A multi-scale assimilation and prediction method and system for FY-4B satellite aerosol observation data

Abstract

The application discloses a FY-4B satellite aerosol observation data multiscale assimilation prediction method and system, relates to the meteorological analysis technical field, and the method comprises the following steps: obtaining land aerosol optical depth observation data inverted by an AGRI sensor carried by a FY-4B stationary meteorological satellite; constructing a double-layer nested based on a WRF-Chem regional weather-chemical coupling mode, and obtaining a collection background field based on a collection emission disturbance strategy; performing 4D-LETKF assimilation analysis based on the land aerosol optical depth observation data and the collection background field, so as to obtain a collection analysis field; and running a short-time WRF-Chem prediction based on the obtained collection analysis field, so as to obtain aerosol short-term prediction results. The application realizes analysis and short-term prediction of regional aerosols with high space-time resolution, and significantly improves the simulation capability for high-resolution spatial distribution characteristics and rapid change processes in the hour scale.

Description

Technical Field

[0001] This invention relates to the field of meteorological analysis technology, and in particular to a multi-scale assimilation forecasting method and system for FY-4B satellite aerosol observation data. Background Technology

[0002] Atmospheric aerosols are key atmospheric components affecting air quality, weather, and climate, exhibiting high spatiotemporal variability in their concentration and optical properties. Satellite remote sensing provides a wide-ranging, near-real-time information source for aerosol observation, while data assimilation techniques that combine satellite observations with numerical models are the primary means of improving the accuracy of aerosol analysis and forecasting.

[0003] Currently, the mainstream aerosol assimilation systems internationally include: NASA's MERRA-2 reanalysis system (based on gridded statistical interpolation GSI for assimilation of multi-source AOTs such as MODIS and AVHRR), ECMWF's CAMS system (using four-dimensional variational 4D-Var to assimilate MODIS and PMAp AOTs), and the Japan Meteorological Institute's scheme based on LETKF for assimilating MODIS and AOTs. These systems have promoted the development of aerosol reanalysis on global or continental scales, but they generally suffer from the following shortcomings: The resolution gap is large. The horizontal resolution of existing operational assimilation systems is mostly 50-200 km, which is 1-2 orders of magnitude different from the original satellite observations (such as FY-4B reaching 4 km). This causes satellite phase cells to be averaged and aggregated at coarse resolution grid points, and signals such as local emissions at the sub-grid point scale are smoothed out.

[0004] Low temporal frequency. Most existing assimilation systems use a 3-hour or 6-hour assimilation window and assimilate several times a day, which is difficult to match the 15-minute high-frequency observation rhythm of geostationary satellites (FY-4B) and cannot characterize the rapid changes in aerosols on an hourly scale.

[0005] The simulation capability for rapid pollution processes is poor. For processes such as sandstorms and sudden pollution that evolve rapidly within hours, the forecast fields of existing coarse-resolution, low-frequency assimilation systems are unable to accurately simulate them. Summary of the Invention

[0006] To address the aforementioned shortcomings in existing technologies, this invention provides a multi-scale assimilation prediction method and system for FY-4B satellite aerosol observation data, which solves the problems of large resolution gaps, low time frequency, and poor simulation capabilities for rapid pollution processes in existing technologies.

[0007] To achieve the above-mentioned objectives, this invention provides a multi-scale assimilation prediction method for FY-4B satellite aerosol observation data, comprising: Acquire land aerosol optical thickness observation data retrieved by the AGRI sensor carried by the FY-4B geostationary meteorological satellite; A two-layer nested model was constructed based on the WRF-Chem regional weather-chemical coupling model, and the ensemble background field was obtained based on the ensemble emission perturbation strategy; 4D-LETKF assimilation analysis was performed based on terrestrial aerosol optical thickness observation data and ensemble background field to obtain the ensemble analysis field; Based on the obtained ensemble analysis field, short-time WRF-Chem forecasts were run to obtain short-term aerosol forecast results.

[0008] Secondly, this invention also provides a system for implementing a multi-scale assimilation prediction method for FY-4B satellite aerosol observation data, comprising: The observation data preprocessing module is used to acquire land aerosol optical thickness observation data retrieved by the AGRI sensor carried by the FY-4B geostationary meteorological satellite; The model ensemble forecast module is used to construct a two-layer nested model based on the WRF-Chem regional weather-chemical coupling model and obtain the ensemble background field based on the ensemble emission perturbation strategy. The 4D-LETKF assimilation analysis module is used to perform 4D-LETKF assimilation analysis based on terrestrial aerosol optical thickness observation data and ensemble background field to obtain the ensemble analysis field. The short-term forecast output module is used to run short-time WRF-Chem forecasts based on the obtained ensemble analysis field to obtain short-term aerosol forecast results.

[0009] The beneficial effects of this invention are as follows: 1. By coupling FY-4B satellite high-frequency land AOT observations with the 4D-LETKF assimilation algorithm, high-frequency observation information from domestic geostationary satellites can be directly and in real-time assimilated and utilized in regional models.

[0010] 2. By constructing a double-layered nested multi-scale aerosol assimilation framework, FY-4B observations can be directly used in the d02 high-resolution region to solve the problems of excessive resolution gap and loss of subgrid information in traditional coarse-resolution assimilation systems.

[0011] 3. By using dedicated observation preprocessing methods for FY-4B, including quality control rules, outlier removal, and gridding, and adapting to the geometric and inversion characteristics of the FY-4B geostationary satellite, it can capture rapid processes such as dust storm movement, formation and dissipation, and diurnal variations in pollution, thereby improving the ability to more realistically depict the evolution of atmospheric pollutants and provide more accurate forecasts.

[0012] 4. By employing a log-space differential perturbation strategy for multiple emission sources to generate a membership set with physically reasonable prior uncertainty, and constructing a flow-dependent background error covariance, we can significantly improve the analysis accuracy of the aerosol initial field by the data assimilation system, thereby improving the quality of short-term forecasts and providing more realistic uncertainty information for ensemble forecasts. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating a multi-scale assimilation prediction method for FY-4B satellite aerosol observation data. Figure 2 This is a timing diagram of the 4D-LETKF assimilation window; Figure 3 This is a comparison chart of key indicators (BIAS, RMSE, CORR, IOA) for FR and DA in the d01 / d02 region. Detailed Implementation

[0014] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0015] like Figure 1 As shown, in one embodiment of the present invention, the multi-scale assimilation prediction method for FY-4B satellite aerosol observation data includes the following steps: S1. Obtain land aerosol optical thickness observation data retrieved by the AGRI sensor carried by the FY-4B geostationary meteorological satellite.

[0016] Specifically, it includes: Data Reading and Decoding: FY-4B L2-level land AOT (Aerosol Optical Thickness) products are read in batches by time slice, extracting pixel center longitude, pixel center latitude, 550nm land AOT retrieved values, Data Quality Flag (DQF), and observation time (UTC). The FY-4B L2-level land AOT product is an atmospheric aerosol optical thickness product retrieved from AGRI sensor observation data. It features high temporal resolution (30 minutes / 15 minutes / 1 minute) and high spatial resolution (2-4 kilometers), making it a core data source for monitoring dust storms and other processes.

[0017] Quality control: Pixels with a DQF value of 3 are selected based on the Data Quality Flag (DQF) bit, and the remaining pixels are removed.

[0018] Outlier removal: Pixels with an aerosol optical thickness of less than 0.02 are further removed from the pixels after quality control (DQF=3, i.e., 'high quality / best quality').

[0019] Gridding: After outlier removal, the aerosol optical thickness value of each remaining pixel is mapped to each grid point of WRF-Chem through projection transformation and spatial interpolation, resulting in a gridded aerosol optical thickness observation field. The model grid point is the smallest computational unit of the WRF-Chem model in the Lambert projection coordinate system. The WRF-Chem model is a coupling of the WRF (Weather Research and Forecasting) model and the atmospheric chemistry module, employing an online coupling design to simultaneously simulate meteorological processes and the transport, transformation, and deposition of chemical substances.

[0020] In this embodiment, each grid point of d01 corresponds to a small square of 36 km × 36 km, and each grid point of d02 corresponds to a horizontal square of 4 km × 4 km. d01 uses area-weighted averaging, while d02 uses nearest-neighbor interpolation to match at near-native resolution.

[0021] The observation error for each grid point is calculated using the following expression:

[0022] In the formula, This is the observation error; This is the inversion error; The representative error is the standard deviation of the aerosol optical thickness values ​​of all valid FY-4B pixels within the grid, reflecting the spatial variability within the grid.

[0023] Among them, inversion error Linearly correlated with the magnitude of AOT: .

[0024] Output a gridded aerosol optical thickness observation field and observation error.

[0025] S2. A double-layer nested model is constructed based on the WRF-Chem regional weather-chemical coupling model, and the ensemble background field is obtained based on the ensemble emission perturbation strategy.

[0026] A two-layer nested model was constructed based on the WRF-Chem regional weather-chemical coupling model, including: Two regions with different resolutions are defined: the outer region d01 and the inner high-resolution region d02. In this embodiment, the outer region (d01) covers most of China and surrounding areas, using Lambert projection, with a horizontal resolution of 36 km, a grid number of 100×81, 45 vertical layers, a model top of 50 hPa, and a time step of 15 s. d01 is mainly used for large-scale transport and weather processes, providing boundary conditions for d02. The inner high-resolution region (d02) is a nested sub-region within d01, primarily covering Shandong Province and adjacent areas, with a horizontal resolution of 4 km, a grid number of 199×127, and the same 45 vertical layers as d01, with a time step of 5 s. d02 is the key region for assimilation verification, capable of resolving local aerosol pollution structures.

[0027] An initial field for the outer region d01 is generated, and spatial interpolation is used to fill each fine grid point of the inner high-resolution d02 with this initial field. At each time step of the simulation, the meteorological evolution of the outer region d01 is calculated and transmitted in real time to the inner high-resolution region d02 as its boundary conditions, resulting in a double-layer nesting. That is, both the initial meteorological field and boundary conditions of d02 are provided by d01, ensuring the constraint of the high-resolution region field by the large-scale background flow.

[0028] The ensemble background field is obtained based on the ensemble emission perturbation strategy, including: Obtain a baseline emission field for the same experiment, and independently apply random perturbations to the seven emission sources in the baseline emission field in logarithmic space to generate k sets of perturbed emission fields.

[0029] The specific emission sources and disturbance magnitudes are shown in Table 1, and the disturbance magnitudes are set differently based on their prior uncertainties.

[0030] Table 1

[0031] In Table 1, E SO2 Indicates carbon dioxide emissions; E BC Indicates black carbon emissions; E OC Indicates organic carbon emissions; E PM2.5 This indicates emissions of fine particulate matter with a diameter of 2.5 μm or less; E PM10 This refers to inhalable particulate matter with a diameter of 10 μm or less; E DUST Indicates dust emissions; E SS This indicates the emission of sea salt aerosols.

[0032] The aforementioned perturbations maintain a certain autocorrelation in the time dimension. Each member generates a ensemble background field of aerosol concentration, AOT, and meteorological variables based on the perturbation-independent emission integral WRF-Chem. The specific data processing procedure is as follows: Let the baseline emission of the s-th emission source control experiment be... , (s = SO2, BC, OC, PM2.5, PM 10 (DUST, SS), for the first Members ( = 1, 2, …, 20), the disturbance emissions of the s-th emission source are:

[0033] In the formula, For the disturbance emissions from the s-th emission source; For the first Control test baseline emissions for Class II emission sources; It is an exponential function with the natural constant e as the base; For the first Random perturbations imposed by each member, , For the first The logarithmic spatial standard deviation of a type of emission represents the magnitude of the prior uncertainty of that type of emission source.

[0034] Each set of disturbance emission fields is simultaneously assigned to the outer region d01 and the inner high-resolution region d02, respectively, as emission inputs for the corresponding domains.

[0035] Based on the grid settings of d01 and d02, a WRF-Chem double-nested integral simulation is performed simultaneously to obtain the meteorological-chemical coupled simulation results for k ensemble members. This yields an ensemble background field with a flow-dependent background error structure, including the three-dimensional aerosol field, aerosol composition, aerosol optical thickness, and meteorological variables for each ensemble member. The ensemble standard deviation characterizes the magnitude of the model background error and will be directly utilized through the calculation of the ensemble background error covariance in subsequent 4D-LETKF assimilation analysis.

[0036] The 3D aerosol field of each ensemble member refers to the spatial distribution data of the mass concentration (or mixing ratio) of each aerosol species (such as sulfate, nitrate, black carbon, organic carbon, dust, sea salt, etc.) at all vertical layers and all horizontal grid points within the simulation region. The 3D aerosol field of each ensemble member is the result of simulation calculation by the WRF-Chem model under specific input conditions through its internal physical and chemical processes.

[0037] In this embodiment, the specific settings for the WRF-Chem mode are shown in Table 2: Table 2

[0038] Specific explanations of the above configuration parameters: The double-layered nested region (d01: 36 km, d02: 4 km) integrates large-scale transport fields and local aerosol structures into the same model framework, determining the accuracy of the assimilation system's background field across multiple scales. The time step settings (d01: 15 s, d02: 5 s) are based on CFL stability conditions to ensure numerical stability at different resolutions. NCEP FNL reanalysis data, used as initial boundary conditions, provides the initial meteorological field and boundary conditions, and is a widely used standard atmospheric reanalysis data in regional aerosol model research.

[0039] In this embodiment, the physical and chemical schemes in the WRF-Chem mode are shown in Table 3: Table 3

[0040] In Table 3, the numbers in parentheses are the corresponding parameters and option numbers in WRF-Chem v4.4 namelist.input. For example, mp_physics=28 is the configuration parameter name of the microphysics scheme, and 28 corresponds to the Thompson aerosol-aware scheme.

[0041] The selection criteria and functions of the above physicochemical schemes are as follows: The Thompson aerosol-aware microphysics scheme is used, which explicitly considers the influence of aerosols on cloud microphysical processes and helps establish aerosol-cloud-radiation coupling; the RRTMG scheme is used for both long-wave and short-wave radiation, which can calculate the direct effects of aerosols on radiative transfer; the YSU boundary layer scheme is used, suitable for regional aerosol simulation at medium resolution; the Revised MM5 Monin-Obukhov scheme, paired with YSU, is used for the near-surface layer to ensure accurate calculation of surface fluxes and atmospheric exchange; the GOCART aerosol chemistry scheme is used, which includes five main aerosol components: black carbon, organic carbon, dust, sulfate, and sea salt, corresponding to the mass mixing ratio state variables in WRF-Chem. Anthropogenic emissions are represented by the HTAP high-resolution anthropogenic emission inventory, and biomass combustion emissions are represented by the FINN inventory; both constitute the main body of the model emission inputs.

[0042] S3. Based on the observation data of land aerosol optical thickness and the ensemble background field, 4D-LETKF assimilation analysis was performed to obtain the ensemble analysis field.

[0043] The background field of the set generated in step S2 is as follows Figure 2As shown, within the assimilation time window, the state vectors of k ensemble members are output every 15 minutes, including aerosol component concentration, AOT and related meteorological variables, forming an ensemble background field for calculating background error and updating the state of the assimilation system; Figure 2 middle This is the transformation matrix in 4D-LETKF, used to transform the background set into the analysis field set.

[0044] Generating simulated observations: Based on the three-dimensional aerosol field for each ensemble member, the simulated aerosol optical thickness is calculated using a forward observation operator, specifically: The mass concentration of aerosols is calculated using the following expression: In the formula, For the first Layer Mass concentration of aerosol-like particles (μg / m³); For the first Layer Mass mixing ratio of aerosol-like particles (kg / kg); The density of dry air (kg / m³); The total extinction coefficient of a single layer is calculated using the following expression: In the formula, For the first The total extinction coefficient of the layer at 550 nm; For the first The mass extinction coefficient of aerosols at 550 nm (m² / g, which is related to aerosol particle size, density, complex refractive index, and hygroscopic growth, etc.). The simulated aerosol optical thickness is calculated using the following expression: In the formula, This represents the simulated 550 nm column aerosol optical thickness, which is the amount directly compared to the FY-4B AOT observation. Indicates the forward observation operator; For the first Vertical thickness of the layer. Calculate the simulated AOT and sample according to the observation time to form an ensemble observation estimate for 4 time periods within the window.

[0045] Localization: For each grid point on WRF-Chem, the localization weight of each FY-4B observation centered at that grid point and within the horizontal localization radius is calculated using a Gaussian decay function. The expression is as follows: In the formula, As localized weights, these weights will be layered and modulated on the observation error covariance matrix R in the subsequent 4D-LETKF transformation matrix solution to reduce the impact of long-distance spurious correlations on the analysis increment. The horizontal distance between the observation location and the analysis grid point; The radius is the horizontal localization radius.

[0046] In the observation subspace of each local analysis point, solve for the 4D-LETKF transformation matrix W to ensure that the mean and bias of the analysis set are optimally fitted to the observations within the observation window, while maintaining an appropriate variance in the ensemble spectrum. Specific calculation process: Assume the assimilation window contains m+1 discrete time points. The i-th member in Background observations at time are estimated as follows The background observation perturbations at all times are vertically stacked to obtain the joint background observation perturbation matrix Y4. b Joint observation error covariance .

[0047] The error covariance is calculated using the following expression: In the formula, For error covariance; The number of members in the set; It is the identity matrix; For the joint background observation perturbation matrix; The joint observation error covariance after localized weighting; The average weight vector is calculated using the following expression: In the formula, This is the average weight vector; For joint observation vectors; This represents the average simulated aerosol optical thickness. The mean is expressed as follows: In the formula, for The average value of the members of the set at any given time; The mean of the background disturbance; The background perturbation matrix; The calculation and analysis of set perturbation is expressed as follows: In the formula, express Analysis of the set perturbation at any given moment; The ensemble perturbation is mapped back to the WRF-Chem grid, updating the ensemble members and related variables such as aerosol concentration, wind field, and temperature. The ensemble analysis field for the assimilation time window is output as the initial field for the next assimilation time window; the ensemble analysis field for the assimilation time window is the average value of the ensemble members. Based on the updated ensemble members, the average value of the ensemble members for the next assimilation time window is calculated until the current window is the last assimilation time window, thus obtaining the ensemble analysis field.

[0048] The corresponding values ​​of some parameters in this embodiment are shown in Table 4: Table 4

[0049] S4. Based on the obtained ensemble analysis field, run the short-time WRF-Chem forecast to obtain the short-term aerosol forecast results.

[0050] Specifically, it includes: The ensemble analysis field is decomposed by members to construct 20 initial values ​​for WRF-Chem restart; Each ensemble member is independently driven to integrate simultaneously over the nested regions d01 and d02, and the aerosol state of each ensemble member is output at fixed time intervals. The ensemble average forecast result is obtained by averaging the aerosol states of all ensemble members.

[0051] In this embodiment, the specific calculation process of the ensemble forward integral is as follows: k=20 members of the analysis ensemble obtained from 4D-LETKF assimilation analysis. As the initial field, each member Independently driven WRF-Chem was simultaneously integrated over a double-layered nested region of d01 (36 km) and d02 (4 km) for 5 h. During the integration, all members used the same emission field (unperturbed controlled emissions) and meteorological boundary conditions (taken from d01), differing only in their initial aerosol concentration fields (from ensemble analysis). The aerosol state of each member was output every 15 minutes, including the mass mixing ratio qs,l of each GOCART component (black carbon, organic carbon, dust, sulfate, sea salt) and the corresponding 550 nm AOT. Finally, the results were obtained from... +15min to The ensemble forecast fields for 20 time intervals over +5 hours are used to obtain the ensemble average forecast by averaging over all members. The ensemble divergence reflects the forecast uncertainty. The k members are independently integrated over 5 hours, with an output frequency of no less than 15 minutes, to obtain the ensemble aerosol forecast (AOT, component mass mixing ratio).

[0052] The ensemble average forecast results are divided into several independent time periods by hour for segmented verification; the data is further divided into +1h, +2h, +3h, +4h, and +5h (EXP1-EXP5) to support targeted verification and business product generation.

[0053] Based on the ensemble average forecast results after segmented testing, two sets of gridded forecast fields, d01 and d02, are output.

[0054] To verify the beneficial effects of this invention, during the experimental period from November 22nd to 30th, 2025, the assimilation analysis effect of this invention (DA) was compared with that of the reference experiment (FR) using FY-4B AOT during the same time period as a reference. The experimental area covered China and surrounding East Asia, and a double-layer nested WRF-Chem v4.4 was used. The assimilation window length was 1 hour, and FY-4B land AOT entered the system at a frequency of 15 minutes within the window. Comparative experiment: FR is the pure model forecast, and DA is the assimilation result of this invention. Validation data included FY-4B AOT during the same period and 14 AERONET sites. Comparison of key indicators (BIAS, RMSE, CORR, IOA) between FR and DA in the d01 / d02 region is as follows: Figure 3 As shown in Table 5, the comparison results are as follows: Table 5

[0055] The improvement in high-resolution d02 is significantly greater than that in d01, indicating that the high-resolution region of the multi-scale assimilation framework of this invention can more accurately simulate the spatial distribution of aerosols.

[0056] Using 14 AERONET sites (AOT) as independent validation data, the present invention significantly reduced RMSE in 11 of these sites (approximately 78.6%), with the average RMSE decreasing from 0.175 to 0.151, an average reduction of 13.7%; the improvement was particularly significant in sites along the coast of South Korea and Japan.

[0057] Meanwhile, analytical field component diagnosis showed that assimilation corrected for the systematically underestimated organic carbon (OC) and black carbon (BC) in the model: OC in the d01 region increased by approximately 59.8 mg / m³. 2 BC increased by approximately 21.7 mg / m³ 2 The OC increase in the d02 region was approximately 170 mg / m³. 2 This result corrects the original model's systematic underestimation of anthropogenic and combustion sources in the east.

[0058] The accuracy of the final short-term forecast was improved, as shown in Table 6: Table 6

[0059] As shown in the table above, the correlation coefficient of d02 EXP1 increased from 0.391 to 0.655 (↑67.5%), indicating a significant enhancement in the short-term forecast capability of high-resolution regions. The RMSE of d02 EXP1 decreased from 0.183 to 0.144 (↓21.3%), while the CORR improved by 67.5%. At EXP5, it still outperformed FR, demonstrating that the high-frequency assimilation analysis field performs well in short-term forecasts over 3-5 hours.

[0060] In a typical dust storm event in November 2025, the FR experiment failed to capture significant changes in AOT over a 3-hour period; however, the DA experiment of this invention showed that the peak AOT was 0.50-0.70 higher than that of FR, and its temporal and spatial consistency with the dust storm passage time observed by ground-based PM10. This demonstrates that this invention has the ability to simulate rapid hourly evolution processes of aerosols.

[0061] In summary, this invention establishes a multi-scale assimilation and forecasting system for FY-4B satellite aerosol observation data based on four-dimensional local ensemble transform Kalman filtering (4D-LETKF). Through the coordinated operation of four modules—observation data preprocessing, model ensemble forecasting, 4D-LETKF assimilation analysis, and short-term forecast output—it achieves high spatiotemporal resolution analysis and short-term forecasting of regional aerosols, significantly improving the simulation capability for high-resolution spatial distribution characteristics and rapidly changing processes on hourly scales.

Claims

1. A multi-scale assimilation prediction method for FY-4B satellite aerosol observation data, characterized in that, include: Acquire land aerosol optical thickness observation data retrieved by the AGRI sensor carried by the FY-4B geostationary meteorological satellite; A two-layer nested model was constructed based on the WRF-Chem regional weather-chemical coupling model, and the ensemble background field was obtained based on the ensemble emission perturbation strategy; 4D-LETKF assimilation analysis was performed based on terrestrial aerosol optical thickness observation data and ensemble background field to obtain the ensemble analysis field; Based on the obtained ensemble analysis field, short-time WRF-Chem forecasts were run to obtain short-term aerosol forecast results.

2. The method according to claim 1, characterized in that, Acquire land aerosol optical thickness observation data retrieved by the AGRI sensor aboard the FY-4B geostationary meteorological satellite, including: Data reading and decoding: Batch reading of FY-4B L2-level land AOT products by time slice, extracting pixel center longitude, pixel center latitude, 550nm land AOT retrieval inversion value, data quality flag (DQF) bit and observation time; Quality control: Pixels with DQF values ​​set to a preset value are filtered based on the Data Quality Flag (DQF) bit, and the remaining pixels are discarded; Outlier removal: Based on the pixels after quality control, pixels with aerosol optical thickness less than the preset value are removed; Gridding: After removing outliers, the aerosol optical thickness value of each remaining pixel is mapped to each grid point of WRF-Chem through projection transformation and spatial interpolation to obtain a gridded aerosol optical thickness observation field; where the aerosol optical thickness value is the 550nm land AOT retrieved and inverted value. The observation error for each grid point is calculated using the following expression: In the formula, This is the observation error; This is the inversion error; The representative error is the standard deviation of the aerosol optical thickness values ​​of all valid FY-4B pixels within the grid. Output a gridded aerosol optical thickness observation field and observation error.

3. The method according to claim 2, characterized in that, A two-layer nested model was constructed based on the WRF-Chem regional weather-chemical coupling model, including: Define two regions with different resolutions, namely the outer region d01 and the inner region d02; the resolution of the inner region is higher than that of the outer region. Generate the initial field of the outer region d01, and fill each grid point of the inner region d02 with the initial field of the outer region through spatial interpolation; At each time step of the simulation, the meteorological evolution of the outer region d01 is calculated and transmitted in real time to the inner region d02 as the boundary condition of the inner region d02, resulting in a double-layer nesting.

4. The method according to claim 3, characterized in that, The ensemble background field is obtained based on the ensemble emission perturbation strategy, including: Obtain a baseline emission field for the same experiment, and independently apply random perturbations to multiple emission sources in the baseline emission field in logarithmic space to generate k sets of perturbation emission fields; Each set of disturbance emission fields is simultaneously assigned to the outer region d01 and the inner region d02, which serve as the emission inputs for the corresponding domains. Based on the grid settings of d01 and d02, a WRF-Chem double-layer nested integral simulation is performed simultaneously to obtain the meteorological-chemical coupled simulation results of k ensemble members, that is, to obtain the ensemble background field with flow-dependent background error structure, including the three-dimensional aerosol field, aerosol composition, aerosol optical thickness and meteorological variables of each ensemble member.

5. The method according to claim 4, characterized in that, The method for calculating the disturbance emissions of each type of emission source is as follows: In the formula, For the disturbance emissions from the s-th emission source; For the first Control test baseline emissions for Class II emission sources; It is an exponential function with the natural constant e as the base; For the first Random perturbations imposed by each member, , For the first The logarithmic spatial standard deviation of a type of emission represents the magnitude of the prior uncertainty of that type of emission source.

6. The method according to claim 5, characterized in that, Based on terrestrial aerosol optical thickness observation data and ensemble background field, 4D-LETKF assimilation analysis was performed to obtain the ensemble analysis field, including: Generating simulated observations: Based on the three-dimensional aerosol field for each set member, the forward observation operator is used to calculate the simulated aerosol optical thickness; Localization: For each grid point on WRF-Chem, the localization weight of each FY-4B observation centered at that grid point and within the horizontal localization radius is calculated using a Gaussian decay function. The expression is as follows: In the formula, For localized weighting; The horizontal distance between the observation location and the analysis grid point; The horizontal localization radius; The error covariance is calculated using the following expression: In the formula, For error covariance; The number of members in the set; It is the identity matrix; For the joint background observation perturbation matrix; The joint observation error covariance after localized weighting; The average weight vector is calculated using the following expression: In the formula, This is the average weight vector; For joint observation vectors; This represents the average simulated aerosol optical thickness. The mean is expressed as follows: In the formula, for The average value of the members of the set at any given time; The mean of the background disturbance; The background perturbation matrix; When the current assimilation time window is not the last assimilation time window: The calculation and analysis of set perturbation is expressed as follows: In the formula, express Analysis of the set perturbation at any given moment; The analysis set perturbation is mapped back to the grid of WRF-Chem, and the set members are updated; the average value of the set members in the current assimilation time window is output as the initial field for the next assimilation time window; The average value of the set members for the next assimilation time window is calculated based on the updated set members, and this process continues until the current window is the last assimilation time window, thus obtaining the set analysis field.

7. The method according to claim 6, characterized in that, Based on the three-dimensional aerosol field for each set member, the simulated aerosol optical thickness is calculated using a forward observation operator, including: The mass concentration of aerosols is calculated using the following expression: In the formula, For the first Layer Mass concentration of aerosol-like particles; For the first Layer Mass mixing ratio of aerosols; The density of dry air; The total extinction coefficient of a single layer is calculated using the following expression: In the formula, For the first The total extinction coefficient of the layer at 550 nm; For the first Mass extinction coefficient of aerosol-like particles at 550 nm; The simulated aerosol optical thickness is calculated using the following expression: In the formula, This represents the simulated optical thickness of a 550 nm column aerosol. Indicates the forward observation operator; For the first Vertical thickness of the layer.

8. The method according to claim 7, characterized in that, Based on the obtained ensemble analysis field, short-time WRF-Chem forecasts were run to obtain short-term aerosol forecast results, including: The ensemble analysis field is decomposed by members to construct k initial values ​​for WRF-Chem restart; Each set member is independently driven to integrate simultaneously on the double-nested regions d01 and d02, and the aerosol state of each set member is output at fixed time intervals. The aerosol states of all ensemble members are averaged to obtain the ensemble average forecast result; The ensemble average forecast results are divided into several independent time periods by hour for segmented testing. Based on the ensemble average forecast results after segmented testing, two sets of gridded forecast fields, d01 and d02, are output.

9. A system for implementing the multi-scale assimilation and prediction method for FY-4B satellite aerosol observation data as described in any one of claims 1 to 8, characterized in that, include: The observation data preprocessing module is used to acquire land aerosol optical thickness observation data retrieved by the AGRI sensor carried by the FY-4B geostationary meteorological satellite; The model ensemble forecast module is used to construct a two-layer nested model based on the WRF-Chem regional weather-chemical coupling model and obtain the ensemble background field based on the ensemble emission perturbation strategy. The 4D-LETKF assimilation analysis module is used to perform 4D-LETKF assimilation analysis based on terrestrial aerosol optical thickness observation data and ensemble background field to obtain the ensemble analysis field; The short-term forecast output module is used to run short-time WRF-Chem forecasts based on the obtained ensemble analysis field to obtain short-term aerosol forecast results.

Images