic satellite infrared hyperspectral observation data. This application is based on the RTTOV fast radiation transfer model, adopts a high-precision spectral line model, introduces the infrared channel spectral response function, simulates the radiation transfer of typical atmospheric profile distributions in different regions and seasons, designs a statistical regression model of atmospheric transmittance, generates a normalized fast transmittance calculation lookup table for HIRAS-II, and completes the fast radiation transfer model required for the four-dimensional variational assimilation of domestic satellite infrared spectral data, as well as the corresponding tangent line / adjoint model training. Specifically includes the following steps:

[0078] 1) Based on new international atmospheric profile datasets such as EC83 and the HITRAN spectral library, an infrared hyperspectral transmittance dataset is generated using the line-by-line integration mode;

[0079] Based on the new international atmospheric profile datasets such as EC83 and the HITRAN spectral library, the infrared hyperspectral transmittance dataset is analyzed and studied using the line-by-line integration model. The versions of the HITRAN hyperspectral atmospheric molecular absorption line database after 2012 are collected and sorted. The high-spectral resolution line-by-line integration atmospheric transmittance model LBL is investigated and analyzed, and mature versions are selected. The atmospheric profile datasets such as EC83 are collected and sorted. The climate representativeness and extreme weather representativeness of the global representative atmospheric sample datasets are compared and analyzed. The hyperspectral atmospheric transmittance is calculated using the high-spectral resolution line-by-line integration atmospheric transmittance model. The line-by-line integration forward model with a resolution of 0.001cm-1 is carried out for the CO2 15μm absorption line, the H2O 6.7μm absorption line and the 8-12μm infrared window area. The CO2, CO, NH4, N2O, NO2, CFC are calculated from the ground to 0.05hPa altitude and within the satellite zenith angle range of 0-70. 11 、CFC 14The monochromatic transmittance of atmospheric absorption components at at least 90 isobaric surfaces, and then CO2, CO, NH4, N2O, NO2, CFC 11 、CFC 14 The transmittance of the uniform mixed gas is synthesized, and the continuous absorption transmittance of H2O in the full spectrum is calculated separately to form a layered atmospheric transmittance data set of uniform mixed gas, water vapor and O3 from the ground to the effective top of the atmosphere;

[0080] 2) Using the established hyperspectral infrared representative atmospheric transmittance dataset, a fast forward operator for clear sky of Fengyun satellite infrared hyperspectral images was established based on the multivariate linear regression method.

[0081] The spectral response function of the infrared hyperspectral channel is established by using the sinc function expansion and the Hanming window function toe-cut. A fast calculation method based on the channel optical thickness is studied and analyzed. Based on the least squares method, a fast channel optical thickness coefficient calculation scheme for uniform mixed gas absorption, water vapor line absorption, water vapor continuous absorption, and O3 absorption is established. Because the channel convolution transmittance of a single gas is different from the channel convolution transmittance shared by all absorbing gases, before the regression calculation, the transmittance of the channel can be expressed as follows using the transmittance of the monochromatic absorbing gas (uniform mixed gas, H2O, and O3):

[0082] ;

[0083] It should be noted that when calculating the monochromatic transmittance of a uniform mixture of gas and water vapor, the previous equation may be divided by 0 due to a small value. For this reason, an empirical constant 1.0e-6 is defined. When the transmittance of any isobaric surface is less than this constant, the transmittance can be corrected using the water vapor content of the adjacent isobaric surface:

[0084] ;

[0085] Among them, constant constant=1.e-6, amount is the absorbed gas content.

[0086] For a given atmospheric state (atmospheric temperature, water vapor and ozone profiles) and surface parameters (surface emissivity, pressure, temperature and surface exposed temperature), the The rapid radiation transfer mode can simulate the channel transmittance from the corresponding pressure layer to the top of the atmosphere. Or the optical thickness of the channel from the corresponding pressure layer to the top of the atmosphere ( ), where the optical thickness from pressure layer j to the top of the atmosphere at beam νi after convolution can be defined as:

[0087] ;

[0088] in, is the prediction factor that depends on the profile, M is the number of prediction factors, is the transmittance calculation coefficient.

[0089] Define the linear regression system:

[0090] ;

[0091] Used for regression calculation of transmittance calculation coefficient .

[0092] After the large-sample learning process described above, this application generated a HIRAS-II infrared hyperspectral rapid radiation transfer model based on RTTOV v11.3, training the HIRAS-II instrument's rapid transmittance coefficients. This model provides the satellite infrared hyperspectral instrument background field simulation information for the four-dimensional variational assimilation system, providing foundational support for subsequent assimilation application research. Currently, the HIRAS-II instrument uses 101 atmospheric layers of transmittance coefficients.

[0093] Under the premise of ensuring the assimilation effect and considering the computational complexity, in one embodiment, this application uses the principal component method to carry out channel screening of satellite infrared hyperspectral atmospheric vertical sounding data. The principal component method is a classic data dimensionality reduction method. Its core idea is to replace the original set of related variables with a smaller number of unrelated variables, namely principal components, to achieve the purpose of reducing the dimension of the data set. In addition, the newly obtained variables can effectively preserve most of the information of the original data. The steps are as follows:

[0094] Step 1: Channel screening method based on the cumulative impact of principal components

[0095] Spaceborne hyperspectral atmospheric vertical sounders typically have thousands of detection channels. Directly assimilating observational data from such a large number of channels presents several significant challenges: First, the computational complexity of assimilation is prohibitive, significantly increasing the computational time and impacting operational performance. Second, the significant correlations between hyperspectral channels can lead to assimilation ill-posedness and information redundancy, resulting in negative assimilation effects. While ensuring both assimilation effectiveness and computational complexity, this application proposes a hyperspectral infrared data channel selection scheme based on the cumulative influence coefficient of the principal component and the observation bias of the assimilation system. This scheme comprehensively considers the influence of the cumulative influence coefficient of the principal component of the detection channel observation information and the initial guess field bias (OB: observed radiance minus model simulated radiance).

[0096] First, based on the channel selection method of the cumulative influence coefficient of the principal component, the importance of the original observation data channels of the infrared hyperspectral data is sorted offline. Assuming that the original infrared hyperspectral observation is : ; . Then use the principal component method to map it to the new variable space : ; .in:

[0097] , ;

[0098] in: : ; , The row vector is the original data The eigenvector of the covariance matrix of . In this project, n is the number of model layers and m is the number of channels. If the coefficient satisfy , At the same time, it must also be ensured are unrelated, and yes The jth largest variance among all linear combinations of Raw data The jth principal component of .

[0099] In the actual processing process, it is necessary to calculate the principal component cumulative influence coefficient for different types of detection channels such as temperature (CO2 absorption band), humidity (water vapor absorption band) and window area (visible light), and then gradually consider the combination of daytime and nighttime detection channels, as well as the influence of sunlight, to finally achieve the optimal confirmation of hyperspectral channels. The following takes the temperature detection channel as an example. After eliminating the channels affected by water vapor and those in the blacklist, the calculation process of the principal component cumulative influence coefficient based on the Jacobian matrix of the temperature detection channel is detailed:

[0100] Step 1: First, obtain the Jacobian matrix H(x) of the temperature detection channel. The variables of each channel are calculated according to the following formula: After the transformation, the standardized matrix is ​​obtained, which is recorded as: ;in are the elements of the normalized matrix, Indicates the number of mode layers, Indicates the number of channels;

[0101] Step 2: Calculation The covariance matrix of ;

[0102] Step 3: Calculation The eigenvalue of and eigenvectors, and the eigenvalues Arrange the values ​​from large to small;

[0103] Step 4: Perform principal component analysis and select the first p principal components. The principle of selecting the first p principal components is: given a certain percentage ε (usually ε is equal to 90%), select p so that , represents the percentage of the total variance accounted for by the first p principal components, that is, the total contribution rate of the p principal components. represents the jth largest eigenvalue;

[0104] Step 5: Map it into a new vector, obtain the main components with the largest influence according to step 4 above, and enter the analysis of cumulative influence coefficient;

[0105] Step 6: If the same channel is selected in different principal components, the influence coefficient s is calculated. It is defined as the product of the principal component's eigenvector and the variance contribution rate of each principal component. The variance contribution rate is defined as the ratio of the corresponding eigenvalue of each principal component to the sum of the total eigenvalues, which is the variance contribution rate of the principal component. Otherwise, the influence coefficient is calculated and finally sorted to select the channels with the greatest impact on temperature. When the number of selected channels reaches the initial number of channels, or the cumulative influence coefficient of a channel is less than the given threshold, stop channel selection and go to step 4.

[0106] Finally, 60 channels were selected from the 3041 channels of FY-3E HIRAS-II.

[0107] A principal component-based approach was used to select 3041 channels from the HIRAS-II hyperspectral data, currently considering only the impact on temperature information. After iteratively looping through the six steps above, the influence coefficients of each HIRAS channel were calculated and ranked for suitability in the 4D-Var assimilation system. Channels were then further selected based on the distribution of the observed-simulated (OB) values ​​for each of the 3041 HIRAS channels. The statistical results of the OB values ​​for each HIRAS-II channel show that, while the mean values ​​(mean) of some channels exhibit significant positive or negative deviations, the mean values ​​for most HIRAS channels are near zero. Furthermore, the root mean square error (std) of the OB values ​​for the HIRAS-II data is very close to that of the internationally comparable IASI instrument. Based on the channel ranking, observational channels with mean (OB) values ​​greater than 0.5 and std (OB) values ​​greater than 3.5 were further excluded, resulting in a final selection of 60 HIRAS-II channels suitable for the 4D-Var assimilation system, as shown in the table below.

[0108] Table 1 List of 60 preferred channels of HIRAS-II

[0109]

[0110] In one embodiment, the present application implements the deviation correction and quality control of HIRAS-II as follows:

[0111] Satellite data biases arise from a wide range of sources, and the brightness temperature and emissivity biases of different detectors share both commonalities and unique characteristics. The bias correction methods used in this application include both offline and online bias correction, including but not limited to static bias, adaptive bias, regression bias, variational bias, bias correction based on radiative transfer models, Kalman filtering, and dynamic bias update techniques.

[0112] Variational assimilation theory requires that model and observation errors be unbiased and Gaussian. However, the statistical deviations between the radiance values ​​observed by satellite instruments and those calculated based on the background field profile are systematic rather than random, and these errors come from a wide range of sources, including:

[0113] 1) Biases in background fields based on model short-term forecasts;

[0114] 2) Systematic deviations in radiative transfer patterns, such as RTTOV;

[0115] 3) Errors in observational data contaminated by clouds, especially precipitation clouds;

[0116] 4) Calibration and positioning errors of observation instruments;

[0117] 5) Changes in the corresponding characteristics of the sensor over time can also cause systematic deviations in radiation measurements.

[0118] The goal of bias correction is to eliminate or reduce these system biases. Currently, common bias correction methods include air mass bias correction, air mass bias correction, and variational bias correction. Bias correction is a challenging task, requiring unique correction methods tailored to different instruments and channels. Existing bias correction methods are primarily implemented in two steps: scanning bias correction and air mass bias correction.

[0119] Satellite channel observations and simulated brightness temperature statistics reveal an overall "limb darkening" phenomenon, i.e., a decrease in brightness temperature on either side of the subsatellite point. This is due to the fact that, for a given channel frequency and atmospheric density, the optical thickness of the atmosphere increases with scanning angle away from the subsatellite point, necessitating a scan bias correction. Furthermore, biases in the forward model, caused by inaccurate rapid radiative transfer models and transmittance coefficient calculations, tend to vary with air mass and surface characteristics, necessitating air mass bias correction. The key issue in air mass bias correction lies in the selection of predictors; different predictors yield different bias correction results. Commonly used predictors include surface temperature, precipitable water content, 1000-300 hPa thickness, 850-300 hPa thickness (also known as tropospheric thickness), and 200-50 hPa thickness (also known as stratospheric thickness).

[0120] The deviation corrections used in this application include:

[0121] 1) Scanning deviation correction

[0122] Scanning bias correction is required for all observation instruments on polar-orbiting satellites. For each observation field of view (FOV), a constant bias correction relative to the center of the scanning band is required. The scanning bias correction scheme of this application plans to adopt the Harris and Kelly scheme, introduce a latitude-dependent scanning bias correction, and increase the latitude by 10 degrees. o The Earth is divided into 18 latitude zones using meridians as a band. The average values ​​for each scan position and each latitude zone are calculated and then used to calculate the scan correction factor. .Right now ;

[0123] in It is a latitude belt;

[0124] To avoid discontinuities in scan bias correction between latitudinal bands, a smoothing technique is used to generate continuous correction coefficients in areas that span latitudinal bands. Once the scan correction coefficients are calculated for each scan position and each latitudinal band, a simple smoothing method is used to generate smooth transitions between latitudinal bands. Smoothed scan correction coefficient calculation method:

[0125] ;

[0126] So far, the scanning deviation correction coefficient has been calculated If you only want to eliminate the relative deviation caused by the scanning position, you can apply it to the following formula to obtain the observation residual after scanning deviation correction:

[0127] .

[0128] 2) Air mass deviation correction

[0129] The air mass bias correction regression scheme uses a set of bias predictors To relate to the air mass deviation, the following linear regression equation is used to calculate each channel Air mass deviation:

[0130] ;

[0131] here, Coefficient and It is calculated by least squares fitting using a large number of samples (usually two weeks of data). The above formula can be written in the following more concise form using vectors and matrices.

[0132] ;

[0133] in yes dimensional predictor vector, coefficient matrix It is given by the following formula:

[0134] ;

[0135] here represents the covariance, yes A vector of . Indicates channel Deviation:

[0136] ;

[0137] The key to correcting air mass biases lies in the choice of predictors. For spaceborne infrared instruments (HIRS, AIRS, CRIS, and IASI), the bias model of Harris and Kelly (2001) is currently used. This model involves a linear regression based on a set of predictors derived from NWP models. The coefficients and offsets corresponding to these predictors can be adjusted to reduce bias in radiosonde analysis.

[0138] 3) Variational bias correction

[0139] The Variational Bias Correction (VarBC) scheme is used for operational assimilation of satellite radiance. The VarBC method is used to estimate the bias parameters required in the radiance bias correction model and these parameters are used as part of the control variables in the main variation analysis procedure.

[0140] The variational bias correction (VarBC) scheme is an adaptive bias correction system that estimates and updates bias parameters within the NWP variational assimilation system and corrects the observed radiance. The regression coefficients (i.e., bias parameters) in the bias correction are used as part of the control variables in the analysis. Alternatively, the variational bias correction method can be considered an extension of the observation operator, which is a function of both the NWP model state and the bias parameters. To avoid fitting any local features of the data (e.g., cloud contamination) based on the bias, a background term for the bias parameter is introduced. After the initial guess of the bias parameter is introduced as a constraint, the background term can be considered an inertial term. Taking three-dimensional variational assimilation as an example, the new objective functional can be written as follows after using the variational bias correction method:

[0141] ;

[0142] in is the radiance deviation parameter vector, with dimension , Identifies the number of predictors, represents the number of sensors assimilated, Indicates the number of channels each sensor has, is the estimate of the coefficient of variation from the previous analysis, and denote the background error covariance of the atmospheric state vector and the bias parameter, respectively, x b is the prior knowledge of the atmosphere (usually called the background field), It contains the deviation factor X and the deviation coefficient A combinatorial polynomial of two parameters.

[0143] The introduction of bias parameters as control variables adds additional degrees of freedom to the variational assimilation system during its convergence to a solution. The primary advantage of this approach is that it allows for the combined consideration of bias parameters and model state variables to determine the optimal estimate of the analysis field. If the introduction of these new control parameters does not lead to new local minima in the objective functional of the multi-source meteorological and hydrological data assimilation system, the variational assimilation system, by incorporating these additional degrees of freedom, will yield a solution that better approximates the best linear unbiased estimate (BLUE). Embedding VarBC within the minimization process does incur a certain degree of computational overhead. This overhead is manageable for limited data types and simple bias models, but becomes more pronounced with more complex bias models.

[0144] Based on the above-mentioned HIRAS-II bias characteristic analysis results, and referring to the variational bias correction schemes of existing infrared hyperspectral instruments (AIRS and IASI), this application designs the following prediction factor selection scheme for the FY-3E HIRAS-II.

[0145] Table 2. Correction scheme for HIRAS-II instrument bias

[0146]

[0147] Table 3. Corresponding parameters of HIRAS-II prediction factors

[0148]

[0149] The table below outlines the HIRAS data quality control scheme. This strategy primarily targets observations from high-latitude regions with mixed surface types, high noise levels, excessively large pixel zenith angles, and large observation errors. Furthermore, pixels with a QA score less than 100 and channel IDs equal to 1 are excluded. Furthermore, a HIRAS-II FOR pixel contains nine pixels, but only the fifth pixel is retained to reduce correlation between observations.

[0150] Table 4 HIRAS-II quality control plan

[0151]

[0152] After applying variational bias correction to the HIRAS-II data, the bias characteristics between the observations and the background field simulations were effectively improved. The above scheme effectively eliminated data with large observation errors at the junction of ascending and descending orbits and in tropical regions.

[0153] In one embodiment, the HIRAS-II infrared hyperspectral channel cloud detection method proposed in this application includes:

[0154] Because clouds can be roughly treated as blackbodies for infrared radiation, cloud detection research is an essential and critical step before assimilation of infrared hyperspectral remote sensing data. In actual observations, the completely cloud-free infrared observation field accounts for only about 10% of the total field of view. Simply discarding data from cloud-affected fields would result in a significant waste. For global forecast models, observations falling within clear sky areas are relatively rare, and utilizing only data from completely clear sky areas is far from sufficient. Secondly, clouds strongly influence thermal infrared radiation observations. For example, in the middle troposphere, near the long-wavelength (15μm) temperature detection channel, the radiance error caused by uncertainties in the atmospheric temperature profile is typically less than a few Kelvin. However, in the presence of clouds in the same area, this error can reach tens of Kelvin. This imbalance makes infrared radiation data within cloud regions difficult to use.

[0155] Cloud detection method and setup for clear-sky channel of HIRAS-II infrared hyperspectral data

[0156] For infrared hyperspectral remote sensing data, a cloud detection scheme based on clear-sky channels can identify clear-sky channels in the field of view, thus avoiding data waste and introducing atmospheric observation information above the cloud tops into the assimilation system. The clear-sky channel-based cloud detection method focuses on identifying whether each channel in the field of view is affected by clouds. Its basic concept is as follows: first, based on band division, the anomalies between the observed data and the simulated radiance under the clear-sky assumption are sorted according to the channel's sensitivity to clouds, so that the cloud signal changes monotonically with the sorting. Second, digital filtering is used to process the anomaly data to reduce the atmospheric and instrument noise signals contained in the anomaly data. Finally, based on the sorted channel sequence, the channels are looped through to find the channel where the cloud signal first becomes significant. Using this channel as the demarcation line, channels ranked above this channel (i.e., channels less sensitive to clouds) are retained, while channels ranked below this channel are discarded.

[0157] refer to Figure 2 The clear-sky channel cloud detection method is based on band separation, channel sorting, and digital filtering techniques. It uses the difference between infrared observation brightness temperature and background field simulation brightness temperature (usually provided by the 6-hour forecast field of the numerical model) to find channels that are not affected by cloud contamination. The specific steps are as follows:

[0158] a. For each field of view point of the hyperspectral data, calculate the observed brightness temperature of each channel minus the simulated brightness temperature to obtain the brightness temperature deviation vector;

[0159] b. Assign channel heights to each channel based on its sensitivity to clouds at each viewing point. Group all channels within the infrared band and sort the channels within each group of separated bands by channel height.

[0160] c. Since the brightness temperature deviation vector contains errors such as instrument noise and model error, the sliding average method is used to filter and eliminate the influence of these errors;

[0161] d. Starting from the channel most sensitive to clouds, gradually search for the following channels. When the brightness temperature deviation is less than a given threshold and the absolute value of the brightness temperature deviation gradient is less than a certain threshold, the channel is identified as a clear sky channel, while channels below the threshold are considered to have clouds.

[0162] The rapid cloud detection algorithm based on the spaceborne hyperspectral atmospheric vertical sounding data of the clear sky channel will perform cloud detection on 137 channels of HIRAS entering the YH4DVAR3.0 assimilation system, setting the brightness temperature threshold to 0.65 and the brightness temperature gradient threshold to 0.1.

[0163] The results of the HIRAS-II data assimilation application and experimental evaluation are as follows: the standard statistical procedure test procedure specified by the World Meteorological Organization Basic Committee was used to statistically calculate the correlation coefficients of the analyzed 500hPa geopotential height and 850hPa temperature anomalies for forecast pairs in different regions around the world, and the degree of improvement of the FY-3E HIRAS data in the forecast results in different regions was systematically analyzed and evaluated.

[0164] Table 5 describes the full-information assimilation test setup for the FY-3E HIRAS-II infrared hyperspectral data. The BASE experiment adds microwave AMSUA and GPS occultation observations to the conventional data. The ALL experiment adds FY-3E HIRAS-II infrared hyperspectral data to the BASE experiment. 40-day forecasting experiments are conducted for the spring and summer of 2023, respectively.

[0165] Table 5 FY-3E HIRAS-II evaluation test settings under full information conditions

[0166]

[0167] The statistical test results of different variables in the experiment of assimilating domestic satellite infrared hyperspectral data (HIRAS-II) under information-poor conditions at different pressure layers all showed consistency superior to the experiment of assimilating conventional observation data (conv). These results demonstrate that the satellite infrared hyperspectral data assimilation function of this application is correct and effective. Assimilating domestic satellite FY-3E infrared hyperspectral HIRAS-II data can significantly improve the analysis fields of all layers of the global atmosphere and effectively enhance the level of numerical forecasting. Overall, the FY-3E satellite HIRAS-II infrared hyperspectral data significantly improved the predicted intensity error of Typhoon Dusurui, and the errors of its development speed and path before landfall were superior to those of the experiment using conventional observation data.

[0168] As used herein, the word "preferred" is intended to serve as an example, instance, or illustration. Any aspect or design described herein as "preferred" is not necessarily to be construed as advantageous over other aspects or designs. Rather, the use of the word "preferred" is intended to present concepts in a concrete manner. As used in this application, the term "or" is intended to mean an inclusive "or" rather than an exclusive "or." That is, unless otherwise specified or clear from the context, "X employs A or B" is intended to mean any of the naturally inclusive permutations. That is, if X employs A; X employs B; or X employs both A and B, then "X employs A or B" is satisfied in any of the foregoing examples.

[0169] Moreover, although the present disclosure has been shown and described with respect to one or implementations, those skilled in the art will think of equivalent variations and modifications based on reading and understanding of this specification and the accompanying drawings. The present disclosure includes all such modifications and variations and is limited only by the scope of the appended claims. In particular, with respect to the various functions performed by the above-mentioned components (such as elements, etc.), the terms used to describe such components are intended to correspond to any component (unless otherwise indicated) that performs the specified function of the component, even if structurally not equivalent to the disclosed structure that performs the function in the exemplary implementation of the present disclosure shown herein. In addition, although the specific features of the present disclosure have been disclosed with respect to only one of several implementations, such features can be combined with one or other features of other implementations that may be desired and advantageous for a given or specific application. Moreover, insofar as the terms "including", "having", "containing" or their variations are used in specific embodiments or claims, such terms are intended to be included in a manner similar to the term "comprising".

[0170] The functional units in the embodiments of the present invention may be integrated into a single processing module, or each unit may exist physically separately, or multiple or more units may be integrated into a single module. The aforementioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium. The aforementioned storage medium may be a read-only memory, a magnetic disk, or an optical disk, etc. The aforementioned devices or systems may execute the storage method in the corresponding method embodiment.

[0171] In summary, the above embodiment is one implementation method of the present invention, but the implementation method of the present invention is not limited to the described embodiment. Any other changes, modifications, substitutions, combinations, and simplifications that deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the scope of protection of the present invention.

Claims

1. A data assimilation method for clear-sky channel infrared hyperspectral data, characterized in that: The following steps are involved: Assimilation preprocessing: obtain clear sky channel infrared hyperspectral data; Use fast radiative transfer models to transform variables from pattern space into linear or nonlinear relationships in observation space; Channel selection: Combine the principal component method to screen the satellite infrared hyperspectral atmospheric vertical detection data channels; Cloud detection: Utilize the difference between infrared observation brightness temperature and background field simulation brightness temperature to find channels that are not affected by cloud contamination; Bias correction: Eliminate the statistical deviation between the radiation values ​​observed by satellite instruments and the radiation values ​​calculated based on the background field profile simulation; Data quality control: Observation data with errors exceeding a preset threshold on mixed surface types are eliminated; Output the data assimilation results of clear sky channel infrared hyperspectral data; The specific steps of the cloud detection method are as follows: For each field of view point of the hyperspectral data, the observed brightness temperature of each channel is subtracted from the simulated brightness temperature to obtain the brightness temperature deviation vector; According to the sensitivity of each channel relative to the cloud in each field of view point, a channel height is assigned to each channel; all channels in the infrared band are grouped, and the channels in each group of separated bands are sorted according to the channel height; The sliding average method is used to eliminate the influence of instrument noise and model error in the brightness temperature deviation vector; Starting from the channel that is most sensitive to clouds, the following channels are gradually searched. When the brightness temperature deviation is less than a given threshold and the absolute value of the brightness temperature deviation gradient is less than a certain threshold, the channel is identified as a clear sky channel, otherwise it is identified as a cloudy channel.

2. The data assimilation method based on the clear sky channel infrared hyperspectral data according to claim 1 is characterized in that: The rapid radiative transfer mode includes: interpolating the model profile to the observation position, then using the rapid radiative transfer mode forward operator to simulate the observation, obtaining the observation increment after comparing it with the actual observation, and then converting the observation increment into an analysis increment through the rapid radiative transfer mode tangent and adjoint operators.

3. The data assimilation method based on the clear sky channel infrared hyperspectral data according to claim 2 is characterized in that: The rapid radiative transfer model is based on the RTTOV rapid radiative transfer model, adopts a high-precision spectral line model, introduces the infrared channel spectral response function, simulates radiative transfer based on the typical atmospheric profile distribution in different regions and seasons, designs a statistical regression model for atmospheric transmittance, generates a normalized rapid transmittance calculation lookup table for HIRAS-II, and completes the rapid radiative transfer model required for the four-dimensional variational assimilation of satellite infrared spectral data, as well as the corresponding tangent linear / adjoint model training. Specifically, the following steps are included: Based on the atmospheric profile dataset and spectral library, the infrared hyperspectral transmittance dataset is generated using the line-by-line integration mode; Using the infrared hyperspectral transmittance dataset, a satellite infrared hyperspectral clear sky fast forward operator is established based on the multivariate linear regression method; Before the regression calculation, the transmittance of the monochromatic absorbing gas and the transmittance of the channel are used. Expressed as: ； Where v is the channel center frequency, j is the mode layer, is the channel mixed gas transmittance from mode layer j to the external space with a center frequency of v, is the transmittance of the mixed gas containing water vapor in the range from mode layer j to the outer space for the channel with the center frequency v from mode layer j to the outer space, is the atmospheric transmittance of water vapor and ozone in the mixed gas from mode layer j to the outer space; When the transmittance of any isobaric surface is less than the preset empirical constant, the transmittance is corrected using the water vapor content of the adjacent isobaric surface: ； Among them, constant constant= , amount is the absorbed gas content, k is the number of pressure layers, is the mixed gas transmission rate containing water vapor in the kth layer, is the mixed gas transmission rate, is the mixed gas transmission rate of the kth layer, is the water vapor content in the kth layer; For a given atmospheric state and surface parameters, along the observation angle The propagation path of the fast radiation transfer mode simulates the channel transmittance from the jth pressure layer to the top of the atmosphere. Or the optical thickness of the path from the jth pressure layer to the top of the atmosphere ,in , after convolution, the optical thickness of beam νi from pressure layer j to the top of the atmosphere is defined as: ； in, is the prediction factor that depends on the profile, M is the number of prediction factors, is the transmittance calculation coefficient; define the following linear regression equation group for regression calculation of transmittance calculation coefficient : ； is the transmittance of the channel from the j-1th pressure layer to the top of the atmosphere in the model.

4. The data assimilation method based on the clear sky channel infrared hyperspectral data according to claim 3 is characterized in that: The principal component method is used to screen satellite infrared hyperspectral atmospheric vertical detection data channels. The steps are as follows: Firstly, based on the channel selection method of the cumulative influence coefficient of the principal component, the importance of the original observation data channels of the infrared hyperspectral data is sorted offline. The original observation data of the infrared hyperspectral data is : ; , and then use the principal component method to map it to the new variable space : ; ,in: ， ； in: : ; , The row vector is the original data The eigenvectors of the covariance matrix, u ij is the feature of the jth channel of the i-th pattern layer after mapping, n is the number of pattern layers, m is the number of channels, x ij is the original observation of the jth channel of the i-th mode layer, y ij is the observation of the jth channel of the i-th mode layer after mapping, X m is the original observation of channel m; if the coefficient satisfy , ; It also guarantees are unrelated, and yes Among all linear combinations of For the original data The jth principal component of The principal component cumulative influence coefficients are calculated for the temperature, humidity, and window area detection channels respectively. Then, the combination of daytime and nighttime detection channels, as well as the influence of sunlight, are gradually considered to ultimately achieve the optimal confirmation of the hyperspectral channel. The influence coefficients of the HIRAS channels of the four-dimensional variational assimilation system were ranked. Then, the channels were further screened based on the observation-simulation distribution characteristics of multiple HIRAS channels. Based on the channel ranking, the observation channels with observation-simulation mean > 0.5 and observation-simulation variance > 3.5 were further excluded. Finally, the HIRAS-II preferred channels suitable for the four-dimensional variational assimilation system were obtained.

5. The data assimilation method based on the clear sky channel infrared hyperspectral data according to claim 4 is characterized in that: For the temperature detection channel, after removing the channels affected by water vapor and those in the blacklist, the cumulative influence coefficient of the principal component is calculated based on the Jacobian matrix of the temperature detection channel. The process is as follows: Step 1: Get the Jacobian matrix H(x) of the temperature detection channel. The variables of each channel are calculated according to the following formula: After the transformation, the standardized matrix is ​​obtained, which is recorded as: ;in are the elements of the normalized matrix, Indicates the number of mode layers, Indicates the number of channels, s ji is the variance of the jth channel in layer i, x ji is the element of the jth channel in the i-th layer, is the mean of the i-th layer; Step 2: Calculate the normalized matrix The covariance matrix of ; Step 3: Calculation The eigenvalue of and eigenvectors, and the eigenvalues Arrange the values ​​from large to small; Step 4: Perform principal component analysis and take the first p principal components; The principle of selecting the first p principal components: given a preset percentage ε, select p so that , represents the percentage of the first p principal components in the total variance, that is, the total contribution rate of the p principal components, where represents the jth largest eigenvalue; Step 5: Map it into a new vector, obtain the previous main components according to Step 4, and enter the analysis of cumulative influence coefficient; Step 6: If the same channel is selected in different principal components, the influence coefficient s is calculated. S is the product of the eigenvector of the principal component and the variance contribution rate of each principal component. The difference contribution rate is defined as the ratio of the corresponding eigenvalue of each principal component to the sum of the total eigenvalues, which is the variance contribution rate of the principal component. Otherwise, the influence coefficient is calculated and finally sorted to select the channels whose influence on temperature is greater than the preset threshold. When the number of selected channels reaches the initially given number of channels, or the cumulative influence coefficient of a channel is less than the given threshold, the channel selection is stopped and the process goes to Step 4.

6. The data assimilation method based on the clear sky channel infrared hyperspectral data according to claim 5 is characterized in that: The bias correction includes at least one of scanning bias correction, air mass bias correction and variational bias correction; The scanning deviation correction includes: Introduces latitude-dependent scan bias corrections and o The Earth is divided into 18 latitude bands using meridians as a band; the average values ​​for each scan position and each latitude band are calculated and then used to calculate the scan correction factor ,Right now: ； in It is a latitude belt, is the observation angle, is the average value of the scan positions, is the average value for the latitudinal band; To avoid discontinuity in the scan bias correction between latitudinal bands, a smoothing technique is used to generate continuous correction coefficients in areas that span latitudinal bands. Once the scan correction coefficients for each scan position and each latitudinal band are calculated, a smoothing method is used to generate a smooth transition between latitudinal bands. The smoothed scan correction coefficient calculation method is: ； At this point, the scanning deviation correction coefficient is calculated ; The following formula is used to eliminate the relative deviation caused by the scanning position and obtain the observation residual after scanning deviation correction: ； The air mass bias correction includes: Using a set of bias predictors To relate to the air mass deviation, the following linear regression equation is used to calculate each channel Air mass deviation: ； here, Coefficient and It is calculated by least squares fitting using a large number of samples; the above formula is simplified using vectors and matrices as follows: ； in yes dimensional predictor vector, coefficient matrix It is given by the following formula: ； here represents the covariance, yes vector of use Indicates channel Deviation: ； Y E is the observed radiance, y(x) is the radiance simulated by the forecast vector x; Adjustments to the coefficients and offsets of the prediction factors are made to reduce bias in the radiosonde analysis; The variational bias correction includes: Variational bias correction estimates and updates the bias parameters within the NWP variational assimilation system and corrects the observed radiance. Specifically, this involves introducing a background term for the bias parameters to avoid fitting any local features of the data based on the bias. For three-dimensional variational assimilation, the new objective functional after applying the variational bias correction method is written as follows: ； in is the radiance deviation parameter vector, with dimension , Identifies the number of predictors, represents the number of sensors assimilated, Indicates the number of channels each sensor has, is the estimate of the coefficient of variation from the previous analysis, and denote the background error covariance of the atmospheric state vector and the bias parameter, respectively, x b It is a priori knowledge of the atmosphere, It contains the deviation factor X and the deviation coefficient A combinatorial polynomial of two parameters.

7. The data assimilation method based on the clear sky channel infrared hyperspectral data according to claim 6 is characterized in that: Data quality control includes: eliminating observation data from high-latitude areas on mixed surface types, noise, pixel zenith angle, and observation error greater than the preset threshold; if the pixel quality score QA is not equal to 100, or the channel identification quality is equal to 1, the pixel or channel will be eliminated.

8. The data assimilation method based on the clear sky channel infrared hyperspectral data according to claim 7 is characterized in that: The clear-sky channel cloud detection method, based on band division, sorts the anomalies between the observed data and the simulated radiance under the clear-sky assumption according to the channel's sensitivity to clouds, so that the cloud signal changes monotonically with the sorting. Secondly, a digital filtering method is used to process the anomaly data to reduce the atmospheric and instrument noise signals contained in the anomaly data. Finally, based on the sorted channel sequence, the channels are cycled to find the channel where the cloud signal first becomes significant. Taking this channel as the boundary, the channels sorted above this channel are retained, and the channels sorted below this channel are discarded.

Images