A modeling method of multi-satellite coordinated spectral radiation observation migration conversion model

Abstract

The application discloses a modeling method of a multi-satellite coordinated spectral radiation observation migration conversion model, first, an input parameter set is acquired and configured, and a forward simulation calculation is performed by using a MODTRAN atmospheric radiation transmission model to obtain hyperspectral radiation brightness data; then, normalized spectral response functions of different generations of multi-spectral sensors in selected satellites are loaded into the MODTRAN atmospheric radiation transmission model, or the hyperspectral radiation brightness data is post-processed and convolved and integrated, and channel radiation brightness values of each channel of the selected satellite under the same input observation scene are output; subsequently, paired analysis is performed to obtain a paired sample set and construct a radiation migration relationship model for training and precision verification; finally, the radiation migration relationship model is applied to selected satellite remote sensing observation data, observation data radiation benchmarks between different generations of multi-spectral sensors of the selected satellite are unified, and a long-time-series radiation data set with time sequence continuity and cross-platform / cross-generation consistency is generated.

Description

Technical Field

[0001] This invention belongs to the technical field of satellite remote sensing, specifically involving a modeling method for a multi-satellite collaborative spectral radiation observation migration and transformation model. It constructs a radiation response migration model between different generations of multispectral sensor channels of meteorological satellites to achieve observation radiation consistency and temporal comparability of radiation imaging remote sensing data from multiple generations of satellites. Background Technology

[0002] The Fengyun series of satellites is a meteorological remote sensing observation platform independently developed and operated by my country. It has formed a pattern of collaborative observation across multiple generations, orbit types (polar orbit, geostationary), and payload types (visible light, infrared, microwave, etc.). To achieve radiometric consistency and temporal comparability of remote sensing data among the multispectral sensors carried by different generations of Fengyun satellites, constructing cross-sensor transfer models has become a key technical requirement in satellite data calibration, fusion, and long-term climate product generation.

[0003] The construction of multi-generational sensor migration models mainly relies on two core technologies: cross-calibration and radiative transfer simulation. The former provides a theoretical framework for establishing transferable radiative response relationships between sensors, while the latter provides simulation support for compensating for differences in spectral response and observation conditions based on physical mechanisms. This model concept has been widely applied internationally and has gradually evolved into a standard methodology for establishing consistency among multi-source remote sensing data. Cross-calibration technology was first proposed by NOAA / NESDIS in the United States and has been systematically developed in the Global Space-based Inter-Calibration System (GSICS). GSICS uses high-precision radiation sensors such as Aqua / MODIS and IASI as reference sources, combined with a Simultaneous Nadir Overpass (SNO) strategy, to achieve calibration parameter transfer between satellites and across generations of sensors. This idea originated from the SNO calibration method proposed by Cao et al. in 2004, and subsequently, Wu et al. proposed a systematic radiative coefficient transfer strategy based on multi-source sensors, which was successfully applied to multiple operational satellite systems.

[0004] On the other hand, to bridge the systematic differences among different sensors in terms of spectral response function (SRF), observation geometry, orbital characteristics, and environmental conditions, radiative transfer models have been introduced to simulate remote sensing observations under ideal scenarios, thereby enhancing the physical consistency and universality of the transfer model. In this direction, MODTRAN and RTTOV are the most commonly used radiative transfer simulation tools internationally. As early as 1993, Gao et al. used MODTRAN to simulate brightness temperature under different atmospheric conditions to support the assessment of spectral consistency among sensors. In 2005, Schmidt et al. proposed the "Observation-Simulation Matching (OSM)" method, which systematically combines scene libraries with sensor response characteristics to construct reversible radiative response transfer models.

[0005] However, existing cross-calibration methods also have corresponding drawbacks. Some existing methods rely heavily on shared observations to build empirical regression models, lacking physical modeling of the causes of observational differences. This results in weak model generalization ability under varying geometric conditions, extreme scenarios, or in the early stages of new satellite deployment. Furthermore, traditional radiometric response conversion models often require extensive retraining with large amounts of on-orbit observation data when new satellites are deployed, leading to lengthy training times. Therefore, there is an urgent need for a method suitable for modeling radiometric observation response relationships and transferring calibration parameters between multispectral remote sensing sensors on different satellite platforms. Summary of the Invention

[0006] This invention aims to address the technical bottlenecks and practical needs in the current process of multi-source remote sensing data fusion and cross-platform radiometric calibration by providing a modeling method for multi-satellite collaborative spectral radiometric observation migration and transformation models. This method integrates radiative transfer simulation and sensor response function difference modeling, and has high versatility. It is applicable to the radiometric transformation modeling needs between various types of sensors, including geostationary orbit and polar orbit, infrared and visible and near-infrared. In particular, it is suitable for the currently popular machine learning methods, facilitating model migration between different satellites.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A modeling method for a multi-star collaborative spectral radiation observation migration and transformation model includes the following steps:

[0009] S1. Scene Parameter Configuration and Input: Obtain and configure the input parameter set for radiative transfer simulation; the input parameter set includes atmospheric state parameters, surface attribute parameters, observation geometric parameters, and observation scene;

[0010] S2. Radiative transfer simulation and spectral convolution: Input the input parameter set into the MODTRAN atmospheric radiative transfer model for forward simulation calculation to obtain hyperspectral radiance data;

[0011] S3. Radiance Calculation and Dataset Generation: The standardized spectral response functions of different generations of multispectral sensors from the selected satellites are loaded into the MODTRAN atmospheric radiative transfer model, or post-processed convolutional integrals are performed on the hyperspectral radiance data output by MODTRAN to output the simulated channel radiance values ​​for each channel of the selected satellite under the same input observation scenario; the channels include thermal infrared channels and visible-near-infrared solar reflection channels.

[0012] S4. Radiation migration modeling: Paired analysis is performed using simulated channel brightness temperature values ​​of various generations of multispectral sensors under the same observation scenario to obtain a paired sample set in the form of a benchmark sensor and a target sensor. A radiation migration model between different generations of multispectral sensors is constructed using statistical modeling methods, and the paired sample set is used for training and accuracy verification to obtain the final radiation migration model. The radiation migration model includes an empirical fitting model and a lookup table model. The empirical fitting model is a linear regression model or a higher-order polynomial regression model.

[0013] The lookup table model construction process is as follows: the lookup table is designed in the form of a multidimensional array, binning indexes are performed according to different angle ranges and different surface attributes in the simulation, and coefficients obtained by fitting through empirical models are integrated to generate more refined transformation coefficients; the corresponding migration model transfer coefficients are found by combining channel identifiers, angle range binning, and surface attribute codes.

[0014] S5. Cross-sensor calibration application: The final radiometric migration relationship model is applied to the selected satellite remote sensing observation data to unify the radiometric benchmark of observation data among different generations of multispectral sensors of the selected satellite, and generate a long-term radiometric dataset with temporal continuity and cross-platform / cross-generational consistency; the application methods include forward transformation and cross-calibration.

[0015] The forward conversion refers to converting historical observation data from one generation of multispectral sensors on the selected satellite to the radiation reference of another generation of multispectral sensors on the same selected satellite.

[0016] The cross-calibration refers to converting the observation data of the selected satellite multispectral sensor to the radiation reference of the selected reference satellite multispectral sensor.

[0017] As a preferred technical solution, the sources of the input parameter set include ERA5 or MERRA-2 reanalysis data, MODIS Land Cover database, or standard atmospheric profiles, and a representative atmospheric-surface input sample library is constructed through combined sampling.

[0018] The atmospheric state parameters include the vertical profiles of atmospheric temperature, atmospheric humidity, ozone concentration, atmospheric visibility, concentration profiles of other atmospheric components, aerosol optical thickness, and cloud parameters.

[0019] The surface attribute parameters include surface temperature, surface infrared emissivity, surface albedo, and surface cover type;

[0020] The observation geometric parameters include the solar zenith angle, the satellite observation zenith angle, and the relative azimuth angle between the sun and the satellite.

[0021] The observation scenarios include clear-sky ocean, clear-sky land, partial cloud coverage, and multi-layered cloud coverage;

[0022] The input parameter set is subjected to data quality control before being input into the MODTRAN atmospheric radiative transfer model, including parameter range checking, missing value handling, and format standardization.

[0023] As a preferred technical solution, obtaining hyperspectral radiance data specifically involves:

[0024] Parameter preprocessing: The input parameter set is converted according to the input format requirements of the MODTRAN atmospheric radiative transfer model, including hierarchical matching of atmospheric profile data, dimensional adaptation of surface attribute parameters, and unit unification of observation geometric parameters;

[0025] Model parameter configuration: Set the simulation core parameters in the MODTRAN atmospheric radiative transfer model, including: radiative transfer calculation mode, spectral resolution, atmospheric model, and underlying surface type;

[0026] Forward simulation calculation: The preprocessed input parameter set and the configured simulation core parameters are input into the MODTRAN atmospheric radiative transfer model to calculate the atmospheric radiative transfer process, and the initial hyperspectral radiance data is output.

[0027] Post-processing of hyperspectral data: Extract the initial hyperspectral radiance data, remove simulated outliers and perform wavenumber coordinate calibration to obtain hyperspectral radiance data with a wavenumber range covering the entire spectrum from visible light to thermal infrared and a wavenumber interval equal to the simulated resolution.

[0028] As a preferred technical solution, the convolution integral is expressed as:

[0029] ,

[0030] Where L(ν) is the hyperspectral radiance data at wavenumber ν output by the MODTRAN atmospheric radiative transfer model, SRF(ν) is the normalized spectral response function at wavenumber ν for a specific channel corresponding to the spectral characteristics of the multispectral sensor, and v min and vmax These represent the effective wavenumber ranges of the normalized spectral response function for that channel.

[0031] During the post-processing convolution integration, the wavenumber coordinates of the hyperspectral radiance data are aligned and interpolated with the defined coordinates of the normalized spectral response function of the multispectral sensor. Specifically:

[0032] Coordinate acquisition and comparison: The wavenumber sequence VL = [νl1, νl2, ..., νl] of the hyperspectral radiance data output by the MODTRAN atmospheric radiative transfer model was extracted. N The wavenumber sequence VS = [νs1, νs2, ..., νs] of the normalized spectral response function of the multispectral sensor. M By comparing the wavenumber ranges of the two wavenumber sequences, the intersection range [νl, νs] is determined, where N is the number of wavenumbers in the hyperspectral radiance data and M is the number of wavenumbers in the normalized spectral response function of the multispectral sensor.

[0033] Reference coordinate establishment: The wavenumber sequence VL of hyperspectral radiance data is used as the reference wavenumber coordinate;

[0034] SRF interpolation matching: The wavenumber sequence of the normalized spectral response function of the multispectral sensor is interpolated onto the reference wavenumber coordinates using the linear interpolation method to obtain the normalized spectral response function of the multispectral sensor after interpolation;

[0035] Post-interpolation verification: Calculate the integral value deviation of the standardized spectral response function of the multispectral sensor before and after interpolation. If the integral value deviation is greater than the set deviation threshold, switch to cubic spline interpolation and re-interpolate to ensure that the energy characteristics of the standardized spectral response function of the multispectral sensor remain unchanged after interpolation.

[0036] As a preferred technical solution, the pairing analysis specifically includes:

[0037] Dataset preprocessing: The simulated brightness temperature datasets from various generations of multispectral sensors are uniformly preprocessed; the preprocessing involves outlier removal.

[0038] Pairing rule establishment: Matching of multispectral instrument features between satellites based on the spectral characteristics of each channel of the selected satellite;

[0039] Pairing quality assessment: Analysis of paired multispectral instrument channels, including differences in spectral response functions and data reflecting the channel matching characteristics between different instruments.

[0040] As a preferred technical solution, the empirical fitting model is constructed as follows:

[0041] Model type selection: The model type is selected based on the scatter distribution characteristics of the paired sample set. If the scatter points are linearly distributed, a linear regression model is selected; if they are nonlinearly distributed, a 2nd-3rd order higher-order polynomial regression model is selected. The higher-order polynomial regression model is constrained by L2 regularization of the absolute value of the coefficients.

[0042] Model parameter solution: The corresponding algorithm is used to solve the model parameters. For linear regression models, the slope and intercept are solved by the least squares method; for higher-order polynomial regression models, the polynomial coefficients are solved by the partial least squares method.

[0043] Model accuracy validation: Samples are randomly selected from the paired sample set according to a set ratio as the validation set, and the rest are used as the training set; the empirically fitted model is trained and validated using the training set and the validation set, the predicted values ​​are calculated and evaluated using accuracy metrics; the accuracy metrics include the coefficient of determination, mean absolute error, root mean square error and bias.

[0044] Model optimization: Optimize the model based on accuracy metrics until the model accuracy requirements are met;

[0045] The process of constructing the lookup table model:

[0046] A lookup table is designed using a multidimensional array, and the coefficients obtained by fitting an empirical model are integrated to obtain the lookup table model.

[0047] As a preferred technical solution, the forward conversion specifically includes:

[0048] Sensor designation: Use a multispectral sensor of one generation of the selected satellite as the reference sensor, and another multispectral sensor of the selected satellite as the target sensor;

[0049] Historical data preprocessing: Historical observation data of the reference sensor L1B level data is acquired and preprocessed; the preprocessing includes quality control and spatiotemporal warping; the quality control is used to remove abnormal data from the historical observation data of the reference sensor and perform parameter range checks; the spatiotemporal warping is used to unify the historical observation data of the reference sensor to the spatial resolution of the target sensor and unify the time reference.

[0050] Model selection and parameter matching: Select the optimal radiation migration relationship model based on the complexity of the observation scenario; use the empirical fitting model to retrieve the regression equation coefficients by channel identifier, and use the lookup table model to retrieve the transformation coefficients by channel identifier + bin combination code;

[0051] Forward conversion calculation: Load the model parameters of the optimal radiative migration relationship model and convert the historical observation data of the reference sensor to the radiative reference of the target sensor;

[0052] Post-processing of the converted results includes accuracy verification, data format conversion, and outlier handling. Accuracy verification involves randomly selecting a set proportion of the converted data and comparing it with the measured data from the target sensor during the same period, calculating a unified accuracy index. If the accuracy requirements are not met, the model parameter matching process is re-checked. The unified accuracy index includes the coefficient of determination, mean absolute error, root mean square error, and bias. Data format conversion involves converting the converted data to the standard data format of the target sensor while retaining the metadata information of the original data corresponding to the converted data. Outlier handling involves removing outliers that exceed the physical range generated during the conversion process and supplementing them using spatial interpolation.

[0053] Results Output and Archiving: The transformed data is archived according to time series, and a transformation quality report is generated; the report includes model type, transformation accuracy index, and outlier ratio.

[0054] As a preferred technical solution, the cross-calibration specifically involves:

[0055] Sensor designation: Select a satellite or generational multispectral sensor with a known radiation reference and high calibration accuracy as the reference sensor, and use the target satellite or target generational multispectral sensor to be calibrated as the target sensor; the reference sensor and the target sensor are different generations of the same satellite or different satellites / generational sensors;

[0056] Target data preprocessing: Acquire and preprocess the observation data of the target sensor; the preprocessing includes quality control and spatiotemporal adjustment; the quality control is used to remove abnormal data from the target sensor observation data and perform parameter range checks; the spatiotemporal adjustment is used to unify the target sensor observation data to the spatial resolution of the reference sensor and unify the time reference.

[0057] Inverse model construction: Based on the final radiative migration relationship model, an inverse model between the reference sensor and the target sensor is constructed;

[0058] Cross-calibration calculation: The inverse model is used to transform the observation data of the target sensor to the radiation reference of the reference sensor to complete the cross-calibration;

[0059] Calibration result verification: The calibration results are compared with the data of the reference sensor during the same period, and a unified accuracy index is calculated to verify the accuracy during the same period. The unified accuracy index includes the coefficient of determination, mean absolute error, root mean square error, and deviation. The calibration results are merged with the historical long-term data of the reference sensor, and the time series continuity is analyzed to verify the long-term trend until the accuracy requirements are met.

[0060] Long-term time series dataset generation: The validated calibrated data is spliced ​​with the historical observation data of the reference sensor, sorted in chronological order and duplicate data is removed to generate a long-term time series dataset with a unified radiation reference, and data source information is added.

[0061] In addition, a modeling system for a multi-satellite collaborative spectral radiation observation migration and transformation model is provided, which is applied to the modeling method of the above-mentioned multi-satellite collaborative spectral radiation observation migration and transformation model. The system includes a parameter set configuration module, a simulation calculation module, a channel radiation processing module, a relationship modeling module, and a calibration application module.

[0062] The parameter set configuration module is used to acquire and configure the input parameter set for radiative transfer simulation; the input parameter set includes atmospheric state parameters, surface attribute parameters, observation geometric parameters, and observation scene;

[0063] The simulation calculation module is used to input the input parameter set into the MODTRAN atmospheric radiative transfer model for forward simulation calculation to obtain hyperspectral radiance data.

[0064] The channel radiation processing module is used to load the standardized spectral response functions of different generations of multispectral sensors in the selected satellites into the MODTRAN atmospheric radiative transfer model, or to perform post-processing convolution integration on the hyperspectral radiance data output by MODTRAN, and output the simulated channel radiance values ​​of each channel of the selected satellite under the same input observation scenario; the channels include thermal infrared channels and visible light-near infrared solar reflection channels;

[0065] The relationship modeling module is used to perform paired analysis on the channel brightness temperature values ​​of simulated multispectral sensors of different generations under the same observation scenario to obtain a paired sample set in the form of a benchmark sensor-target sensor; a radiation migration relationship model between multispectral sensors of different generations is constructed through statistical modeling methods, and the paired sample set is used for training and accuracy verification to obtain the final radiation migration relationship model; the radiation migration relationship model includes an empirical fitting model and a lookup table model; the empirical fitting model is a linear regression model or a higher-order polynomial regression model;

[0066] The calibration application module is used to apply the final radiation migration relationship model to the selected satellite remote sensing observation data, so as to unify the radiation benchmark of the observation data among different generations of multispectral sensors of the selected satellite, and generate a long-term radiation dataset with temporal continuity and cross-platform / cross-generational consistency; the application methods include forward transformation and cross-correction.

[0067] The forward conversion refers to converting historical observation data from one generation of multispectral sensors on the selected satellite to the radiation reference of another generation of multispectral sensors on the same selected satellite.

[0068] The cross-calibration refers to converting the observation data of the selected satellite multispectral sensor to the radiation reference of the selected reference satellite multispectral sensor.

[0069] In addition, the present invention also provides a computer-readable storage medium storing a program that, when executed by a processor, implements the modeling method for the above-mentioned multi-star cooperative spectral radiation observation migration and transformation model.

[0070] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0071] 1. Integrating physical constraints and data-driven mechanisms to improve the physical consistency and stability of the conversion: Existing cross-calibration methods mainly rely on common observations to build empirical regression models, lacking physical modeling of the causes of observational discrepancies. This results in weak model generalization ability under changing geometric conditions, extreme scenarios, or in the early stages of new satellite deployment. This invention introduces the MODTRAN atmospheric radiative transfer model for physical simulation, effectively compensating for the differences in the sources of observational inconsistencies and achieving a more physically plausible response alignment.

[0072] 2. Possesses rapid model transfer and cross-sensor adaptation capabilities: Traditional models often require a large amount of on-orbit observation data for retraining when a new satellite is launched. However, this invention establishes a parameter library through the simulation-observation relationship of existing satellites (such as FY-4A) and combines it with a small number of observation samples from the new satellite to quickly fit and form a transfer model, significantly reducing model training time and data dependence, and possessing stronger practicality and engineering adaptability.

[0073] 3. Ensuring the continuity and consistency of long-term climate observation sequences: This invention supports parameter migration between multiple FY series geostationary satellites, ensuring the radiometric comparability of observation data from different generations of sensors, facilitating the construction of long-term, stable, and consistent brightness temperature / reflectivity time series, and serving various high-consistency scenarios such as climate monitoring, ecological remote sensing, and land surface change analysis.

[0074] 4. The method is highly versatile and applicable to a variety of sensor types: This invention is not only applicable to the response relationship modeling between FY-4A and FY-4B / AGRI, but can also be extended to other polar-stationary sensors (such as MERSI and AGRI), and can even be used in international multi-platform cross-calibration scenarios. Attached Figure Description

[0075] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0076] Figure 1 This is a flowchart illustrating a modeling method for a multi-star collaborative spectral radiation observation migration and conversion model in an embodiment of the present invention.

[0077] Figure 2 This is a schematic diagram of the process for simulating radiative transfer in an embodiment of the present invention.

[0078] Figure 3 This is a schematic diagram of the process for synthesizing FY-4A / AGRI and FY4B AGRI data in an embodiment of the present invention.

[0079] Figure 4 This is a schematic diagram of the data preprocessing process of the FY-3 series MERSI instrument in an embodiment of the present invention.

[0080] Figure 5 This is a schematic diagram of the corresponding channel fitting results of two FY-4 series satellite instruments in an embodiment of the present invention.

[0081] Figure 6 This is a schematic diagram of the FY-3 series satellite channel fitting results in an embodiment of the present invention.

[0082] Figure 7 This is a schematic diagram illustrating the difference in spectral response between the FY-3D and FY-3F MESRI instruments in an embodiment of the present invention. Detailed Implementation

[0083] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative effort are within the scope of protection of the present application.

[0084] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application can be combined with other embodiments.

[0085] The technical terms used in this application are defined below. Other undefined terms should be understood in accordance with the general technical knowledge of those skilled in the art:

[0086] The European Centre for Medium-Range Weather Forecasts (ECMWF) is an international organization founded in 1975. Its main business includes developing global numerical weather prediction, Earth system monitoring and meteorological data storage, operating integrated forecasting systems and artificial intelligence forecasting systems, and providing medium-range weather forecast products to member countries.

[0087] Geostationary satellites typically refer to satellites that operate in a geosynchronous orbit (GSO) above the equator, approximately 35,786 kilometers above the Earth's surface. The unique characteristic of this orbit is that its orbital period is exactly equal to the Earth's rotation period (approximately 23.934 hours). From the ground, its orbital projection appears almost fixed at a single point on the equator, hence the term "stationary." This orbit allows geostationary satellites to continuously observe the same region, providing extremely high temporal resolution, making them particularly suitable for continuous monitoring of large-scale meteorological systems (such as typhoons, fronts, and convective systems).

[0088] Polar-orbiting satellites typically refer to satellites operating in near-Earth polar orbits (usually sun-synchronous orbits, SSOs). Their orbital altitude is typically between 700 and 850 kilometers, with an inclination of nearly 90 degrees, enabling them to cover the North and South Poles. Because their orbital paths are almost perpendicular to the equator, and the Earth's rotation causes the ground to gradually move from west to east, polar-orbiting satellites can complete global coverage within multiple orbital periods. They possess excellent spatial coverage capabilities and high spatial resolution, making them suitable for long-term, high-resolution monitoring and fine-scale analysis on a global scale.

[0089] MODTRAN (MODerate resolution atmospheric TRANsmission) is a radiative transfer model widely used in remote sensing, meteorology, aerospace, and defense. It is used to simulate the propagation of electromagnetic radiation in the atmosphere, especially the radiative transfer characteristics under different atmospheric compositions, observation conditions, and surface properties. The model was jointly developed by the U.S. Air Force Research Laboratory (AFRL) and Spectral Sciences Inc. The current version is MODTRAN6, which has higher spectral resolution, stronger aerosol simulation capabilities, and richer atmospheric database support.

[0090] Transfer Model: In fields such as remote sensing, machine learning, and cross-sensor calibration, it refers to a modeling method that transfers existing model knowledge or parameter structure from one data domain, sensor platform, or task scenario to another target domain. Its core concept is to use existing knowledge in the source domain to achieve rapid modeling, performance improvement, or reduced sample requirements in the target domain.

[0091] The Advanced Geosynchronous Radiation Imager (AGRI) is one of the main payloads carried on China's new generation of geostationary meteorological satellites, the FY-4 series. It is a multi-channel scanning imaging radiometer whose main mission is to achieve continuous high-frequency observation of the atmosphere, the Earth's surface, and clouds, meeting the needs of modern meteorological operations for mesoscale disaster monitoring, weather forecasting, convection identification, and atmospheric parameter inversion.

[0092] The Medium Resolution Spectral Imager (MERSI) is one of the main payloads independently developed by China and carried on the Fengyun-3 (FY-3) series of polar-orbiting meteorological satellites. It is dedicated to remote sensing observation of Earth's surface and atmospheric parameters. The instrument has multispectral and multichannel observation capabilities, covering the visible, near-infrared, mid-infrared, and thermal infrared bands. MERSI has several versions, with the current representative models being: MERSI-I: used on FY-3A / B satellites; and MERSI-II: used on FY-3C / D / E, featuring improved channel settings and calibration capabilities.

[0093] The Moderate Resolution Imaging Spectroradiometer (MODIS) is a crucial remote sensing instrument developed by NASA and carried on the Terra (AM-1) and Aqua (PM-1) Earth observation satellites. It is an important component of the global observation system. Its key features include: 36 spectral channels covering the visible to thermal infrared spectrum (0.4–14.4 μm); spatial resolutions of 250 m (2 channels), 500 m (5 channels), and 1 km (the remaining channels); continuous observation of land, ocean, and atmosphere; and wide applications in land cover classification, vegetation dynamics, cloud and aerosol characteristics, sea surface temperature, snow cover, and fire detection. Its long-term continuous observation capabilities make it a vital data source for international climate change and Earth system research. MODIS data serves as a crucial benchmark for multinational collaborations and calibrations, including those with GSICS, CEOS, NASA Earth Science, NOAA, and ESA.

[0094] This embodiment uses the Advanced Geostationary Radiometric Imager (AGRI) aboard Fengyun-4A and Fengyun-4B satellites as examples to illustrate in detail the modeling method for a multi-satellite collaborative spectral radiometric observation migration and conversion model proposed in this application. Figure 1 As shown, it includes the following steps:

[0095] S1. Scene Parameter Configuration and Input: Obtain and configure the set of input parameters for radiative transfer simulation, which includes atmospheric state parameters, surface attribute parameters, observation geometry parameters, and observation scene.

[0096] Furthermore, the input parameters and acquisition sources of this invention include, but are not limited to: reanalysis data such as ERA5 and MERRA-2, the MODIS Land Cover database, or standard atmospheric profiles (such as mid-latitude summer data), and a representative atmospheric-surface input sample library is constructed through combined sampling. Wherein:

[0097] Atmospheric state parameters include the vertical profile of atmospheric temperature, the vertical profile of atmospheric humidity (water vapor mixing ratio or relative humidity), the ozone concentration profile, atmospheric visibility, the concentration profiles of other atmospheric components, aerosol optical thickness (including type and vertical distribution), and cloud parameters (cloud cover, cloud top height, cloud optical thickness).

[0098] Surface properties include surface temperature, surface infrared emissivity (for thermal infrared channels), surface albedo (for solar reflection channels), and surface cover type (e.g., ocean, desert, vegetation, city, etc.).

[0099] The observation geometric parameters include the solar zenith angle, the satellite observation zenith angle, and the relative azimuth angle between the sun and the satellite.

[0100] The observation scenarios are designed to cover typical and diverse global observation conditions, including but not limited to clear-sky ocean, clear-sky land (desert, plateau, farmland, forest, city), partial cloud cover, and multi-layered cloud cover.

[0101] Before inputting the parameter set into the MODTRAN atmospheric radiative transfer model, data quality control is performed, including parameter range checking, missing value handling, and format standardization, to ensure that the parameters input into MODTRAN are valid and have reasonable physical meaning.

[0102] In this embodiment, the atmospheric profiles for atmospheric state parameters are selected from 83 standard atmospheric profiles provided by ECMWF, covering a variety of typical atmospheric states from the tropics and temperate zones to the polar regions, including different water vapor content, temperature profiles, ozone concentrations, etc., which effectively improves the regional representativeness and model generalization ability of subsequent MODTRAN simulations.

[0103] S2. Radiative Transfer Simulation and Spectral Convolution: The input parameter set is fed into the MODTRAN atmospheric radiative transfer model for forward simulation calculation to obtain hyperspectral radiance data (unit: W / (m²·sr·cm)). -1 )).

[0104] Furthermore, the steps for acquiring hyperspectral radiance data are as follows:

[0105] S21. Parameter Preprocessing: The quality-controlled input parameter set is transformed according to the input format requirements of the MODTRAN model. This includes hierarchical matching of atmospheric profile data (converting the pressure levels of the reanalysis data to the MODTRAN default 1km / 2km vertical resolution levels), dimensional adaptation of land surface attribute parameters (converting 2D land cover type data to single-point / regional average parameters that the model can recognize), and unit unification of observation geometric parameters (ensuring that zenith angle and azimuth angle are in degrees). This ensures the rationality of the input parameters, removes outliers, and supplements missing items.

[0106] S22. Model Parameter Configuration: Set the simulation core parameters in the MODTRAN atmospheric radiative transfer model, including: radiative transfer calculation mode (solar reflection mode / thermal radiation mode, select according to channel type), spectral resolution (set to 0.1 cm⁻¹). -1 The high spectral resolution covers the full response range of the multispectral sensor, atmospheric modes (select preset modes such as mid-latitude summer / winter, tropical, etc., according to the observation scenario, or load custom atmospheric profiles), and underlying surface type (select ocean / land / mixed underlying surface mode according to surface attribute parameters).

[0107] S23. Forward simulation calculation: Input the preprocessed input parameter set and configuration parameters into the MODTRAN atmospheric radiative transfer model to calculate the atmospheric radiative transfer process, and output the initial hyperspectral radiance data, including the values ​​of physical processes such as atmospheric absorption, scattering, and surface reflection of solar radiation, and atmospheric emission and transmission of thermal radiation.

[0108] S24. Hyperspectral Data Post-processing: Extract the initial hyperspectral radiance data output by the model, remove simulated outliers (such as abrupt noise points appearing in the spectral range, identified using the 3σ criterion and replaced with the mean of adjacent wavenumbers), and perform wavenumber coordinate calibration (using the standard wavenumber scale built into the model as a reference to perform linear correction on the wavenumber axis of the output data), finally obtaining wavenumbers covering the entire visible to thermal infrared spectrum (300-30000 cm⁻¹). -1 (Corresponding wavelengths 0.3-33μm), wavenumber interval is the analog resolution (analog resolution uses 0.1cm). -1 The value is the highest precision supported by MODTRAN and is suitable for this scenario to ensure accuracy. The hyperspectral radiance data L(ν) is used to effectively remove abnormal radiance values ​​that are outside the physical range and extract the hyperspectral radiance data of the effective band.

[0109] S3. Radiance Calculation and Dataset Generation: The standardized spectral response functions of different generations of multispectral sensors in the selected satellites are loaded into the MODTRAN atmospheric radiative transfer model, or the hyperspectral radiance data output by MODTRAN is post-processed and convolved to output the simulated channel radiance values ​​of each channel of the selected satellite under the same input observation scenario; the channels include thermal infrared channels and visible-near infrared solar reflection channels.

[0110] Furthermore, after acquiring hyperspectral radiance data, the standardized spectral response functions of different generations of multispectral sensors from the selected satellites can be loaded into the MODTRAN atmospheric radiative transfer model, outputting the simulated channel radiance values ​​for each channel of the selected satellite under the same input observation scenario. In this embodiment, the standardized spectral response functions of the AGRI instruments carried by FY-4A and FY-4B are loaded into the MODTRAN atmospheric radiative transfer model, outputting the simulated channel radiance values ​​for each channel of FY-4A and FY-4B under the same input observation scenario. Based on the channel type of the AGRI instrument, they are divided into two main categories: solar reflection channels (0.4-2.3 μm) and infrared thermal emission channels (3.7-14 μm). The former is mainly dominated by solar radiation and needs to consider the solar angle and surface reflectivity; the latter focuses more on atmospheric temperature distribution and surface thermal emission characteristics. The two categories require separate simulation strategies to ensure simulation accuracy and applicability.

[0111] On the other hand, the hyperspectral radiance data output by MODTRAN can be post-processed using convolution integrals to output the simulated channel radiance values ​​for each channel of the selected satellite under the same input observation scenario; the convolution integral is expressed as:

[0112] ,

[0113] Where L(ν) is the hyperspectral radiance data at wavenumber ν output by the MODTRAN atmospheric radiative transfer model, SRF(ν) is the normalized spectral response function at wavenumber ν for a specific channel corresponding to the spectral characteristics of the multispectral sensor, and v min and v max These represent the effective wavenumber range (i.e., the wavenumber interval where SRF(ν)>0) of the normalized spectral response function of the channel.

[0114] Furthermore, during post-processing convolution integration, the wavenumber (or wavelength) coordinates of the hyperspectral radiance data are aligned and interpolated with the defined coordinates of the normalized spectral response function of the multispectral sensor to minimize computational errors. Specifically:

[0115] Coordinate acquisition and comparison: The wavenumber sequence VL = [νl1, νl2, ..., νl] of the hyperspectral radiance data output from the MODTRAN atmospheric radiative transfer model was extracted. N (0.1cm interval) -1 The wavenumber sequence VS = [νs1, νs2, ..., νs] of the normalized spectral response function of the multispectral sensor. M (The interval is usually 1-5cm) -1 (This varies depending on the sensor model). Compare the wavenumber ranges of the two wavenumber sequences to determine the intersection range [νl, νs], where N is the number of wavenumbers in the hyperspectral radiance data, and M is the number of wavenumbers in the normalized spectral response function of the multispectral sensor. (See attached reference.) Figure 7 Taking the visible-near-infrared solar reflection channels of Fengyun-3 MERSI and AQUA MODIS as examples, if the wavelength range of the thermal infrared channel is too wide to be convenient to demonstrate, the same applies to FY4A ARGI and FY4B AGRI used as examples. It can be seen that channels that can perform spectral matching and then transmit spectral features need to have similar basic features, that is, have similar center wavelengths and bandwidths. Similar spectral response functions are a prerequisite for ensuring the smooth transmission of spectral features.

[0116] Reference coordinates were established: the wavenumber sequence VL of hyperspectral radiance data was used as the reference coordinates because it has higher resolution and can fully cover the response characteristics of SRF.

[0117] SRF interpolation matching: The wavenumber sequence of the normalized spectral response function of the multispectral sensor is interpolated onto the reference wavenumber coordinates using linear interpolation, ensuring a one-to-one correspondence between the coordinates of the two, and thus obtaining the normalized spectral response function of the multispectral sensor after interpolation.

[0118] Post-interpolation verification: Calculate the integral deviation of the standardized spectral response function (SRF) of the multispectral sensor before and after interpolation. If the integral deviation exceeds a set deviation threshold, switch to cubic spline interpolation and re-interpolate to ensure that the energy characteristics of the SRF remain unchanged after interpolation. In this embodiment, if the integral deviation > 1%, switch to cubic spline interpolation and re-interpolate to ensure that the energy characteristics of the SRF remain unchanged after interpolation.

[0119] Finally, the simulated channel radiance values ​​of each channel of FY-4A and FY-4B under the same input observation scenario are output through any of the above methods.

[0120] S4. Radiation migration modeling: Paired analysis is performed using simulated channel brightness temperature values ​​of different generations of multispectral sensors under the same observation scenario to obtain a paired sample set in the form of a benchmark sensor and a target sensor; a radiation migration model between different generations of multispectral sensors is constructed using statistical modeling methods, and the paired sample set is used for training and accuracy verification to obtain the final radiation migration model, including an empirical fitting model and a lookup table model.

[0121] Furthermore, step S4 aims to establish the response transfer function of the FY-4B / AGRI channel relative to the FY-4A / AGRI channel to support unified product output and cross-satellite calibration. Therefore, it first requires pairing analysis using the generated simulated brightness temperature datasets of various generations of multispectral sensors corresponding to the same physical scenario. Specifically:

[0122] S411. Dataset Preprocessing: The simulated brightness temperature datasets from various generations of multispectral sensors undergo unified preprocessing, primarily including outlier removal, to ensure the simulations possess realistic and reliable physical characteristics. In this embodiment, outlier removal employs box plotting to eliminate outliers from each dataset (removing samples beyond 1.5 times the interquartile range), preventing some radiance values ​​that do not conform to physical characteristics from affecting the regression results.

[0123] S412. Pairing rule establishment: Match the multispectral instrument features between satellites based on the spectral characteristics of each channel of the selected satellite. The rules mainly cover the center wavelength or center wavenumber and bandwidth of the multispectral channels.

[0124] S413. Analyze the paired multispectral instrument channels. The main analytical features include differences in spectral response functions and other data that can reflect the channel matching characteristics between different instruments.

[0125] Next, a model of radiative migration relationships between different generations of multispectral sensors was constructed using statistical modeling methods.

[0126] S421. For empirical fitting models, the construction method is as follows:

[0127] S4211. Model Type Selection: The model type is selected based on the scatter distribution characteristics of the paired sample set. If the scatter points exhibit a linear distribution, a linear regression model is selected; if they exhibit a non-linear distribution, a 2nd-3rd order higher-order polynomial regression model is selected. L2 regularization is used to constrain the absolute value of the coefficients to avoid overfitting. In this embodiment, a confidence level of α=0.05 indicates a linear distribution of the scatter points; the regularization parameter λ ranges from 0.01 to 0.1.

[0128] S4212. Model Parameter Solution: The model parameters are solved using the corresponding algorithm. For linear regression models, the slope and intercept are solved using the least squares method; for polynomial regression models, the polynomial coefficients are solved using partial least squares. Specifically:

[0129] For a linear regression model: the model form is y = a×x + b, where a is the slope and b is the intercept. It is solved using the least squares method, with the following formula: , Where n is the number of samples in the paired sample set. For higher-order polynomial regression models: the model form is y = a1 + a2x + … + a k x k (k=2 or 3), solve for the coefficients a1, a2, ..., a using partial least squares method. k This reduces the effects of multicollinearity among variables.

[0130] S4213. Model Accuracy Validation: Randomly select samples from the paired sample set as the validation set according to a set proportion, and use the remainder as the training set; use the training set and validation set to train and validate the constructed empirically fitted model, calculate the predicted values, and evaluate them using accuracy metrics; accuracy metrics include coefficient of determination R², mean absolute error MAE, root mean square error RMSE, and bias. In this embodiment, 20% of the validation samples are extracted from the paired sample set; for the accuracy metrics, the coefficient of determination R²... 2 To measure the explanatory power of the model, R² ≥ 0.9; the mean absolute error (MAE) is |y - y|. pre The average value of | is required, and the MAE should be less than 1 / 3 of the sensor's radiation accuracy. pre The predicted value is the value after applying the transfer model; the bias is (y - y pre The average value of ) is required to be |Bias| ≤ 1% of the average radiance of the channel.

[0131] S4214. Model Optimization: Optimize the model based on accuracy metrics until the required accuracy is met. Specifically, for high-order polynomial regression models, adjust the polynomial order (e.g., reduce from 3rd to 2nd or increase to 4th); introduce interaction terms (e.g., add an interaction term between observed geometric parameters and radiance values); finally, remove outliers from the training set and resolve the parameters until the accuracy meets the requirements.

[0132] S422, The process for constructing the lookup table model is as follows:

[0133] A multidimensional array-based lookup table is designed, with bins indexed according to different simulated angle ranges (including observation zenith angle 0°-60°, solar zenith angle 0°-60°, etc.) and different surface attributes (including ocean, desert, vegetation, city, ice and snow, etc.). Coefficients obtained through empirical model fitting are integrated to generate more refined conversion coefficients. The corresponding transfer model coefficients can be found by combining channel name, center wavelength, angle range bins, and surface attribute codes. This step saves the tedious calculation of coefficients through code and improves the applicability accuracy of the model under different observation conditions.

[0134] S5. Cross-sensor calibration application: The final radiometric migration relationship model is applied to the remote sensing observation data of the selected satellite to unify the radiometric reference of the observation data among different generations of multispectral sensors of the selected satellite, generating a long-term radiometric dataset with temporal continuity and cross-platform / cross-generational consistency. The application methods include forward conversion and cross-correction. Forward conversion refers to converting the historical observation data of one generation of multispectral sensors of the selected satellite to the radiometric reference of another generation of multispectral sensors of the same selected satellite. Cross-correction refers to converting the observation data of the multispectral sensors of the selected satellite to the radiometric reference of the multispectral sensors of the selected reference satellite.

[0135] Furthermore, the forward conversion specifically involves:

[0136] S511. Sensor designation: Select one generation of multispectral sensors from the chosen satellite as the reference sensor Sensor_O, and another generation of multispectral sensors from the chosen satellite as the target sensor Sensor_N.

[0137] S512. Historical Data Preprocessing: Acquire historical observation data from the reference sensor at L1B level and preprocess it. Preprocessing includes quality control and spatiotemporal adjustment. Quality control is used to remove abnormal data from the historical observation data of the reference sensor. Based on data quality identifiers (such as MODIS QC codes), pixels contaminated by clouds, shadows, sensor anomalies, etc. are removed, and parameter range checks are performed. Spatiotemporal adjustment is used to unify the historical observation data of the reference sensor to the spatial resolution of the target sensor (such as resampling 1km resolution data to 500m) and unify the time reference (such as converting to UTC time).

[0138] S513. Model selection and parameter matching: Select the optimal radiation migration relationship model based on the complexity of the observation scenario; for empirical fitting models, retrieve the regression equation coefficients by channel identifier, and for lookup table models, retrieve the transformation coefficients by channel identifier + bin combination code.

[0139] S514, Forward Conversion Calculation: Load the model parameters of the optimal radiation migration relationship model and convert the historical observation data of the reference sensor to the radiation reference of the target sensor.

[0140] S515. Post-process the converted results, including accuracy verification, data format conversion, and outlier handling. Accuracy verification involves randomly selecting a set proportion (e.g., 1%) of the converted data and comparing it with the measured data of the target sensor from the same period to calculate a unified accuracy index. If the accuracy requirements are not met, the model parameter matching process is re-checked. The unified accuracy index includes the coefficient of determination, mean absolute error, root mean square error, and bias. Data format conversion refers to converting the converted data into the standard data format of the target sensor (e.g., HDF-EOS format) while retaining the metadata information of the original data corresponding to the converted data (e.g., observation time, orbital number). Outlier handling involves removing outliers that exceed the physical range generated during the conversion process and supplementing them using spatial interpolation.

[0141] S516. Results Output and Archiving: Archive the transformed data by time series and generate a transformation quality report; the report includes model type, transformation accuracy indicators, and outlier ratio, etc.

[0142] Furthermore, the backward correction specifically refers to:

[0143] S521. Sensor Designation: Select a satellite or generational multispectral sensor with a known radiation reference and high calibration accuracy as the reference sensor (Sensor_O), and designate the target satellite or target generational multispectral sensor to be calibrated as the target sensor (Sensor_N). The reference sensor and the target sensor may be from the same satellite but different generations, or from different satellites but different generations. (See attached reference.) Figure 6The conversion results of the seven visible-near-infrared solar reflection channels of MERSI and MODIS are highly correlated. At the same time, the higher radiation standard of MODIS can be used as a reference for correction. The correction scheme can also preserve the original spectral characteristics and information of MERSI.

[0144] S522. Target data preprocessing: Acquire the observation data from the target sensor and preprocess it, including quality control and spatiotemporal curvature; the processing method is the same as described above.

[0145] S523. Inverse Model Construction: Since backward correction is the inverse process of forward transformation, an inverse model is constructed based on the final radiative transfer relationship model. If an empirical fitting model is used, and the forward model is y = a×x + b (x is Sensor_O data, y is Sensor_N data), then the inverse model is x = (y - b) / a (ensuring a≠0). If a lookup table model is used, a reverse lookup table needs to be constructed (using Sensor_N data as input, mapping to Sensor_O data).

[0146] S524, Cross-calibration calculation: Load the constructed inverse model to convert the observation data of the target sensor Sensor_N to the radiation reference of the reference sensor Sensor_O, and complete the cross-calibration across satellites / generations.

[0147] S525. Verification of Calibration Results: Compare the calibrated results with the concurrent data of the reference sensor, and calculate the unified accuracy index to verify the concurrent accuracy. The unified accuracy index includes the coefficient of determination, mean absolute error, root mean square error, and deviation. Then, merge the calibrated results with the historical long-term time series data of the reference sensor, analyze the time series continuity to verify the long-term trend, until the accuracy requirements are met. Among them, the comparison of concurrent data refers to calculating the root mean square error (RMSE) and coefficient of determination (R²) of the concurrent observation data of the reference sensor and the target sensor. 2 Based on the iterative correction of the index, R is required to 2 ≥0.95 (or set the accuracy threshold according to the actual scenario); Long-term trend verification refers to merging the calibrated observation data with the historical long-term time series data of the benchmark sensor, and analyzing the temporal continuity of the data to ensure the consistency of the trend described in scenarios such as climate monitoring.

[0148] S526. Generation of long-term time series dataset: The validated calibrated data is spliced ​​with the historical observation data of the reference sensor, sorted in chronological order and duplicate data is removed (such as the mean of overlapping data from the same period) to generate a long-term time series dataset of a unified radiation reference, and data source information is added (such as the model version used for calibration and the source of parameters).

[0149] This embodiment is based on the underlying surface modeling capabilities supported by MODTRAN version 5.3. Taking into account the differences in reflectivity, thermodynamic properties, and spectral response of typical surface types, the following typical underlying surface types are preferred for combination configuration:

[0150] Ocean Water: Characterizes large-area water scenes with low reflectivity and significant angle dependence; the underlying surface has low reflectivity and is significantly affected by wind speed, making it particularly suitable for simulating the observation response of coastal areas, tropical ocean surfaces, and typhoon formation areas.

[0151] Desert and bare land surfaces: These are highly reflective, arid regions with strong surface thermal radiation characteristics. They are often used to simulate hot and dry continental scenarios, such as typical observation environments in North Africa, Central Asia, and the Australian outback.

[0152] Vegetation / Field / Forest Covered Areas: Moderate reflectance, exhibiting distinct spectral absorption bands, such as the vegetation jump characteristics between 0.66 μm and 0.86 μm; suitable for scenarios such as mid-to-low latitude terrestrial areas and atmospheric parameter inversion accuracy assessment.

[0153] Urban building surface: It has mixed material coverage, high surface roughness, and complex thermal response, which can be used to simulate the satellite's multi-angle observation response to high-density urban clusters.

[0154] Snow / Ice Covered Areas: High albedo, high scattering, and low thermal emission, especially with significant differences between shortwave infrared and thermal infrared bands, making them suitable for high-latitude polar monitoring areas.

[0155] All the aforementioned underlying surface types were implemented using the SURFACE REFLECTIVITY parameter configuration in MODTRAN 5.3, combined with standard surface templates or custom reflectance spectrum input methods in the SURFACE TYPE option for modeling. Each surface type was combined with different atmospheric profiles (such as tropical, mid-latitude summer / winter, polar, etc.) for combined simulation, forming a broad-coverage input sample library containing multi-dimensional space (latitude zone, underlying surface type, observation geometry, etc.), which significantly enhanced the robustness and universality of the fitted model in cross-channel transformations.

[0156] Furthermore, to further improve the physical accuracy of the simulation, this application also adopts the following strategies in the MODTRAN input configuration:

[0157] Viewing Geometry Sampling: Separate viewing grids are designed for the solar reflection channel and the thermal infrared channel (e.g., observation zenith angle 0°–60°, solar zenith angle 0°–60°) to fully cover a variety of observation scenarios.

[0158] Dynamic adjustment of surface albedo: For heterogeneous underlying surface scenarios (such as urban areas and water body edges), an adjustable reflectance spectrum (via a user-defined reflectance table .rfl file) is introduced to improve the modeling capability for high-contrast boundary regions.

[0159] Although this application uses clear sky (No Clouds) conditions in most basic modeling to ensure that the fitted model focuses on the basic radiative transfer processes of the Earth-atmosphere system, considering that clouds and aerosols have a non-negligible interference with the sensor response in actual observations, this application also extends the model to include MODTRAN simulation modeling with different types of clouds and aerosols to enhance the model's generalization ability and adaptability to complex scenarios.

[0160] In practical implementation, the following aspects should be considered for configuration and simulation:

[0161] Simulation of typical cloud types:

[0162] Based on the cloud model building capabilities supported by MODTRAN 5.3, the following typical cloud layers are configured for comparative simulation:

[0163] Low-level clouds (such as stratocumulus): with a cloud base height of about 1–2 km, a relatively large optical thickness, and a significant impact on the solar reflection channel.

[0164] Mid-level clouds (such as altocumulus): with a cloud base of about 3–5 km, can affect both visible and infrared wavelengths.

[0165] High-altitude clouds (such as cirrus clouds): The cloud top can reach an altitude of 8–12 km, and mainly affect the short-wave infrared and thermal infrared channels through thin cloud scattering.

[0166] Cloud optical thickness and liquid content: By setting parameters such as CLDALT, CLDTHK, and CLD in MODTRAN, the system simulates scene combinations with different thicknesses and cloud heights, thereby improving the diversity of spectral response simulation.

[0167] Simulation of aerosol distribution and types:

[0168] Aerosols have a particularly critical impact on radiative transfer processes in the visible and shortwave infrared channels. To more accurately characterize various atmospheric pollution and natural process scenarios, this application simulates the following standard aerosol types in MODTRAN 5.3: Rural, Urban, Maritime, Desert / Dust, and Background Tropospheric. Combined with settings such as AEROSOLMODEL, VIS (visibility), and ALTITUDE (aerosol layer height), the aerosol concentration and vertical distribution in the simulation are controlled, thereby realistically reproducing the impact of typical pollution, dust, and marine humidity environments on channel responses.

[0169] Combinatorial Context Construction Strategies:

[0170] To construct channel matching and transfer relationships with engineering application value, this application cross-combines the aforementioned different cloud and aerosol conditions with various atmospheric profiles, underlying surface types, and observation geometry to form a multivariate physical simulation matrix. Each set of parameters corresponds to a simulation scenario, outputting simulated radiation data for the corresponding band for subsequent fitting modeling and transfer function extraction.

[0171] Multi-scenario fitting strategy fusion:

[0172] For the simulation results obtained under the influence of cloud aerosols, this application adopts two strategies in parallel evaluation: scenario-specific fitting and full-scenario hybrid fitting.

[0173] Scenario-specific fitting: This involves establishing a dedicated transfer function under similar cloud or aerosol conditions, suitable for applications requiring environmental prediction. Hybrid fitting: This involves fitting all sample data together to obtain a model with the best overall fitting performance, suitable for applications requiring a balance between real-time accuracy and computational efficiency.

[0174] The above strategy effectively solves the problem of missing response information caused by a single surface setting in traditional simulations, and provides comprehensive and robust data support for subsequent spectral fitting, channel migration and response transformation modeling.

[0175] To achieve high-precision cross-platform radiation consistency assessment and calibration verification, Figure 2 This paper introduces the MODTRAN atmospheric radiative transfer model to perform numerical simulations of the Earth-atmosphere system radiative response under target observation conditions. Based on input parameters such as solar zenith angle, satellite zenith angle, relative azimuth angle, aerosol optical thickness, and water vapor content from the sensor's perspective, this module constructs the corresponding atmospheric profile model and calculates the transmission response relationship between surface reflectance and sensor radiance in the target band.

[0176] By introducing MODTRAN simulation results, we can compensate for observational differences caused by variations in geometry and aerosols, and also construct theoretical conversion relationships between the benchmark sensor and the target sensor. This process significantly improves the robustness of the calibration system to atmospheric disturbances, provides physical consistency constraints for subsequent cross-calibration regression analysis, and enhances the interpretability and universality of the results.

[0177] To achieve data compatibility, cross-satellite consistency processing, and continuous operational integration of observation products from the Advanced Geostationary Radiometric Imager (AGRI) aboard the Fengyun-4 series satellites (FY-4A and FY-4B), this application further designs an FY-4B data reconstruction method based on the "transfer coefficient" and proposes a complete data synthesis and format standardization process. The entire process is as follows: Figure 3 As shown, it can be divided into the following two sub-modules:

[0178] 1. FY-4A Standard Disk Synthesis Process

[0179] FY-4A / AGRI data serves as the data standard template for the first satellites in the series, and its product format and channel organization structure are used as a benchmark reference for cross-satellite fusion. Its disk data synthesis process mainly includes the following steps:

[0180] Data preprocessing and quality control:

[0181] After the FY-4A / AGRI raw L1 level observation data is flattened and calibrated (converted to physical units of reflectance / brightness temperature), it is first subjected to quality screening to remove null values, occlusions, and edge interference pixels to ensure data integrity and validity.

[0182] Geographic mapping and observational geometric correction:

[0183] Using the accompanying GEO positioning file, the latitude and longitude information of each pixel is retrieved based on the scan line / field of view relationship, and geometric distortions caused by the viewing angle are corrected in conjunction with Earth ellipsoid parameters and attitude information. Solar angle normalization is performed on the visible light channel to eliminate the influence of variations in solar incidence at different imaging times.

[0184] Disk region resampling synthesis:

[0185] Following the standard projection rules of the FY-4A / AGRI ground receiving system (such as equal latitude / longitude conics), the entire orbit observation data is resampled and projected onto a unified grid to generate a full-disk image with a resolution of 0.01° / pixel (or other specified spatial accuracy). This synthesis process maintains the continuity of the field of view and radiometric consistency, and is the basis for the generation of FY-4 AGRI service products.

[0186] Standardized output of product structure:

[0187] The final output standard FY-4A / AGRI product file includes: a radiometric inversion layer (reflectivity / brightness temperature) output synchronously from multiple channels, a geographic information layer (latitude, longitude, and altitude), a quality control layer (valid identifiers, angle information, etc.), and metadata records (channel settings, timestamps, orbital numbers, product levels, etc.).

[0188] 2. FY-4B Disk Data Compatibility Reconstruction Process Based on Channel Transfer Coefficient

[0189] To achieve compatibility between the FY-4B / AGRI observation products and the FY-4A product structure, and to ensure the comparability, stitching capability, and substitutability of the data from the two satellites, this application is based on the aforementioned channel response transfer model (see...). Figure 1 The FY-4B refactoring process is designed as follows:

[0190] FY-4B Raw Observation Data Analysis and Preprocessing:

[0191] Decode the FY-4B / AGRI L1B observation data to obtain calibrated reflectivity / brightness temperature values, and simultaneously read auxiliary parameters such as viewing angle, solar angle, and timestamp. To ensure the applicability of subsequent transfer coefficients, it is necessary to match the scene type used in the MODTRAN simulation (e.g., atmospheric profile, solar zenith angle range, etc.) as a prerequisite for model application.

[0192] Channel-level response conversion processing:

[0193] For each FY-4B observation channel, find its corresponding FY-4A target channel, and call the fitting function (linear, polynomial, nonlinear regression function, etc.) trained based on MODTRAN simulation to convert the FY-4B channel values ​​RB. B Convert to the value R of the simulated FY-4A response A That is: R A =aR B +b, R A =aR B 2 +bR B +c.

[0194] All transfer functions originate from Figure 1 The spectral response fitting module has its parameters pre-defined in the configuration table and is automatically invoked.

[0195] Geometric correction and coordinate alignment:

[0196] Repeat the geometric mapping process in the FY-4A data processing, and reproject the converted FY-4B channel response values ​​onto the standard disk grid according to a unified projection rule to ensure that the FY-4B product and FY-4A are aligned in spatial position.

[0197] Multichannel synthesis and structure regularization:

[0198] All converted channel data are synthesized and output, and the files are organized strictly according to the FY-4A product's structural specifications, including band arrangement order, resolution settings, projection parameters, and metadata formats. If necessary, a "Source Identifier" tag is added to indicate that the current data originates from FY-4B and whether it has undergone transmission and conversion.

[0199] 3. Cross-satellite joint data fusion and business expansion

[0200] Through the above data reconstruction process, the FY-4B / AGRI sensor output product constructed in this application not only maintains consistency with FY-4A in the spatial dimension, but also achieves near physical equivalence in the spectral response dimension through the transfer coefficient, ensuring that the two satellite products have good inheritance and continuity.

[0201] This method can be directly applied to the following typical scenarios: cross-satellite time-series monitoring (such as changes in surface temperature / NDVI), spatiotemporal tracking and path fusion analysis of typhoon cloud systems, and development of a unified platform for multi-satellite collaborative meteorological inversion algorithms. In addition, since the spectral response transfer mechanism of the corresponding channels of multispectral instruments between satellites proposed in this application is based on physical modeling and spectral simulation rather than pure empirical interpolation, its model also has the ability to be transferred to subsequent satellites in the future, and has good engineering scalability and versatility.

[0202] This invention addresses the difference in radiometric response between a target satellite and a reference satellite. Based on the consistency of the channel center band, it selects the spectral response function (SRF) of the corresponding channel to simulate the reference radiometric data. Specifically, by performing a weighted convolution between the simulated atmospheric radiometric data and the spectral response function of the reference satellite, the equivalent radiometric values ​​for each channel are obtained to approximate the observation output of the target satellite in the corresponding channel. This processing effectively restores the equivalent observation characteristics under different sensor responses. Subsequently, based on the data relationship between the convolution result and the actual observation output of the target satellite in the corresponding channel, a regression fitting method is used to construct a transfer model. The fitting results are established channel by channel, forming a set of input-output mapping coefficients, i.e., the transfer coefficients shown in the figure. These transfer coefficients can be directly used for subsequent calibration transformations to achieve accurate conversion of reference satellite radiometric results to target satellite observation data. Figure 4 The distribution characteristics of the fitting coefficients obtained under different channels are shown, verifying the feasibility and stability of the method of the present invention in terms of multi-channel matching and conversion accuracy.

[0203] To further verify the effectiveness of the established transfer model, the established transfer coefficient model was applied to the full-disk observation data of the FY-4B / AGRI sensor to generate fitted radiometric values ​​corresponding to FY-4A / AGRI. Figure 5 The figure shows the comparison results of FY-4B / AGRI data and FY-4A / AGRI data in typical channels after applying the transfer model. As can be seen from the figure, after model conversion, the two maintain good consistency in overall amplitude and spatial distribution, achieving effective conversion and comparison of radiometric observation data from different generations of geostationary meteorological satellites, providing support for long-term sequence construction and consistency monitoring.

[0204] Further Figure 6 This model applies to the conversion model of AQUA / MODIS and FY3D / MERSI-II, showing that channels with similar spectral center wavelengths between the two can also achieve bidirectional conversion through this model.

[0205] The last Figure 7 This method uses the spectral response functions of seven channels from MERSI and MODIS with similar center wavelengths as its strength. This model can only achieve spectral response transfer between instruments when the center wavelengths and bandwidths are similar. The prior conditions and limitations of this method are indicated.

[0206] It should be noted that, for the sake of simplicity, the aforementioned method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously.

[0207] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0208] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A modeling method for a multi-star collaborative spectral radiation observation migration and conversion model, characterized in that, Includes the following steps: S1. Scene Parameter Configuration and Input: Obtain and configure the input parameter set for radiative transfer simulation; the input parameter set includes atmospheric state parameters, surface attribute parameters, observation geometric parameters, and observation scene; S2. Radiative transfer simulation and spectral convolution: Input the input parameter set into the MODTRAN atmospheric radiative transfer model for forward simulation calculation to obtain hyperspectral radiance data; S3. Calculation of Radiance and Generation of Datasets: The standardized spectral response functions of different generations of multispectral sensors in the selected satellites are loaded into the MODTRAN atmospheric radiative transfer model, or the hyperspectral radiance data output by MODTRAN is post-processed and convolved to output the simulated channel radiance values ​​of each channel of the selected satellite under the same input observation scenario; the channels include thermal infrared channels and visible-near infrared solar reflection channels. S4. Radiation migration modeling: Paired analysis is performed using simulated channel brightness temperature values ​​of various generations of multispectral sensors under the same observation scenario to obtain a paired sample set in the form of a benchmark sensor and a target sensor. A radiation migration model between different generations of multispectral sensors is constructed using statistical modeling methods, and the paired sample set is used for training and accuracy verification to obtain the final radiation migration model. The radiation migration model includes an empirical fitting model and a lookup table model. The empirical fitting model is a linear regression model or a higher-order polynomial regression model. S5. Cross-sensor calibration application: The final radiometric migration relationship model is applied to the selected satellite remote sensing observation data to unify the radiometric benchmark of observation data among different generations of multispectral sensors of the selected satellite, and generate a long-term radiometric dataset with temporal continuity and cross-platform / cross-generational consistency; the application methods include forward transformation and cross-calibration. The forward conversion refers to converting historical observation data from one generation of multispectral sensors on the selected satellite to the radiometric reference of another generation of multispectral sensors on the same selected satellite. The cross-calibration refers to converting the observation data of the selected satellite multispectral sensor to the radiation reference of the selected reference satellite multispectral sensor.

2. The modeling method according to claim 1, characterized in that, The input parameter set is obtained from sources including ERA5 or MERRA-2 reanalysis data, the MODIS Land Cover database, or standard atmospheric profiles, and a representative atmospheric-surface input sample library is constructed through combined sampling. The atmospheric state parameters include the vertical profiles of atmospheric temperature, atmospheric humidity, ozone concentration, atmospheric visibility, concentration profiles of other atmospheric components, aerosol optical thickness, and cloud parameters. The surface attribute parameters include surface temperature, surface infrared emissivity, surface albedo, and surface cover type; The observation geometric parameters include the solar zenith angle, the satellite observation zenith angle, and the relative azimuth angle between the sun and the satellite. The observation scenarios include clear-sky ocean, clear-sky land, partial cloud coverage, and multi-layered cloud coverage; The input parameter set is subjected to data quality control before being input into the MODTRAN atmospheric radiative transfer model, including parameter range checking, missing value handling, and format standardization.

3. The modeling method according to claim 1, characterized in that, The specific steps for obtaining the hyperspectral radiance data are as follows: Parameter preprocessing: The input parameter set is converted according to the input format requirements of the MODTRAN atmospheric radiative transfer model, including hierarchical matching of atmospheric profile data, dimensional adaptation of surface attribute parameters, and unit unification of observation geometric parameters; Model parameter configuration: Set the simulation core parameters in the MODTRAN atmospheric radiative transfer model, including: radiative transfer calculation mode, spectral resolution, atmospheric model, and underlying surface type; Forward simulation calculation: The preprocessed input parameter set and the configured simulation core parameters are input into the MODTRAN atmospheric radiative transfer model to calculate the atmospheric radiative transfer process, and the initial hyperspectral radiance data is output. Post-processing of hyperspectral data: Extract the initial hyperspectral radiance data, remove simulated outliers and perform wavenumber coordinate calibration to obtain hyperspectral radiance data with a wavenumber range covering the entire spectrum from visible light to thermal infrared and a wavenumber interval equal to the simulated resolution.

4. The modeling method according to claim 1, characterized in that, The convolution integral is expressed as: ， Where L(ν) is the hyperspectral radiance data at wavenumber ν output by the MODTRAN atmospheric radiative transfer model, SRF(ν) is the normalized spectral response function at wavenumber ν for a specific channel corresponding to the spectral characteristics of the multispectral sensor, and v min and v max These represent the effective wavenumber ranges of the normalized spectral response function for that channel. During the post-processing convolution integration, the wavenumber coordinates of the hyperspectral radiance data are aligned and interpolated with the defined coordinates of the normalized spectral response function of the multispectral sensor. Specifically: Coordinate acquisition and comparison: The wavenumber sequence VL = [νl1, νl2, ..., νl] of the hyperspectral radiance data output by the MODTRAN atmospheric radiative transfer model was extracted. N The wavenumber sequence VS = [νs1, νs2, ..., νs] of the normalized spectral response function of the multispectral sensor. M By comparing the wavenumber ranges of the two wavenumber sequences, the intersection range [νl, νs] is determined, where N is the number of wavenumbers in the hyperspectral radiance data and M is the number of wavenumbers in the normalized spectral response function of the multispectral sensor. Reference coordinate establishment: The wavenumber sequence VL of hyperspectral radiance data is used as the reference wavenumber coordinate; SRF interpolation matching: The wavenumber sequence of the normalized spectral response function of the multispectral sensor is interpolated onto the reference wavenumber coordinates using the linear interpolation method to obtain the normalized spectral response function of the multispectral sensor after interpolation; Post-interpolation verification: Calculate the integral value deviation of the standardized spectral response function of the multispectral sensor before and after interpolation. If the integral value deviation is greater than the set deviation threshold, switch to cubic spline interpolation and re-interpolate to ensure that the energy characteristics of the standardized spectral response function of the multispectral sensor remain unchanged after interpolation.

5. The modeling method according to claim 1, characterized in that, The pairing analysis specifically includes: Dataset preprocessing: The simulated brightness temperature datasets from various generations of multispectral sensors are uniformly preprocessed; the preprocessing involves outlier removal. Pairing rule establishment: Matching of multispectral instrument features between satellites based on the spectral characteristics of each channel of the selected satellite; Pairing quality assessment: Analysis of paired multispectral instrument channels, including differences in spectral response functions and data reflecting the channel matching characteristics between different instruments.

6. The modeling method according to claim 1, characterized in that, The empirical fitting model is constructed as follows: Model type selection: The model type is selected based on the scatter distribution characteristics of the paired sample set. If the scatter points are linearly distributed, a linear regression model is selected; if they are nonlinearly distributed, a 2nd-3rd order higher-order polynomial regression model is selected. The higher-order polynomial regression model is constrained by L2 regularization of the absolute value of the coefficients. Model parameter solution: The corresponding algorithm is used to solve the model parameters. For linear regression models, the slope and intercept are solved by the least squares method; for higher-order polynomial regression models, the polynomial coefficients are solved by the partial least squares method. Model accuracy validation: Samples are randomly selected from the paired sample set according to a set ratio as the validation set, and the rest are used as the training set; the empirically fitted model is trained and validated using the training set and the validation set, the predicted values ​​are calculated and evaluated using accuracy metrics; the accuracy metrics include the coefficient of determination, mean absolute error, root mean square error and bias. Model optimization: Optimize the model based on accuracy metrics until the model accuracy requirements are met; The lookup table model construction process is as follows: the lookup table is designed in the form of a multidimensional array, binning indexes are performed according to different angle ranges and different surface attributes in the simulation, and coefficients obtained by fitting through empirical models are integrated to generate more refined transformation coefficients; the corresponding migration model transfer coefficients are found by combining channel identifiers, angle range binning, and surface attribute codes.

7. The modeling method according to claim 1, characterized in that, The forward transition specifically refers to: Sensor designation: Use a multispectral sensor of one generation of the selected satellite as the reference sensor, and another multispectral sensor of the selected satellite as the target sensor; Historical data preprocessing: Historical observation data of the reference sensor L1B level data is acquired and preprocessed; the preprocessing includes quality control and spatiotemporal warping; the quality control is used to remove abnormal data from the historical observation data of the reference sensor and perform parameter range checks; the spatiotemporal warping is used to unify the historical observation data of the reference sensor to the spatial resolution of the target sensor and unify the time reference. Model selection and parameter matching: Select the optimal radiative migration relationship model based on the complexity of the observation scenario; The empirical fitting model calls the regression equation coefficients according to the channel identifier, while the lookup table model reads the conversion coefficients according to the channel identifier + angle range binning and the combined encoding of land surface type. Forward conversion calculation: Load the model parameters of the optimal radiative migration relationship model and convert the historical observation data of the reference sensor to the radiative reference of the target sensor; Post-processing of the converted results includes accuracy verification, data format conversion, and outlier handling. Accuracy verification involves randomly selecting a set proportion of the converted data and comparing it with the measured data from the target sensor during the same period, calculating a unified accuracy index. If the accuracy requirements are not met, the model parameter matching process is re-checked. The unified accuracy index includes the coefficient of determination, mean absolute error, root mean square error, and bias. Data format conversion involves converting the converted data to the standard data format of the target sensor while retaining the metadata information of the original data corresponding to the converted data. Outlier handling involves removing outliers that exceed the physical range generated during the conversion process and supplementing them using spatial interpolation. Results Output and Archiving: The transformed data is archived according to time series, and a transformation quality report is generated; the report includes model type, transformation accuracy index, and outlier ratio.

8. The modeling method according to claim 1, characterized in that, The cross-calibration specifically refers to: Sensor designation: Select a satellite or generational multispectral sensor with a known radiation reference and high calibration accuracy as the reference sensor, and use the target satellite or target generational multispectral sensor to be calibrated as the target sensor; The reference sensor and the target sensor are from the same satellite but different generations or are sensors from different satellites across satellites / generations; Target data preprocessing: Acquire and preprocess the observation data from the target sensor; the preprocessing includes quality control and spatiotemporal shaping; The quality control is used to remove abnormal data from the target sensor observation data and to perform parameter range checks; the spatiotemporal adjustment is used to unify the target sensor observation data to the spatial resolution of the reference sensor and to unify the time reference. Inverse model construction: Based on the final radiative migration relationship model, an inverse model between the reference sensor and the target sensor is constructed; Cross-calibration calculation: The inverse model is used to transform the observation data of the target sensor to the radiation reference of the reference sensor to complete the cross-calibration; Calibration result verification: The calibration results are compared with the data of the reference sensor during the same period, and a unified accuracy index is calculated to verify the accuracy during the same period. The unified accuracy index includes the coefficient of determination, mean absolute error, root mean square error, and deviation. The calibration results are merged with the historical long-term data of the reference sensor, and the time series continuity is analyzed to verify the long-term trend until the accuracy requirements are met. Long-term time series dataset generation: The validated calibrated data is spliced ​​with the historical observation data of the reference sensor, sorted in chronological order and duplicate data is removed to generate a long-term time series dataset with a unified radiation reference, and data source information is added.

9. A modeling system for a multi-star collaborative spectral radiation observation migration and conversion model, characterized in that, A modeling method applied to the multi-star collaborative spectral radiometric observation migration and transformation model according to any one of claims 1-8, the system comprising a parameter set configuration module, a simulation calculation module, a channel radiometric processing module, a relationship modeling module, and a calibration application module; The parameter set configuration module is used to acquire and configure the input parameter set for radiative transfer simulation; the input parameter set includes atmospheric state parameters, surface attribute parameters, observation geometric parameters, and observation scene; The simulation calculation module is used to input the input parameter set into the MODTRAN atmospheric radiative transfer model for forward simulation calculation to obtain hyperspectral radiance data. The channel radiation processing module is used to load the standardized spectral response functions of different generations of multispectral sensors in the selected satellites into the MODTRAN atmospheric radiative transfer model, or to perform post-processing convolution integration on the hyperspectral radiance data output by MODTRAN, and output the simulated channel radiance values ​​of each channel of the selected satellite under the same input observation scenario; the channels include thermal infrared channels and visible light-near infrared solar reflection channels; The relationship modeling module is used to perform paired analysis on the channel brightness temperature values ​​of simulated multispectral sensors of different generations under the same observation scenario to obtain a paired sample set in the form of a benchmark sensor-target sensor; a radiation migration relationship model between multispectral sensors of different generations is constructed through statistical modeling methods, and the paired sample set is used for training and accuracy verification to obtain the final radiation migration relationship model; the radiation migration relationship model includes an empirical fitting model and a lookup table model; the empirical fitting model is a linear regression model or a higher-order polynomial regression model; The calibration application module is used to apply the final radiation migration relationship model to the selected satellite remote sensing observation data, so as to unify the radiation benchmark of the observation data among different generations of multispectral sensors of the selected satellite, and generate a long-term radiation dataset with temporal continuity and cross-platform / cross-generational consistency; the application methods include forward transformation and cross-correction. The forward conversion refers to converting historical observation data from one generation of multispectral sensors on the selected satellite to the radiometric reference of another generation of multispectral sensors on the same selected satellite. The cross-calibration refers to converting the observation data of the selected satellite multispectral sensor to the radiation reference of the selected reference satellite multispectral sensor.

10. A computer-readable storage medium storing a program, characterized in that, When the program is executed by the processor, it implements the modeling method of the multi-star cooperative spectral radiation observation migration and transformation model according to any one of claims 1-8.

Images