A wide range of fast optical satellite sensor on-board observation radiance simulation method

Abstract

The application discloses a wide-range fast optical satellite sensor on-satellite observation radiance simulation method and belongs to the technical field of satellite sensing data processing.The application is a method for quickly obtaining wide-range optical satellite sensor on-satellite observation radiance by combining an atmospheric radiation transmission model with a machine learning method.First, a large number of high-precision and representative atmospheric profiles are used as input parameters of atmospheric radiation transmission model simulation to obtain a large number of simulated on-satellite radiance values and corresponding ground state, observation angle and atmospheric state parameter data sets as training data sets for constructing a machine learning algorithm, a machine learning algorithm is used to fit the nonlinear relationship between the on-satellite radiance and the influence factors in the simulation results, and a machine learning model capable of quickly predicting the on-satellite observation radiance of the optical satellite sensor is obtained.The application solves the problem that the sensor has no corresponding global channel reflectivity, and greatly shortens the simulation time and computing resources required.

Description

Technical Field

[0001] This invention belongs to the field of satellite sensor data processing technology, specifically relating to a method for simulating the radiance of a large-scale, fast optical satellite sensor on-board observations. Background Technology

[0002] With the rapid development of satellite sensor performance, the imaging methods of visible light band sensors have gradually shifted from traditional diurnal imaging to all-day imaging (diurnal + lunar imaging). A key challenge in this transition is pre-setting the sensor's dynamic range (the radiance range within a single sensor image can span ten orders of magnitude) to ensure the reliability of on-orbit observations. To address this, during the sensor band design phase, it's necessary to understand the range of on-board radiance variations under all possible observation conditions within a single image during on-orbit operation to assist in setting the sensor's dynamic range. On-board radiance simulation based on atmospheric radiative transfer theory is an economical and rapid solution to this problem. Most existing satellite sensor on-board radiance simulation schemes obtain "point-like" atmospheric parameters at specific geographical locations through field experiments. While this method offers high accuracy, it suffers from low data acquisition frequency, small data coverage, and high cost, thus failing to achieve large-scale, high-temporal-resolution on-board radiance simulation to support sensor dynamic range setting. Currently, some simulation methods extract "area-like" atmospheric state information from products or reanalysis data of water vapor, ozone, and aerosol optical thickness obtained from high-performance and calibrated satellite sensors, enabling large-scale on-board observations from specific optical sensors. However, the following problems still exist in achieving large-scale, high-temporal-resolution on-board radiance simulation of optical satellite sensors based on satellite atmospheric parameter inversion products or reanalysis data:

[0003] (1) In the design of the dynamic range of satellite sensors, it is necessary to understand the magnitude of the on-board radiance of all pixels in all observed images during satellite operation and the distribution of pixels with different radiance magnitudes in the observed images, which requires a large number of simulations. If the traditional atmospheric radiative transfer model driven by physical processes is used for simulation, it will consume a huge amount of time and computer resources. Furthermore, the spectral response function of the sensor needs to be adjusted multiple times to achieve the design goal. The existing simulation methods are obviously no longer able to meet the requirements of fast and efficient use.

[0004] (2) For sensors still in the design phase, there are no directly available surface reflectance products to provide reliable surface background field information for simulation. Due to the differences in spectral response functions between different sensors, the surface reflectance products of existing sensors cannot be directly applied to other sensors. Summary of the Invention

[0005] This invention provides a method for simulating the radiance of on-board observations from large-scale, rapid optical satellite sensors, which can be used to improve the simulation accuracy of on-board observations from large-scale satellite sensors.

[0006] The technical solution adopted in this invention is as follows:

[0007] A method for simulating radiance from on-board observations using a large-scale, fast optical satellite sensor, characterized by the following steps:

[0008] Step S1: Construct a typical ground object spectral library, and select sensors whose spectral range overlaps with the spectral response function range of the sensor to be simulated from the spectral response function library of satellite sensors. Integrate the spectral response functions of both sensors with the typical ground object spectra in the ground object spectral library to obtain the channel reflectance ρ of band λ. b :

[0009]

[0010] Where ρ(λ) represents the continuous spectral curve of the ground object, f(λ) represents the spectral response function of band λ, and λ max and λ min These represent the upper and lower bounds of the spectral response function's band range, respectively.

[0011] Using a machine learning algorithm to fit the conversion formula between the two channel reflectivities, then ρ b =f(ρ1,ρ2...ρ n ), where ρ1, ρ2…ρ n f(ρ1,ρ2…ρ) represents the channel reflectivity of the sensor that provides the reflectivity background field. n The figure represents the conversion relationship between the background field reflectance and the channel reflectance of a certain band.

[0012] Step S2: Filter the global atmospheric profile dataset and construct an atmospheric profile dataset:

[0013] Filter and remove profiles with relative humidity exceeding a specified value: In high-latitude regions, profiles are removed based on a set first threshold; in mid- and low-latitude regions, profiles are removed based on a set second threshold.

[0014] Then remove profiles whose altitude is below the specified altitude threshold;

[0015] Step S3, Radiative Transfer Simulation:

[0016] Based on the atmospheric profile dataset constructed in step S2, the zenith angle of the radiation source (sun and moon) is set to M values ​​that vary with a step size a° in the interval [0°, 180°], the zenith angle observed by the sensor is set to N values ​​that vary with a step size b° in the interval [0°, 90°], and the lunar phase angle observed by the sensor is set to L values ​​that vary with a step size c° in the interval [0°, 180°]. A total of K×M×N×(L+1) simulations are performed, and the simulation results are used to construct the training dataset for the machine learning model.

[0017] Set the total radiation L at the sensor entrance pupil. toa (θ1,θ2,ρ t The total radiation L(θ1,θ2,ρ) at the sensor entrance pupil t ) and sensor channel radiance L c (θ1,θ2,ρ t ):

[0018] L toa (θ1,θ2,ρ t ) = L s (θ1,θ2,ρ t )+L m (θ1,θ2,ρ t );

[0019]

[0020]

[0021] The radiation sources include the sun and the moon, θ1 represents the solar zenith angle, θ2 represents the sensor zenith angle, and ρ t L represents the surface reflectance at the observed target location. s (θ1,θ2,ρ t L represents the solar radiation at the sensor's entrance pupil. m (θ1,θ2,ρ t ) represents lunar radiation at the sensor's entrance pupil, L0(θ1,θ2) represents atmospheric path radiation, and F d (θ1) represents the incident radiation at the Earth's surface, T(θ2) represents the upward radiation transmittance of the atmosphere, and S(θ1,θ2) represents the spherical albedo of the lower surface of the atmosphere.

[0022] Step S4, Simulation based on machine learning model:

[0023] Machine learning algorithms were used to establish atmospheric path radiation L0(θ1,θ2) and surface incident radiation F. d The nonlinear mapping relationship between (θ1), the spherical albedo S(θ1,θ2) of the lower atmospheric surface, the upward atmospheric radiative transmittance T(θ2), and the simulation input:

[0024] L0(θ1,θ2)=f 1 (AOD,TQV,O3,θ1,θ2);

[0025] F d (θ1)=f 2 (AOD,TQV,O3,θ1);

[0026] S(θ1,θ2)=f 3 (AOD,TQV,O3,θ1,θ2);

[0027] T(θ2)=f 4 (AOD,TQV,O3,θ2);

[0028] Where AOD represents the aerosol optical thickness of the reanalysis data, TQV represents the atmospheric water vapor content of the reanalysis data, O3 represents the ozone content of the reanalysis data, and f 1 () represents the nonlinear relationship between aerosol optical thickness, atmospheric water vapor content, ozone content, solar zenith angle, sensor zenith angle, and atmospheric path radiation. 2 () represents the nonlinear relationship between aerosol optical thickness, atmospheric water vapor content, ozone content, solar zenith angle, and incident radiation at the Earth's surface. 3 () represents the nonlinear relationship between aerosol optical thickness, atmospheric water vapor content, ozone content, solar zenith angle, sensor zenith angle, and balloon surface albedo. 4 () represents the nonlinear relationship between aerosol optical thickness, atmospheric water vapor content, ozone content, sensor zenith angle, and atmospheric transmittance;

[0029] Step S5: Based on the orbital parameters of the satellite carrying the sensor to be simulated, obtain the sensor observation zenith angle and observation time for each pixel in the simulation area. Calculate the radiation source zenith angle at each pixel based on the observation time, and combine it with AOD, TQV, and O3 through the nonlinear relationship f. 1 ()~f 4 () Calculate atmospheric path radiation L0(θ1,θ2) and surface incident radiation F d The simulation results of (θ1), the spherical albedo S(θ1,θ2) of the lower atmospheric surface and the atmospheric upward radiation transmittance T(θ2) are substituted into the total radiation L(θ1,θ2,ρ) at the sensor entrance pupil. t The expression for ) is used to obtain the total radiation L(θ1,θ2,ρ) at the entrance pupil of the sensor to be simulated based on a machine learning model. t ), which is the observed value of the satellite's radiance.

[0030] Furthermore, the first threshold includes the longitude difference threshold, latitude difference threshold, elevation difference threshold, and atmospheric water vapor content difference; the second threshold includes the longitude difference threshold, latitude difference threshold, elevation difference threshold, atmospheric water vapor content difference, and month difference threshold.

[0031] Furthermore, the zenith angle of the radiation source at each pixel is calculated based on the observation time as follows:

[0032] α=|M S -M L ±180°|;

[0033] sin(H s )=sin(φ)·sin(δ)+cos(φ)·cos(δ)·cos(t);

[0034]

[0035] Where α represents the lunar phase angle, M S M represents the apparent ecliptic longitude of the sun. L H represents the ecliptic longitude of the moon. s The solar altitude angle is represented by φ, the local latitude by δ, the solar declination by t, and the local solar time by A. s It indicates the azimuth angle of the sun.

[0036] The technical solution provided by this invention brings at least the following beneficial effects:

[0037] In this invention, the MODTRAN radiative transfer model is used to construct a training dataset for the machine learning algorithm. This dataset includes all possible satellite observation states and simultaneously considers the influence of solar and lunar radiation on the satellite observation results. After the machine learning model is trained, using this model to replace the traditional radiative transfer simulation process significantly reduces the simulation time and allows for the rapid acquisition of simulation results of onboard observations from visible light satellite sensors over large areas. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention, 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a schematic diagram of the processing procedure for a method of simulating the radiance of a large-scale, rapid optical satellite sensor on-board observation provided in an embodiment of the present invention;

[0040] Figure 2This is a schematic diagram of the reflectance conversion verification results of a large-scale, fast optical satellite sensor on-board observation radiance simulation method provided in an embodiment of the present invention.

[0041] Figure 3 This is a simulation accuracy verification result diagram using the radiative transfer model in an embodiment of the present invention. Among them, (3-a) is the verification result with observations from ground stations; (3-b) is the verification result with observations from MODIS Band 4; and (3-c) is the verification result with observations from VIIRS DNB band.

[0042] Figure 4 This is a test set verification diagram of the prediction model for atmospheric path radiation, spherical albedo, surface incident radiation, and atmospheric upward transmittance established by the machine learning algorithm constructed in this embodiment of the invention. Wherein, (4-a) is surface incident radiation; (4-b) is spherical albedo; (4-c) is path radiation; and (4-d) is atmospheric upward transmittance.

[0043] Figure 5 This is a comparison chart of the simulation results using the method of the present invention and the actual on-board observations in the VIIRS DNB band in an embodiment of the present invention. Wherein, (a) is the actual on-board observation in the VIIRS DNB band; (b) is the simulation result using the method of the present invention; (c) is the lunar zenith angle during simulation; and (d) is the solar zenith angle during simulation. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0045] The purpose of this invention is, firstly, to generate global surface reflectance products for the sensors to be simulated using existing global surface reflectance products, based on the constructed typical ground object spectral library and satellite sensor spectral response function library, thus providing stable and reliable surface background field information for the simulation; secondly, to obtain radiance simulation results under different observation conditions using high-precision and representative atmospheric profile inputs based on a physics-driven radiative transfer model, which is then used to construct a training dataset for the machine learning model, ensuring the accuracy of the simulation scheme; and thirdly, to use a machine learning model to fit the nonlinear relationship between each radiative component and each influencing factor, ensuring the efficiency of the simulation scheme.

[0046] The large-scale, fast on-board observation radiance simulation method for optical satellite sensors provided in this invention can solve the following problems:

[0047] (1) Construct a complete spectral library of typical ground features to ensure the reliability of the sensor channel reflectance conversion relationship obtained based on the spectral library;

[0048] (2) Based on the atmospheric radiation transfer simulation results, a complete training dataset is constructed for the machine learning algorithm so that the training dataset can represent different atmospheric states encountered during satellite observation.

[0049] like Figure 1 As shown in the figure, the method for simulating the radiance of a large-scale, fast-speed optical satellite sensor on-board observations provided by this invention includes:

[0050] Step 1: Surface reflectance conversion.

[0051] To ensure the reliability of the channel reflectance conversion relationship constructed by the method, this embodiment collected a large number of typical ground object spectra in the following three ways: First, combined with a large number of domestic and foreign typical ground object spectral libraries, the spectral data of different typical ground objects provided in the spectral library were downloaded and screened; second, field field typical ground object spectral measurement experiments were conducted to collect some high-quality typical ground object spectral curves; finally, spectral information of some special ground objects, such as sea ice and clouds, was obtained through literature review. After the typical ground object spectral library was constructed, sensors whose band range overlapped with the band range of the spectral response function of the sensor to be simulated (global reflectance products already exist) were selected from the satellite sensor spectral response function library. The spectral response functions of the two were integrated with the typical ground object spectra in the ground object spectral library (Equation (1)) to obtain the corresponding channel reflectance. The conversion formula between the two channel reflectances was fitted using a machine learning algorithm (Equation (2)).

[0052] (2)

[0054] ρ b =f(ρ1,ρ2…ρ n )

[0055] In the formula, ρ(λ) is the continuous spectral curve of the ground object, ρ b Let f(λ) be the channel reflectance in a certain band, and f(λ) be the spectral response function in that band. max and λ min ρ1, ρ2…ρ3 represent the upper and lower bounds of the spectral response function band range, respectively; ρ1, ρ2…ρ3 represent the channel reflectance of the sensor providing the reflectance background field; f(ρ1, ρ2…ρ3) represents the reflectance of the spectral response function band range. n () represents the conversion relationship between background field reflectivity and channel reflectivity in a certain band.

[0056] Step 2: Construction of atmospheric profile dataset.

[0057] To reduce redundant simulations and shorten simulator time while ensuring the diversity of atmospheric states during the simulation process, this embodiment filters the global atmospheric profile dataset in the following way to form a global atmospheric profile database:

[0058] Profiles with relative humidity exceeding a specified value (e.g., 85%) are filtered out to remove the influence of clouds on the profiles. In high-latitude regions, profiles are eliminated based on a set first threshold to remove redundant profiles with high similarity. The first threshold includes a longitude difference threshold, a latitude difference threshold, an elevation difference threshold, and an atmospheric water vapor content difference. In this embodiment, profiles with longitude differences ≤60°, latitude differences ≤15°, elevation differences ≤1000m, and atmospheric water vapor content differences ≤0.5g·cm³ are filtered out. -2 The profiles are used to remove redundant profiles with high similarity;

[0059] In mid-to-low latitude regions, profiles are filtered out based on a set second threshold to remove redundant profiles with high similarity. This second threshold includes longitude difference threshold, latitude difference threshold, elevation difference threshold, atmospheric water vapor content difference threshold, and month difference threshold. In this embodiment, profiles with longitude differences ≤30°, latitude differences ≤10°, elevation differences ≤1000m, and atmospheric water vapor content differences not exceeding 0.5 g·m are filtered out. -2 Profiles with a month difference of ≤2 months are selected to remove redundant profiles with high similarity.

[0060] After filtering according to the above conditions, profiles with altitudes below a specified altitude threshold (e.g., 0m) in the atmospheric profile database obtained after passing the above filtering conditions are removed. The remaining atmospheric profiles are the constructed global atmospheric profile database.

[0061] Step 3: Radiative transfer simulation.

[0062] The radiation at the entrance pupil of the satellite sensor for the optical sensor includes solar radiation and lunar radiation (Equation (3)). Both types of radiation can be decomposed into the form of Equation (4), which can be represented by atmospheric path radiation, surface albedo, spherical albedo, surface incident radiation, and atmospheric ascending transmittance. Integrating Equation (4) with the sensor's spectral response function converts the radiation at the entrance pupil of the satellite sensor into the observed radiance value on the sensor satellite (Equation (5)). Among them, surface reflectance is an inherent property of the Earth's surface. Assuming that surface reflectance is isotropic, the other components are only related to the atmospheric state, the zenith angle of the radiation source, and the zenith angle observed by the sensor during the simulation process. Therefore, in this embodiment, the K atmospheric profile data selected in step 2 are used to characterize the atmospheric state during the simulation process. The zenith angle of the radiation source (sun and moon) is set to M values ​​that vary with a step size a° in the interval [0°, 180°]. The zenith angle observed by the sensor is set to N values ​​that vary with a step size b° in the interval [0°, 90°]. (When performing lunar radiation simulation, the lunar phase angle also needs to be considered as L values ​​that vary with a step size c° in the interval [0°, 180°]. A total of K×M×N×(L+1) simulations are performed, and the simulation results are used to construct the training dataset for the machine learning model. (3)

[0064] L toa (θ1,θ2,ρ t ) = L s (θ1,θ2,ρ t )+L m (θ1,θ2,ρ t )

[0065]

[0066]

[0067] In the formula, θ1 is the solar zenith angle, θ2 is the sensor zenith angle, and ρ t L is the surface reflectance at the target location. toa (θ1,θ2,ρ t L represents the total radiation at the sensor's entrance pupil. s (θ1,θ2,ρ t L represents the solar radiation at the sensor's entrance pupil. m (θ1,θ2,ρ t ) represents the lunar radiation at the sensor's entrance pupil, L(θ1,θ2,ρ t ) represents the total radiation at the sensor entrance pupil; L0(θ1,θ2) represents the atmospheric path radiation; F d (θ1) represents the incident radiation at the Earth's surface, T(θ2) represents the upward radiation transmittance, S(θ1,θ2) represents the spherical albedo of the lower atmospheric surface, and L c (θ1,θ2,ρ t f(λ) represents the radiance of the sensor channel; f(λ) represents the spectral response function of the sensor; λ min λ is the minimum wavelength of the sensor's spectral response function. max The maximum wavelength of the sensor's spectral response function is given by equations (3) to (5). Equations (3) to (5) decompose the on-board radiance into several different parameters, which can then be used for model simulation through machine learning.

[0068] (4) Simulation based on machine learning models.

[0069] Based on step 3, machine learning algorithms are used to establish nonlinear mapping relationships between atmospheric path radiation, balloon surface albedo, surface incident radiation and atmospheric upward transmittance and simulation input ((6)-(9)). (6)

[0071] L0(θ1,θ2)=f 1 (AOD,TQV,O3,θ1,θ2) (7)

[0073] F d (θ1)=f2 (AOD,TQV,O3,θ1) (8)

[0075] S(θ1,θ2)=f 3 (AOD,TQV,O3,θ1,θ2) (9)

[0077] T(θ2)=f 4 (AOD,TQV,O3,θ2)

[0078] In the formula, AOD is the aerosol optical thickness of the reanalysis data, TQV is the atmospheric water vapor content of the reanalysis data, O3 is the ozone content of the reanalysis data, and f 1 () represents the nonlinear relationship between aerosol optical thickness, atmospheric water vapor content, ozone content, solar zenith angle, sensor zenith angle, and atmospheric path radiation. 2 () represents the nonlinear relationship between aerosol optical thickness, atmospheric water vapor content, ozone content, solar zenith angle, and incident radiation at the Earth's surface. 3 () represents the nonlinear relationship between aerosol optical thickness, atmospheric water vapor content, ozone content, solar zenith angle, sensor zenith angle, and balloon surface albedo. 4 () represents the nonlinear relationship between aerosol optical thickness, atmospheric water vapor content, ozone content, sensor zenith angle, and atmospheric transmittance. When large-scale rapid simulation is required, aerosol optical thickness (AOD), atmospheric water vapor content (TQV), and ozone content (O3) from reanalysis data are used as atmospheric parameter inputs for simulation. The sensor observation zenith angle and observation time for each pixel in the simulation area are obtained according to the orbital parameter settings of the satellite carrying the sensor to be simulated. The radiation source zenith angle at each pixel is calculated using equations (10)-(12) based on the observation time. Substituting these into equations (6)-(9) yields the simulation results of atmospheric path radiation, balloon albedo, surface incident radiation, and atmospheric upward transmittance for each pixel. Substituting the simulation results and the surface reflectance in step 1 into equations (4)-(5) yields the total radiation L(θ1,θ2,ρ) at the entrance pupil of the sensor to be simulated based on the machine learning model. t ). (10)

[0080] α=|M S -M L ±180°| (11)

[0082] sin(H s )=sin(φ)·sin(δ)+cos(φ)·cos(δ)·cos(t)

[0083]

[0084] In the formula, α is the lunar phase angle, and M S M is the apparent ecliptic longitude of the sun. L H is the apparent ecliptic longitude of the moon. s The solar altitude angle is φ, where φ represents the local latitude, δ represents the solar declination, and t represents the local solar time. s This is the solar azimuth angle.

[0085] This invention provides a method for rapidly acquiring on-board radiance data from a large-scale optical satellite sensor by combining an atmospheric radiative transfer model with machine learning. First, a large number of high-precision and representative atmospheric profiles are used as input parameters for the atmospheric radiative transfer model simulation. This yields a large dataset of simulated on-board radiance values ​​and corresponding surface conditions, observation angles, and atmospheric state parameters, which serves as the training dataset for constructing the machine learning algorithm. Subsequently, the machine learning algorithm is used to fit the nonlinear relationship between the on-board radiance and its influencing factors in the simulation results, resulting in a machine learning model capable of rapidly predicting on-board radiance from optical satellite sensors.

[0086] To further verify the performance of the large-scale fast optical satellite sensor on-board observation radiance simulation method provided in this embodiment of the invention, the following simulation experiment was conducted.

[0087] The data selected for this experiment mainly include the surface reflectance product MOD09GA from the Moderate-Resolution Imaging Spectroradiometer (MODIS); hourly aerosol optical thickness, atmospheric water vapor content, and ozone content products from MERRA-2 reanalysis data; typical ground object spectral data from domestic and international ground object spectral libraries such as the Jet Propulsion Laboratory (JPL), The Johns Hopkins University (JHU), the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER), and the Hyperspectral Image Processing and Analysis System (HIPAS); a basic atmospheric profile database was constructed using atmospheric profiles from the SeeBor V5.1 atmospheric profile database; and the Moderate Resolution Atmospheric Transmittance algorithm was used. The MODTRAN algorithm is used to simulate radiative transfer and construct a training dataset for the machine learning algorithm. The day / night band of the Visible Infrared Imaging Radiometer (VIIRS) is used as the sensor to be simulated. The implementation can be divided into the following four steps.

[0088] (1) Surface reflectance conversion.

[0089] The sensor spectral response functions of the VIIRS DNB band and MODIS 1-4 bands were integrated with the spectra of typical land cover in the typical land cover spectral library constructed using this method to obtain the corresponding channel reflectances. The conversion relationship between the MODIS 1-4 band channel reflectances and the VIIRS DNB band channel reflectances was then fitted using the random forest algorithm. The verification results are shown in […]. Figure 2 Subsequently, global surface reflectance data from MODIS bands 1-4 were used as input to obtain global surface reflectance data from the VIIRSDNB band.

[0090] (2) Construction of atmospheric profile dataset.

[0091] After collecting the above atmospheric profile dataset, the atmospheric profile filtering scheme in step 2 of the technical solution was used to filter 15,704 atmospheric profiles in the SeeBor V5.1 atmospheric profile database, resulting in 549 representative atmospheric profile datasets.

[0092] (3) Radiative transfer simulation.

[0093] The atmospheric state during the simulation was characterized using 549 selected atmospheric profiles. For solar radiation, the solar zenith angle was set to vary in steps of 2° within the interval [0°, 180°], and the sensor-observed zenith angle was set to vary in steps of 2° within the interval [0°, 90°]. For lunar radiation, the lunar zenith angle was set to vary in steps of 2° within the interval [0°, 180°], and the sensor-observed zenith angle was set to vary in steps of 2° within the interval [0°, 90°]. The lunar phase angle was set to vary in steps of 15° within the interval [0°, 180°]. A total of 549 × 90 × 45 × (12 + 1) = 28,904,850 simulations were performed.

[0094] (4) Simulation based on machine learning models.

[0095] (3) After the simulation, a random forest model was used to fit the mapping relationships between atmospheric path radiation, balloon albedo, surface incident radiation, and atmospheric upward transmittance with aerosol optical thickness, atmospheric water vapor content, ozone content, radiation source angle, and sensor angle (model accuracy is as follows). Figure 3 (As shown). After the model was established, the global aerosol optical thickness, atmospheric water vapor content, and ozone content products provided by MERRA-2, the observation angle parameters of the VIIRS sensor, and the predicted global surface reflectance of the VIIRS DNB were used as model inputs to obtain the predicted values ​​of atmospheric path radiation, balloon albedo, surface incident radiation, and atmospheric upward transmittance on a global scale. The predicted values ​​were then substituted into equation (4) to obtain the predicted value of global on-board radiance in the VIIRS DNB band. Figure 4 The prediction results of this method and the actual observations in the VIIRS DNB band are presented respectively. Figure 5 This demonstrates the simulation results of on-board radiance of a large-scale satellite sensor obtained using this method. Figure 5 -(b)) and actual observed values ​​( Figure 5 The comparison results of -(a)) show that the simulation results are highly consistent with the actual observations in terms of both the magnitude and spatial distribution of the values.

[0096] This invention can use the surface reflectance products, atmospheric profile products, and atmospheric parameters from reanalysis data of on-orbit sensors to simulate on-board observations of optical satellite sensors. Compared with traditional simulation methods, this method solves the problem of sensors lacking corresponding global channel reflectance during pre-research, and significantly reduces the simulation time and computing resources required while meeting the accuracy requirements of sensor dynamic range settings.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

[0098] The above descriptions are merely some embodiments of the present invention. Those skilled in the art can make various modifications and improvements without departing from the inventive concept of the present invention, and these all fall within the scope of protection of the present invention.

Claims

1. A method for simulating radiance from on-board observations using a large-scale, rapid optical satellite sensor, characterized in that... Includes the following steps: Step S1: Construct a typical ground object spectral library, and select sensors whose spectral response function band range overlaps with that of the sensor to be simulated from the satellite sensor spectral response function library. Integrate the spectral response functions of both sensors with the ground object spectra in the ground object spectral library to obtain the band range. channel reflectivity : ； in, Represents the continuous spectral curve of ground features. Indicates band The spectral response function, λ max and λ min These represent the upper and lower bounds of the spectral response function's band range, respectively. A machine learning algorithm was used to fit the conversion formula between the two channel reflectances, and the fitted channel reflectances were obtained. ,in, This represents the channel reflectivity of the sensor that provides the reflectivity background field. This represents the conversion relationship between the fitted background field reflectance and the reflectance of a certain band channel; Step S2: Filter the global atmospheric profile dataset to select K atmospheric profile data points and construct an atmospheric profile dataset: Filter and remove profiles with relative humidity exceeding a specified value: In high-latitude regions, profiles are removed based on a set first threshold; in mid- and low-latitude regions, profiles are removed based on a set second threshold. Then remove profiles whose altitude is below the specified altitude threshold; Step S3, Radiative Transfer Simulation: Based on the atmospheric profile dataset constructed in step S2, the zenith angle of the radiation source is set to M values ​​that vary with a step size a in the interval [0º, 180º], the zenith angle observed by the sensor is set to N values ​​that vary with a step size b in the interval [0º, 90º], and the lunar phase angle observed by the sensor is set to L values ​​that vary with a step size c in the interval [0º, 180º]. A total of K×M×N×(L+1) simulations are performed, and the simulation results are used to construct the training dataset for the machine learning model. Set the total radiation at the sensor entrance pupil Total radiation at the sensor entrance pupil and sensor channel radiance : ； ； ； The radiation sources include the sun and the moon. Indicates the zenith angle of the sun. Indicates the zenith angle of the sensor. Indicates the surface reflectance at the observed target location. This indicates the solar radiation at the sensor's entrance pupil. This indicates lunar radiation at the sensor's entrance pupil. Indicates atmospheric path radiation. Indicates incident radiation on the Earth's surface. Indicates atmospheric upward radiation transmittance. Represents the spherical albedo of the lower surface of the atmosphere; Step S4, Simulation based on machine learning model: Atmospheric path radiation was established using machine learning algorithms. Surface incident radiation spherical albedo of the lower atmospheric surface Atmospheric up-rising radiation transmittance Nonlinear mapping relationship between simulation input and simulation input: ； ； ； ； Where AOD represents the aerosol optical thickness of the reanalysis data, and TQV represents the atmospheric water vapor content of the reanalysis data. This indicates the ozone content in the reanalyzed data. This represents the nonlinear relationship between aerosol optical thickness, atmospheric water vapor content, ozone content, solar zenith angle, sensor zenith angle, and atmospheric path radiation. This represents the nonlinear relationship between aerosol optical thickness, atmospheric water vapor content, ozone content, solar zenith angle, and incident radiation at the Earth's surface. This represents the nonlinear relationship between aerosol optical thickness, atmospheric water vapor content, ozone content, solar zenith angle, sensor zenith angle, and balloon surface albedo. This represents the nonlinear relationship between aerosol optical thickness, atmospheric water vapor content, ozone content, sensor zenith angle, and atmospheric transmittance. Step S5: Based on the orbital parameters of the satellite carrying the sensor to be simulated, obtain the sensor observation zenith angle and observation time for each pixel in the simulation area. Calculate the radiation source zenith angle at each pixel based on the observation time, and combine it with... , and Through nonlinear relationships ~ Calculate atmospheric path radiation Surface incident radiation spherical albedo of the lower atmospheric surface and atmospheric upward radiation transmittance The simulation results were substituted into the total radiation at the sensor entrance pupil. The expression is used to obtain the radiation at the entrance pupil of the sensor to be simulated based on a machine learning model. By integrating this with the sensor's spectral response function, the radiation at the satellite sensor's entrance pupil can be converted into the sensor's on-board radiance.

2. The method as described in claim 1, characterized in that, The first threshold includes the longitude difference threshold, latitude difference threshold, elevation difference threshold, and atmospheric water vapor content difference. The second threshold includes the longitude difference threshold, latitude difference threshold, elevation difference threshold, atmospheric water vapor content difference, and month difference threshold.

3. The method as described in claim 1 or 2, characterized in that, The specific zenith angle of the radiation source at each pixel is calculated based on the observation time: ； ； ； in, Indicates the lunar phase angle. Indicates the ecliptic longitude of the sun. The ecliptic longitude of the moon. Indicates the solar altitude angle. Indicates the local latitude. Indicates the solar declination. Indicates local solar time. It indicates the azimuth angle of the sun.

Images