A multi-platform satellite thermal infrared hyperspectral atmospheric ammonia monitoring method and system

Abstract

This invention provides a multi-platform satellite thermal infrared hyperspectral atmospheric ammonia monitoring method and system, belonging to the field of satellite remote sensing technology. The method includes: collecting thermal infrared hyperspectral data, atmospheric state parameters, surface parameters, and instrument characteristic parameters from geostationary or polar-orbiting satellites; based on the physical mechanism of thermal infrared radiative transfer, inputting preprocessed normalized parameters into a fast radiative transfer forward model to obtain a simulated spectrum consistent with the measured data format from geostationary or polar-orbiting satellites; based on the simulated and measured spectra, and combined with optimization estimation theory, constructing a cost function that includes observation error constraints and NH3 prior information constraints; using the Levenberg-Marquardt iterative algorithm to solve for the minimum value of the cost function, and inverting to obtain the atmospheric ammonia concentration profile; performing quality control and column concentration conversion on the inversion results, and verifying the accuracy using multi-source observation data to obtain an atmospheric ammonia concentration dataset.

Description

Technical Field

[0001] This invention relates to the field of satellite remote sensing technology, and in particular to a multi-platform satellite thermal infrared hyperspectral atmospheric ammonia monitoring method and system. Background Technology

[0002] As a key trace gaseous pollutant, atmospheric ammonia emissions exacerbate environmental problems such as acid rain and also affect the global nitrogen cycle and climate change. Therefore, achieving high-precision, wide-range, and timely monitoring of atmospheric ammonia concentration is of great significance.

[0003] Satellite remote sensing technology has become an important means of monitoring atmospheric ammonia due to its advantages such as wide coverage and strong spatiotemporal continuity. Currently, various satellite ammonia retrieval technologies have been developed both domestically and internationally, primarily based on infrared hyperspectral sensors carried by polar-orbiting satellites. However, existing technologies have significant shortcomings: First, they lack specificity. Most retrieval schemes are either adapted to foreign satellite sensors or limited to a single satellite model, lacking a universal adaptation design for thermal infrared hyperspectral sensors on different platforms, including my country's Fengyun satellites and other domestic and international geostationary and polar-orbiting satellites, thus failing to fully leverage the observation advantages of various satellites. Second, they suffer from poor compatibility. Existing methods are mostly designed for single sensors, lacking a unified retrieval framework adapted to multiple types and platforms of sensors, making it difficult to collaboratively utilize data from different satellites and resulting in low data utilization. Third, retrieval accuracy and reliability need improvement. The standardized preprocessing procedures of some schemes are incomplete, the cross-satellite adaptability of forward radiative transfer models and the stability of optimized retrieval algorithms are insufficient, and there is a lack of systematic quality control and multi-source data verification systems, making it difficult to meet the common needs of accurate monitoring from different satellites. Summary of the Invention

[0004] This invention provides a multi-platform satellite thermal infrared hyperspectral atmospheric ammonia monitoring method, applicable to thermal infrared hyperspectral sensors of various geostationary and polar-orbiting satellites. The method includes: collecting thermal infrared hyperspectral data, atmospheric state parameters, surface parameters, and instrument characteristic parameters from the geostationary or polar-orbiting satellites; based on the physical mechanism of thermal infrared radiative transfer, inputting the preprocessed normalized parameters into a fast radiative transfer forward model to obtain a simulated spectrum consistent with the measured data format of the geostationary or polar-orbiting satellites; based on the simulated spectrum and the measured spectrum of the satellites, and based on optimization estimation theory, constructing a cost function that includes observation error constraints and NH3 prior information constraints, and using the Levenberg-Marquardt iterative algorithm to solve for the minimum value of the cost function to obtain an atmospheric ammonia concentration profile; performing quality control and column concentration conversion on the obtained atmospheric ammonia column concentration, and verifying the accuracy by combining multi-source observation data to obtain an atmospheric ammonia concentration dataset.

[0005] According to one embodiment of the present invention, the method further includes constructing a fast radiative transfer forward model, specifically including: based on the principle of thermal infrared radiative transfer, using the observation geometric parameters from the standardized atmospheric profile, surface parameters, and instrument characteristic parameters obtained after standardized preprocessing as dynamic inputs, and using an absorption coefficient lookup table as auxiliary calculation parameters; decomposing the uplink radiation received by the satellite into three components, namely the component of surface-emitted radiation transmitted through the atmosphere, the atmospheric uplink emission component, and the component of atmospheric downlink radiation reflected by the surface and then transmitted through the atmosphere; and generating a simulated spectrum consistent with the format of the satellite measured data through forward mapping from atmospheric state variables to channel radiance or radiative spectrum.

[0006] According to one embodiment of the present invention, the method further includes constructing the cost function, specifically including: establishing a construction basis based on the deviation between the simulated spectrum and the satellite measured spectrum, using the optimization estimation theory as the core; introducing observation error constraints and NH3 prior information constraints, and weighting the observation bias and prior bias by combining the corresponding spectral error covariance matrix and NH3 prior covariance matrix respectively; and combining the weighted observation bias term with the prior bias term to construct a quadratic cost function that combines observation error constraints and prior information constraints.

[0007] According to one embodiment of the present invention, the process of quality control and column concentration conversion of the retrieved atmospheric ammonia column concentration, combined with multi-source observation data to verify accuracy, to obtain an atmospheric ammonia concentration dataset includes: performing multi-dimensional quality screening and column concentration conversion on the retrieved atmospheric ammonia concentration profile based on reliability indicators, inversion error thresholds, and effective pixel criteria of the inversion process, eliminating invalid and abnormal inversion results; selecting multi-source independent data matching the spatiotemporal range of satellite observations for accuracy verification, wherein the multi-source independent data includes high-precision ground and airborne observation data, mature polar-orbiting satellite inversion data, and atmospheric chemical transport model simulation data; wherein, the high-precision ground and airborne observation data are derived from solar tracking Fourier transform infrared spectroscopy technology and tunable... The data were obtained through field measurements using laser absorption spectroscopy or observations using airborne hyperspectral sensors. Mature polar-orbiting satellite inversion data were derived from industry-verified satellite ammonia inversion products, and atmospheric chemical transport model simulation data were derived from global or regional-scale atmospheric chemical transport model outputs. The consistency between the inversion results and multi-source validation data was quantitatively evaluated by calculating three statistical indicators: correlation coefficient, root mean square error, and spatiotemporal distribution similarity. Consistency was deemed satisfactory when the correlation coefficient was greater than the preset correlation coefficient, the root mean square error was less than the preset root mean square error, and the spatiotemporal distribution similarity was greater than the preset spatiotemporal distribution similarity. Ammonia concentration data that met the preset consistency criteria and of acceptable quality were retained, and metadata such as observation time, spatial coordinates, and inversion error were added to form a standardized atmospheric ammonia concentration dataset.

[0008] According to one embodiment of the present invention, the rapid radiative transfer forward model generates a comparison benchmark, comprising: inputting preprocessed standardized thermal infrared hyperspectral radiance spectral data, standardized atmospheric state parameters, and standardized surface parameters, combined with a constructed absorption coefficient lookup table, NH3 prior profile, and satellite instrument characteristic parameters, into the rapid radiative transfer forward model; the rapid radiative transfer forward model rapidly calculates the absorption effect of atmospheric gases in the thermal infrared band based on the absorption coefficient lookup table, coupling atmospheric radiative transfer and surface radiative emission processes, and simulating the generation of a simulated thermal infrared hyperspectral radiance spectrum of the top atmosphere; performing spectral resolution resampling and band matching processing on the simulated thermal infrared hyperspectral radiance spectrum to generate a radiance comparison spectrum that perfectly matches the detection band, spectral resolution, and observation geometry of the actual satellite observation spectrum, and using this comparison spectrum as a benchmark for deviation comparison with the actual satellite observation spectrum during the ammonia concentration inversion process.

[0009] According to one embodiment of the present invention, the method includes preprocessing the collected data, and performing radiometric calibration on the thermal infrared hyperspectral data to convert the original satellite detection data into radiance values ​​of the top layer of the atmosphere; performing cloud masking and clear-sky pixel determination to remove invalid pixels; completing bad pixel removal and effective detection band screening to generate standardized thermal infrared hyperspectral radiance spectral data.

[0010] According to one embodiment of the present invention, the method includes preprocessing the collected data, and performing multi-source data fusion and spatiotemporal registration on the atmospheric state parameters and the surface parameters to unify the spatiotemporal scale of the data and ensure that the two types of parameters are accurately matched with the spatiotemporal location of satellite thermal infrared hyperspectral observations; performing quality control to remove abnormal data and invalid values, and verifying the rationality of the parameters in conjunction with the observation scenario; and performing standardized format conversion and parameter optimization to generate standardized atmospheric state parameters and standardized surface parameters that are adapted to the input requirements of the fast radiative transfer forward model.

[0011] According to one embodiment of the present invention, the method includes constructing an absorption coefficient lookup table, further comprising: based on a row-by-row radiative transfer model, combining spectral line parameters from a high-resolution molecular absorption transmission database and spectral line files generated by a line-by-line absorption coefficient calculation program, coupling a preset version of a continuous absorption model for Moeller-Tobin water vapor and other gases, constructing an absorption optical thickness lookup table with a preset resolution; using the absorption optical thickness lookup table to simulate the continuous absorption effect of CO2, O3, and N2O gases in the thermal infrared band; using the absorption optical thickness lookup table to calculate the water vapor absorption coefficient of H2O under preset dry air conditions and preset humid air conditions, and estimating the water vapor absorption value of any concentration by linear interpolation.

[0012] According to one embodiment of the present invention, the method includes setting an NH3 prior profile and a covariance matrix, and further includes: the NH3 prior profile is derived from simulation data of an Earth system composition prediction model; profile data corresponding to pressure layers below a first preset pressure are extracted; the profile data undergoes outlier removal, regional clustering analysis based on geographical and meteorological features, and preset processing of equal-interval vertical interpolation at preset distances to generate a standardized NH3 prior profile adapted to the target observation area; the covariance matrix is ​​constructed based on the standardized NH3 prior profile, using a preset scaling factor to amplify the error variance of the prior profile, and setting a preset vertical correlation distance to describe the error correlation in the vertical direction of the profile; the preset scaling factor and the preset vertical correlation distance are determined according to the tropospheric distribution characteristics of NH3 and the convergence requirements of the inversion algorithm.

[0013] This invention also provides a multi-platform satellite thermal infrared hyperspectral atmospheric ammonia monitoring system, comprising: a data collection module for collecting thermal infrared hyperspectral data, atmospheric state parameters, surface parameters, and instrument characteristic parameters from geostationary or polar-orbiting satellites; a fast radiative transfer forward model module for receiving pre-processed normalized parameters based on the physical mechanism of thermal infrared radiative transfer and inputting a fast radiative transfer forward model with radiative component decomposition, cross-satellite adaptation, and comparative benchmark generation capabilities to obtain a simulated spectrum consistent with the measured data format of the geostationary and polar-orbiting satellites; an optimization estimation and inversion module for constructing a cost function containing observation error constraints and NH3 prior information constraints based on the simulated spectrum and the measured spectrum of the satellites, and using the optimization estimation theory, solving for the minimum value of the cost function using the Levenberg-Marquardt iterative algorithm to invert the atmospheric ammonia concentration profile; and a post-processing and error assessment module for quality control and column concentration conversion of the inverted atmospheric ammonia column concentration, verifying accuracy by combining multi-source observation data, and obtaining an atmospheric ammonia concentration dataset.

[0014] This invention effectively addresses the problems of poor compatibility, limited accuracy, and insufficient specificity in existing technologies through full-chain technology optimization and cross-satellite adaptation design, achieving significant application results. The method is applicable to thermal infrared hyperspectral sensors of various geostationary and polar-orbiting satellites. Through standardized preprocessing, radiative component decomposition, and cross-satellite parameter matching, it enables the collaborative utilization of sensor data from different platforms, significantly improving data utilization and observation coverage. The optimized fast radiative transfer forward model and dual-constraint optimization inversion algorithm, combined with a multi-source independent data verification system, improve the accuracy and stability of atmospheric ammonia concentration inversion, ensuring the physical rationality and numerical reliability of the results. Simultaneously, the technical solution balances operational processing efficiency with the needs of domestic applications, providing independent and controllable technical support for atmospheric environmental monitoring and promoting the development of atmospheric ammonia monitoring towards greater precision and multi-satellite collaboration. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating the multi-platform satellite thermal infrared hyperspectral atmospheric ammonia monitoring method provided by the present invention.

[0017] Figure 2 This is a flowchart of the method for constructing the fast radiative transfer forward model provided by the present invention.

[0018] Figure 3 This is a flowchart of the cost function construction method provided by the present invention.

[0019] Figure 4 This is a flowchart of obtaining an atmospheric ammonia concentration dataset through inversion, provided by the present invention.

[0020] Figure 5 This is a flowchart of the fast radiative transfer forward model generation comparison benchmark provided by the present invention.

[0021] Figure 6 This is a flowchart of the preprocessing of thermal infrared hyperspectral data provided by the present invention.

[0022] Figure 7 This is a flowchart of the preprocessing of atmospheric state parameters and surface parameters provided by the present invention.

[0023] Figure 8 This is a flowchart of constructing an absorption coefficient lookup table provided by the present invention.

[0024] Figure 9 This is a flowchart of the method for setting the NH3 prior profile and covariance matrix provided by the present invention.

[0025] Figure 10 This is a block diagram of the multi-platform satellite thermal infrared hyperspectral atmospheric ammonia monitoring system provided by the present invention.

[0026] Figure label:

[0027] 100: Multi-platform satellite thermal infrared hyperspectral atmospheric ammonia monitoring system; 110: Data collection module; 120: Fast radiative transfer forward model module; 130: Optimization estimation and inversion module; 140: Post-processing and error assessment module. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0029] The following is combined Figures 1 to 10 This invention describes a multi-platform satellite thermal infrared hyperspectral atmospheric ammonia monitoring method and system.

[0030] Figure 1 This is a flowchart illustrating the multi-platform satellite thermal infrared hyperspectral atmospheric ammonia monitoring method provided by the present invention. Figure 1 As shown, the method includes the following steps: In step S100, thermal infrared hyperspectral data, atmospheric state parameters, surface parameters, and instrument characteristic parameters of geostationary or polar-orbiting satellites are collected; In step S200, based on the physical mechanism of thermal infrared radiative transfer, the preprocessed normalized parameters are input into the fast radiative transfer forward model to obtain a simulated spectrum consistent with the measured data format of geostationary and polar-orbiting satellites; In step S300, based on the simulated spectrum and the measured spectrum of the satellites, and based on the optimization estimation theory, a cost function containing observation error constraints and NH3 prior information constraints is constructed, and the Levenberg-Marquardt iterative algorithm is used to solve for the minimum value of the cost function to obtain the atmospheric ammonia concentration profile; In step S400, the quality control and column concentration conversion of the obtained atmospheric ammonia column concentration are performed, and the accuracy is verified by combining multi-source observation data to obtain the atmospheric ammonia concentration dataset.

[0031] Specifically, step S100 is the foundational data support step in the entire monitoring process. The collected data covers satellite observation signals, atmospheric background, surface radiation characteristics, and instrument performance parameters, comprehensively supporting subsequent preprocessing and inversion calculations. These data come from a wide range of sources and have undergone targeted screening to ensure compatibility with the thermal infrared hyperspectral sensors of various geostationary and polar-orbiting satellites. This lays the foundation for the collaborative use of data from different platforms and avoids inversion biases caused by missing data types or insufficient compatibility.

[0032] Step S200 is a crucial bridge connecting the raw data and the inversion calculation. The preprocessing stage, through multi-source data fusion, spatiotemporal registration, and quality control, transforms the collected raw data into standardized parameters with a unified format and reliable accuracy, eliminating scale differences and error interference between different data sets. The three functions of the fast radiative transfer forward model work together: radiative component decomposition fully restores the physical process of thermal infrared radiative transfer; cross-satellite adaptation overcomes the application limitations of single sensors; and the generation of a comparison benchmark provides a unified reference for subsequent bias analysis. The final output simulated spectrum is consistent with the satellite's measured format, ensuring it can be directly used for subsequent error calculation and inversion optimization.

[0033] Step S300 is the crucial step for accurately retrieving ammonia concentration. Optimization estimation theory provides a scientific mathematical framework for the inversion. By constructing a dual-constraint cost function, it fully utilizes real-time information from satellite observation data while using prior NH3 information to avoid non-physical values ​​in the inversion results. The Levenberg-Marquardt iterative algorithm boasts advantages such as convergence stability and computational efficiency, enabling it to quickly solve for the minimum value of the cost function and accurately derive the vertical profile of atmospheric ammonia concentration, achieving a quantitative conversion from radiation signal to gas concentration.

[0034] Step S400 is the final step to ensure the reliability of monitoring results. Multi-dimensional quality control effectively eliminates invalid and abnormal profiles through constraints such as inversion reliability indicators and error thresholds, ensuring the physical rationality of the retained data. Multi-source observation data verification comprehensively evaluates the accuracy and consistency of the inversion results from three dimensions: measured benchmark, comparison with similar satellites, and physical model verification. The final standardized atmospheric ammonia concentration dataset, supplemented with complete metadata, can directly support the application needs of regional air quality monitoring, emission source inventory optimization, and other scenarios.

[0035] Figure 2 This is a flowchart illustrating the method for constructing the fast radiative transfer forward model provided by this invention. Figure 2 As shown, according to an embodiment of the present invention, the implementation further includes constructing a fast radiative transfer forward model, specifically including: in step S210, based on the principle of thermal infrared radiative transfer, using the observation geometric parameters from the standardized atmospheric profile, surface parameters, and instrument characteristic parameters obtained after standardized preprocessing as dynamic inputs, and using an absorption coefficient lookup table as auxiliary calculation parameters; in step S220, decomposing the uplink radiation received by the satellite into three components, namely, the component of surface-emitted radiation transmitted through the atmosphere, the atmospheric uplink emission component, and the component of atmospheric downlink radiation reflected by the surface and then transmitted through the atmosphere; in step S230, generating a simulated spectrum consistent with the format of the satellite measured data by forward mapping from atmospheric state variables to channel radiance or radiative spectrum.

[0036] Specifically, step S210 lays the foundation for the input system of the model, ensuring data adaptability and computational efficiency. The principle of thermal infrared radiation transmission is the physical basis of the model, determining the transmission law of radiation signals between the atmosphere and the Earth's surface. The design of all input parameters revolves around this principle. After standardized preprocessing, the three types of dynamically input parameters have a unified format and controllable accuracy. The standardized atmospheric profile includes information on atmospheric temperature, humidity, and the vertical distribution of gases such as H2O, CO2, and O3, which directly affects the absorption and scattering effects during radiation transmission. The surface parameters, including surface temperature, surface emissivity, and surface pressure, are key to characterizing the surface radiation characteristics. The accuracy of the surface emissivity directly determines the calculation accuracy of the surface radiation components. The observation geometric parameters include the observation zenith angle and the observation azimuth angle. The size of the zenith angle changes the path length of radiation through the atmosphere, thus affecting the degree of atmospheric attenuation of radiation. The azimuth angle is related to the directionality of radiation reflected from the surface. These two types of parameters together ensure the accuracy of the radiation transmission path calculation. The auxiliary calculation parameter absorption coefficient lookup table is key to improving the model's computational efficiency. Based on the line-by-line radiative transfer model, it contains gas absorption optical thickness data at a preset resolution, which can quickly provide the absorption coefficients of various gases under different atmospheric conditions. This avoids the tedious process of calculating the absorption effect line by line during model operation, significantly reducing the time required for single-pixel radiation simulation and meeting the needs of operational batch processing.

[0037] Step S220 accurately decomposes the uplink radiation components to fully reconstruct the physical process of thermal infrared radiation transmission, providing support for the accuracy of the simulated spectrum. The uplink radiation received by the satellite is a superposition of multiple radiation sources. Without component decomposition, aliasing errors will occur in the radiation simulation, affecting the accuracy of subsequent inversion. To quantify the contribution of each component, the following formula is used for calculation:

[0038] .

[0039] left side of the formula This represents the uplink radiation spectrum ultimately received by the satellite sensor. The first item on the right corresponds to the component of surface-emitted radiation after atmospheric transmission, where... It is the optical thickness of the entire atmosphere. The corresponding Planck function describes the thermal radiation characteristics of the Earth's surface. This is the overall atmospheric transmittance, which quantifies the effective contribution of surface thermal radiation after atmospheric attenuation during upward transmission. The second term on the right corresponds to the upward atmospheric emission component. The integral variable represents the optical thickness of each atmospheric layer. The integral term describes the total contribution of thermal emission radiation from each atmospheric layer after transmission through the upper atmosphere, reflecting the influence of atmospheric vertical structure on the observed spectrum. The third term on the right corresponds to the component of downward atmospheric radiation that is reflected from the Earth's surface and then transmitted through the atmosphere. For surface reflectance, this term is first calculated by integrating the total amount of radiation transmitted downwards from the atmosphere. After reflection at the surface, it is then attenuated by the transmittance of the entire atmosphere, ultimately quantifying the contribution of this reflected radiation to satellite observations. The precise decomposition and quantification of the three components solves the problem of incomplete physical processes in radiation simulation in existing technologies, making the simulated spectrum more closely match the physical nature of satellite measurements.

[0040] Step S230, which quantitatively maps atmospheric conditions to satellite observation signals, is the final step in model building. The forward mapping logic is based on the physical relationship of thermal infrared radiation transmission. It combines dynamically input atmospheric state variables such as ammonia concentration, temperature, and humidity with auxiliary parameters such as absorption coefficients. Through the radiation component calculation in step S220, it is transformed into channel radiance or radiation spectrum observable by satellite sensors. This mapping process needs to be adapted to the characteristics of different satellite sensors. By adjusting parameters such as spectral resolution resampling and band matching, it is ensured that the generated simulated spectrum is completely consistent with the satellite's measured data in terms of data format, detection band, and spectral resolution. For example, for the high spectral resolution characteristics of the Fengyun satellite's GIIRS sensor, more spectral details are preserved during the mapping process; for the observation band settings of the HIRAS sensor, the radiance calculation channel is adjusted to achieve cross-satellite adaptability. This mapping design not only solves the pain points of existing technologies, which are mostly designed for single sensors and have poor compatibility, but also provides a directly usable data foundation for comparing simulated and measured spectra in subsequent optimization inversion, ensuring the accuracy of bias calculation and the reliability of inversion results.

[0041] Figure 3 This is a flowchart of the cost function construction method provided by the present invention. Figure 3 As shown, according to an embodiment of the present invention, the implementation further includes constructing a cost function, specifically including: in step S310, based on the optimal estimation theory, establishing the construction basis with the deviation between the simulated spectrum and the satellite measured spectrum as the center; in step S320, introducing observation error constraints and NH3 prior information constraints, and weighting the observation bias and prior bias by combining the corresponding spectral error covariance matrix and NH3 prior covariance matrix respectively; in step S330, combining the weighted observation bias term and the prior bias term to construct a quadratic cost function that combines observation error constraints and prior information constraints.

[0042] Specifically, step S310 lays the main logical foundation for constructing the cost function. Optimal estimation theory provides a scientific mathematical framework, enabling the finding of an optimal balance between observed data and prior information, thus avoiding extreme values ​​or non-physical interpretations in the inversion results. The focus is on the deviation between the simulated spectrum and the satellite's measured spectrum because this deviation directly reflects the difference between the assumed atmospheric state and the actual atmospheric state. The simulated spectrum, generated by the fast radiative transfer forward model, carries information on state variables such as atmospheric ammonia concentration, temperature, and humidity. The measured spectrum, on the other hand, is the actual observation signal from the satellite sensor. The magnitude of this deviation determines the reasonableness of the atmospheric state assumption. Constructing a function around this deviation essentially minimizes this difference by adjusting the state variables, thereby achieving an accurate inversion of the actual atmospheric ammonia concentration.

[0043] Step S320 improves the reliability of the cost function through a dual-constraint mechanism, solving the inversion uncertainty problem caused by relying solely on observation data or prior information. The introduction of observation error constraints is based on the inherent characteristics of satellite sensors. The spectral error covariance matrix is ​​constructed by analyzing the spatiotemporal variation characteristics of brightness-temperature differences. The elements in the matrix quantify the observation noise level of different spectral channels. The preset noise threshold is determined based on sensor technical specifications; for example, the spectral noise of GIIRS and HIRAS sensors is controlled within 0.01K. The inverse of this matrix is ​​used to weight observation biases, meaning that spectral channels with lower noise and higher reliability have greater bias weights, while channels with higher noise have weaker weights, avoiding invalid observations interfering with the inversion results. The NH3 prior information constraint provides a reasonable boundary for the inversion. The prior covariance matrix is ​​constructed based on the error statistics of the Earth system composition prediction model simulation data, quantifying the uncertainty of the prior profile. Similarly, it is weighted by the inverse matrix; regions with more reliable prior information, such as areas with stable ammonia emissions, have greater weights, guiding the inversion results to converge to a physically reasonable range and avoiding distortion of the inversion results due to insufficient observation data or excessive noise.

[0044] Step S330 combines the weighted bias terms to form a complete quadratic cost function. This form conforms to the mathematical requirements of optimization estimation theory and is compatible with the solution logic of the Levenberg-Marquardt iterative algorithm. The specific expression of the cost function is as follows:

[0045] ,

[0046] In the formula, The cost function value, The state variables to be inverted include atmospheric ammonia concentration profiles, surface parameters, etc. The spectrum is the actual measured spectrum from the satellite. The simulated spectrum generated for the forward model, Fix parameters for the model, such as absorption coefficient and observation geometry. Let NH3 be the prior state variable. Here is the spectral error covariance matrix. Let be the prior covariance matrix of NH3. The first term is the weighted observation bias term, quantifying the difference between the simulated and measured spectra under observation error constraints; the second term is the weighted prior bias term, characterizing the deviation between the estimated state variable values ​​and the prior values ​​under prior information constraints. The quadratic structure ensures that the function has a unique minimum and can be quickly solved by differentiation, providing an efficient and stable computational foundation for subsequent iterative algorithms. This combined design fully utilizes the dynamic information of real-time observation data while avoiding the problem of multiple solutions in the inversion through prior constraints, ensuring that the inversion results closely match both the measured data and atmospheric physical laws.

[0047] Figure 4 This is a flowchart illustrating the process of obtaining an atmospheric ammonia concentration dataset through inversion, as provided by this invention. Figure 4 As shown, according to an embodiment of the present invention, the atmospheric ammonia column concentration obtained by inversion is subjected to quality control and column concentration conversion, and the accuracy is verified by combining multi-source observation data to obtain an atmospheric ammonia concentration dataset. This includes: in step S410, based on the reliability index of the inversion process, the inversion error threshold, and the effective pixel determination criteria, multi-dimensional quality screening and column concentration conversion are performed on the obtained atmospheric ammonia concentration profile to remove invalid and abnormal inversion results; in step S420, multi-source independent data matching the spatiotemporal range of satellite observations are selected for accuracy verification. The multi-source independent data includes high-precision ground and airborne observation data, mature polar-orbiting satellite inversion data, and atmospheric chemical transport model simulation data; wherein, the high-precision ground and airborne observation data originates from solar-tracking Fourier transform infrared spectroscopy. The data are obtained through field measurements using tunable laser absorption spectroscopy or observations using airborne hyperspectral sensors. Mature polar-orbiting satellite inversion data are sourced from industry-verified satellite ammonia inversion products, while atmospheric chemical transport model simulation data are sourced from global or regional-scale atmospheric chemical transport model outputs. The consistency between the inversion results and multi-source verification data is quantitatively evaluated by calculating three statistical indicators: correlation coefficient, root mean square error, and spatiotemporal distribution similarity. Consistency is deemed to be met when the correlation coefficient is greater than the preset correlation coefficient, the root mean square error is less than the preset root mean square error, and the spatiotemporal distribution similarity is greater than the preset spatiotemporal distribution similarity. In step S430, ammonia concentration data that meets the quality standards and satisfies the preset consistency criteria are retained, and metadata such as observation time, spatial coordinates, and inversion error are added to form a standardized atmospheric ammonia concentration dataset.

[0048] Specifically, step S410 completes the initial quality screening through multi-dimensional rigid constraints, including: removing data that has not converged within 10 iterations; ensuring that the root mean square error (RMSE) of the fit between the measured spectrum and the simulated spectrum is less than 0.2K, and reducing the data... The absolute difference between the prior surface temperature and the inverted surface temperature is less than 10K, eliminating invalid data at the source and laying a reliable foundation for subsequent accuracy verification. Reliability indicators for the inversion process include spectral degrees of freedom and iterative convergence. Spectral degrees of freedom are calculated by the sum of the diagonal elements of the average kernel function matrix, with a preset value greater than 1. A value less than 1 indicates insufficient effective information provided by the observation data, and the inversion results overly rely on prior information, requiring its rejection. Iterative convergence requires that the difference between the state variables before and after iteration, after normalization by the posterior error covariance matrix, is much smaller than the number of elements in the state variable vector. Profiles that do not meet this condition indicate that the algorithm has not found the optimal solution, and the inversion results lack stability. The inversion error threshold is preset at 200%, calculated by the square root of the diagonal elements of the inversion error covariance matrix. This threshold is determined based on industry-standard accuracy requirements and measured verification results. Profiles exceeding this threshold mean that uncertainty exceeds an acceptable range and cannot support the needs of precise monitoring. The criteria for determining effective pixels focus on the bottom-level value of the total column average kernel function, with a preset value greater than 0.1. This indicator reflects the sensitivity of the inversion algorithm to near-surface ammonia concentration. The near-surface area is the main emission and distribution area for ammonia. Pixels with a bottom-level value below 0.1 cannot effectively capture near-surface concentration changes, which can easily lead to data distortion. The three types of constraints complement each other, completing the screening from three dimensions: information sufficiency, result stability, and physical sensitivity, to ensure the basic reliability of the retained profile.

[0049] Step S420 integrates three dimensions: measured benchmarks, cross-platform comparisons, and physical corroboration, constructing a comprehensive multi-source verification system to fully quantify the accuracy and consistency of the inversion results. The data selection process must meet strict spatiotemporal matching requirements, with the time window controlled within one hour before and after the satellite observation time. The spatial range is registered using a regular 0.1°×0.1° latitude and longitude grid to ensure the spatiotemporal correlation between the verification data and satellite pixels, avoiding verification deviations due to spatiotemporal misalignment. All sources of multi-source independent data are mature industry solutions, while ample space for equivalent replacement is reserved: high-precision ground and airborne observation data, in addition to tunable laser absorption spectroscopy, are applicable to trace gas observation technologies with equivalent high precision, such as differential absorption spectroscopy and Fourier transform infrared spectroscopy; mature polar-orbiting satellite inversion data are not limited to specific products, as long as they are industry-verified and their accuracy indicators match the requirements of this invention; atmospheric chemical transport models can also be flexibly selected according to the observation area scale, with both local-scale high-precision models and global models meeting the physical corroboration requirements. Accuracy assessment is achieved through three types of statistical indicators for quantitative determination: preset correlation coefficient, preset root mean square error, and preset spatiotemporal distribution similarity. These are all determined based on the characteristics of multi-source validation data and the need for precision in atmospheric ammonia monitoring. The correlation coefficient reflects the degree of linear correlation between the inversion results and the validation data; a larger value indicates a more consistent trend. The root mean square error quantifies the numerical deviation between the two; a smaller value indicates higher quantitative accuracy. The spatiotemporal distribution similarity is calculated by comprehensively considering the overlap of spatial patterns and the consistency of time series data; a larger value indicates that the inversion results can more accurately reproduce the spatiotemporal distribution characteristics of ammonia. All three indicators must be met simultaneously to be considered consistent and qualified, avoiding misjudgments caused by the bias of a single indicator.

[0050] Step S430 completes data integration and standardization, forming a final dataset that is both usable and universal. Ammonia concentration data that meets quality and consistency standards requires supplementary complete metadata: observation time accurate to the minute to ensure the accuracy of time-series analysis; spatial coordinates clearly defining the latitude and longitude boundaries of pixels for easy overlay with other geographic data, ensuring accurate geographic positioning; and inversion error metadata including absolute and relative errors at each altitude layer, providing a reliability reference for data users. The integration process adopts a standardized format commonly used in the remote sensing data field. This format must support massive multidimensional data storage and have good cross-platform compatibility, adapting to mainstream remote sensing data processing software and geographic information systems. The final atmospheric ammonia concentration dataset achieves a unified format output of observation results from different satellite sensors, facilitating collaborative analysis and direct comparison of multi-sensor data. Through full-process quality control and multi-source verification, the accuracy and reliability of the data are ensured, directly supporting applications in multiple scenarios such as regional air quality early warning, emission source inventory optimization, and global nitrogen cycle research.

[0051] Figure 5This is a flowchart illustrating the generation of a comparative benchmark for the fast radiative transfer forward model provided by this invention. For example... Figure 5 As shown, according to an embodiment of the present invention, the fast radiative transfer forward model generates a comparison benchmark, including: in step S240, preprocessed standardized thermal infrared hyperspectral radiance spectral data, standardized atmospheric state parameters, and standardized surface parameters, combined with a constructed absorption coefficient lookup table, NH3 prior profile, and satellite instrument characteristic parameters, are input into the fast radiative transfer forward model; in step S250, the fast radiative transfer forward model rapidly calculates the absorption effect of atmospheric gases in the thermal infrared band based on the absorption coefficient lookup table, couples atmospheric radiative transfer with surface radiative emission processes, and simulates and generates a simulated thermal infrared hyperspectral radiance spectrum of the top atmosphere; in step S260, the simulated thermal infrared hyperspectral radiance spectrum is subjected to spectral resolution resampling and band matching processing to generate a radiance comparison spectrum that completely matches the detection band, spectral resolution, and observation geometry of the actual satellite observation spectrum, and this comparison spectrum is used as a benchmark for deviation comparison with the actual satellite observation spectrum during the ammonia concentration inversion process.

[0052] Specifically, step S240 constructs a comprehensive and standardized input parameter system, providing accurate data support for the generation of the comparison benchmark. All input parameters are preprocessed and standardized, with a unified format and controllable accuracy, avoiding simulation deviations caused by data heterogeneity. Standardized thermal infrared hyperspectral radiance spectrum data is the fundamental signal for radiation simulation, carrying the raw radiation information from satellite observations. Standardized atmospheric state parameters include information such as temperature, humidity, and vertical gas distribution, directly determining the absorption and transmission characteristics of atmospheric thermal infrared radiation. Standardized surface parameters cover surface temperature and emissivity, characterizing the surface radiation contribution. Absorption coefficient lookup tables provide efficient support for calculating atmospheric gas absorption effects, avoiding the tedious process of line-by-line calculations. The NH3 prior profile provides initial physical constraints for ammonia concentration inversion, ensuring that the simulated spectrum closely matches the actual distribution of ammonia. Satellite instrument characteristic parameters include observation geometry and band settings, providing a basis for subsequent matching processing and ensuring that the comparison benchmark is compatible with satellite observation characteristics. The coordinated input of various parameters comprehensively covers the main influencing factors of the radiative transmission process, laying the foundation for the accuracy of the simulated spectrum.

[0053] Step S250 generates a physically complete simulated spectrum by accurately calculating the gas absorption effect and coupled radiation process. The absorption coefficient lookup table is crucial for rapid calculation. Based on this table, the model can directly obtain the absorption optical thickness of gases such as H2O, CO2, O3, and N2O in the thermal infrared band, efficiently quantifying the absorption contribution of various gases, significantly improving computational efficiency, and meeting operational processing requirements. Simultaneously, the model rigorously couples atmospheric radiative transport and surface radiative emission processes. The atmospheric radiative transport process encompasses the transmission and attenuation of uplink and downlink radiation, while the surface radiative emission process reflects the surface's own thermal radiation characteristics. This coupling completely recreates the entire physical mechanism of thermal infrared radiation generated from the surface, transmitted through the atmosphere, to the satellite sensor, avoiding radiation simulation distortion caused by separating the two processes. The final generated simulated spectrum of thermal infrared hyperspectral radiance at the top of the atmosphere truly reflects the comprehensive radiation characteristics of the atmosphere-surface system, providing high-quality basic data for subsequent benchmark construction.

[0054] Step S260 transforms the simulated spectrum into a standard reference that can be directly used for deviation comparison through targeted adaptation processing. Spectral resolution resampling adjusts the spectral characteristics of different satellite sensors, such as the high spectral resolution of the GIIRS sensor and the specific resolution requirements of the HIRAS sensor, using interpolation or downsampling to ensure the resolution of the simulated spectrum is completely consistent with the satellite's actual measurements. Band matching processing precisely aligns the satellite's detection band range, ensuring a one-to-one correspondence between the observation channels of the comparison spectrum and the measured spectrum, avoiding misjudgments caused by band misalignment. Observation geometry matching, combined with satellite instrument characteristic parameters, adjusts the observation angle response of the simulated spectrum to make the geometric characteristics of the comparison spectrum consistent with the actual satellite observation scenario. The radiance comparison spectrum generated after these three types of processing achieves complete matching with the satellite's measured spectrum in terms of detection band, spectral resolution, and observation geometry, eliminating comparison obstacles caused by differences in instrument characteristics. This comparative spectrum serves as a benchmark for deviation comparison, providing a unified reference standard for subsequent optimization and inversion. This makes the deviation calculation between the simulated spectrum and the measured spectrum more accurate, effectively supporting the quantitative inversion of atmospheric ammonia concentration. At the same time, it solves the problem of limited inversion accuracy caused by insufficient sensor adaptability in existing technologies.

[0055] Figure 6 This is a flowchart of the preprocessing of thermal infrared hyperspectral data provided by the present invention. For example... Figure 6As shown, according to an embodiment of the present invention, the method includes preprocessing the collected data. For thermal infrared hyperspectral data: in step S211, radiometric calibration is performed to convert the original satellite detection data into radiance values ​​of the top layer of the atmosphere; in step S212, cloud masking and clear-sky pixel determination are performed to remove invalid pixels; in step S213, bad pixel removal and effective detection band screening are completed to generate standardized thermal infrared hyperspectral radiance spectral data.

[0056] Specifically, the radiometric calibration process in step S211 is a fundamental step in data preprocessing, providing a quantitative physical basis for all subsequent analyses. Raw satellite data is mostly composed of electrical signal counts, lacking direct physical meaning and unsuitable for direct use in radiative transfer simulation and concentration inversion. Through radiometric calibration, using the sensor's preset calibration coefficients and calibration model, the raw electrical signals are converted into radiance values ​​at the top of the atmosphere. This physical quantity directly reflects the intensity of thermal infrared radiation received by the satellite sensor, and the accuracy of its quantification directly affects the precision of subsequent correction and inversion, ensuring that all data are processed on a uniform physical scale. For the raw hyperspectral data, the Norton-Beer function is used as the apodization function for convolution processing to suppress the amplitude of ripples on both sides of the response function.

[0057] Step S212 accurately filters valid observation data through cloud masking and clear-sky pixel determination. Thermal infrared hyperspectral data is extremely sensitive to cloud cover. The presence of clouds can block surface radiation and generate strong thermal radiation itself, causing the observation data to fail to accurately reflect the radiation state of the atmosphere and the surface. Direct use of such data for inversion would result in serious errors. Cloud masking identifies and marks cloud-covered pixels by analyzing spectral characteristics, radiation intensity, and other indicators. Clear-sky pixel determination, based on preset rules, filters out valid pixels with no clouds or very low cloud cover, and removes invalid pixels marked as cloud-covered. This ensures that the retained data can accurately support atmospheric ammonia concentration inversion and avoids interference from invalid data in subsequent processes. The determination criterion is that at least 80% of the cloud detection product pixels in the corresponding area are marked as "clear sky" or "possibly clear sky" when they are considered valid pixels.

[0058] Step S213 completes the final optimization and standardization of the data, generating input data adapted to the forward model. During observation, satellite sensors may generate bad pixels due to instrument malfunctions, electronic noise, etc. These pixels have abnormal radiation values, which will affect the overall data quality if not handled. Bad pixels are repaired by methods such as interpolation of adjacent effective pixels and mean replacement to ensure the continuity and integrity of the data. The effective detection band selection is based on the band characteristics of thermal infrared hyperspectral data and the requirements for ammonia monitoring. Bands with low signal-to-noise ratio, severe interference from other gases, or insensitivity to ammonia are removed, and bands that can effectively reflect the absorption characteristics of ammonia are retained, specifically 955–975 cm⁻¹. -1 The bands enhance the relevance and effectiveness of the data. After repair and screening, standardized thermal infrared hyperspectral radiance spectrum data with uniform format, reliable quality, and strong relevance are finally generated, which fully adapts to the input requirements of the fast radiative transfer forward model and provides high-quality data support for the generation of subsequent simulated spectra.

[0059] Figure 7 This is a flowchart of the preprocessing of atmospheric state parameters and surface parameters provided by the present invention. For example... Figure 7 As shown, according to an embodiment of the present invention, the method includes preprocessing the collected data for atmospheric state parameters and surface parameters: In step S214, multi-source data fusion and spatiotemporal registration are performed to unify the spatiotemporal scale of the data and ensure that the two types of parameters are accurately matched with the spatiotemporal location of satellite thermal infrared hyperspectral observations; In step S215, quality control is performed to remove abnormal data and invalid values, and the parameters are validated for rationality in conjunction with the observation scenario; In step S216, standardized format conversion and parameter optimization are performed to generate standardized atmospheric state parameters and standardized surface parameters that are adapted to the input requirements of the fast radiative transfer forward model.

[0060] Specifically, step S214 addresses data heterogeneity through multi-source data fusion and spatiotemporal registration, laying the foundation for the collaborative utilization of both types of parameters and satellite observations. Atmospheric state parameters and surface parameters often originate from different platforms and observation methods. Data from different sources differ in temporal sampling frequency and spatial coverage scale. Direct use of these data can lead to spatiotemporal misalignment between the parameters and satellite thermal infrared hyperspectral observations, affecting the accuracy of radiative transfer simulations. Multi-source data fusion integrates the advantages of data from different sources, compensating for the observational limitations of a single data source and improving the completeness and reliability of the parameters. Spatiotemporal registration uses the spatiotemporal range of satellite thermal infrared hyperspectral observations as a benchmark, unifying the temporal and spatial resolutions of the two types of parameters. This ensures that the parameters accurately correspond to the satellite data in terms of observation time and area, avoiding simulation deviations caused by spatiotemporal mismatches and ensuring that all subsequent model input data are within a unified spatiotemporal framework.

[0061] Step S215 ensures the reliability and physical rationality of parameters through quality control and rationality verification. During data collection, factors such as instrument performance and observation environment may lead to abnormal data or meaningless invalid values ​​that exceed the normal range. Such data can directly interfere with subsequent model calculations and inversion results. The quality control step identifies and removes abnormal data and invalid values ​​by setting reasonable screening rules, reducing the negative impact of data noise. Rationality verification is carried out in conjunction with specific observation scenarios. Based on the geographical characteristics and climatic conditions of the observation area, it judges whether the parameter values ​​conform to physical laws and reality. For example, for surface parameters in arid areas, it verifies whether the humidity-related indicators are within a reasonable range; for atmospheric state parameters in high-altitude areas, it verifies whether the temperature, air pressure, and other indicators are consistent with the atmospheric characteristics of that altitude. This dual control ensures the quality of the retained parameters, providing a reliable foundation for subsequent parameter optimization.

[0062] Step S216 generates input data fully adapted to the forward model through standardized format conversion and parameter optimization. The fast radiative transfer forward model has specific requirements for the format of input parameters. Original parameters from different sources differ in data structure and storage format, making them unusable directly by the model. Standardized format conversion unifies various parameters into a model-recognizable format, eliminating adaptation barriers caused by format differences and ensuring smooth input of parameters for calculation. Parameter optimization, based on the computational needs of the forward model, makes targeted adjustments to the parameters, correcting potential biases and optimizing the numerical range and accuracy of the parameters. This makes the parameters more closely match the physical process of thermal infrared radiative transfer, improving the accuracy of the model simulation. After format conversion and optimization, standardized atmospheric state parameters and standardized surface parameters are finally generated. Atmospheric profile data covers 37 pressure layers within the range of 1–1000 hPa, with the US standard atmospheric profile used above 1 hPa. Surface emissivity needs to be interpolated to the NH3 thermal infrared absorption band, corresponding to wavenumbers 699.2, 826, 925, 1075, and 1204.8 cm⁻¹. -1 Furthermore, each inversion channel's state vector requires four coefficients, which are multiplied by the first four basis functions of the Legendre polynomial to describe the emissivity variation with wavenumber. This fully meets the input requirements of the fast radiative transfer forward model, providing high-quality and highly adaptable parameter support for the generation of subsequent simulated spectra.

[0063] The entire preprocessing process is progressively advanced around data synergy, reliability, and adaptability. Data synergy is achieved through multi-source fusion and spatiotemporal registration, data reliability is ensured through quality control and rationality verification, and model adaptability is achieved through format conversion and parameter optimization. This systematically solves the problems of heterogeneity, noise interference, and insufficient adaptability of the original atmospheric state parameters and surface parameters. It addresses the pain points of the imperfect standardized preprocessing process in existing technologies and provides a solid guarantee for the stable operation and accurate simulation of the subsequent fast radiative transfer forward model.

[0064] Figure 8 This is a flowchart of constructing an absorption coefficient lookup table provided by the present invention. For example... Figure 8 As shown, according to an embodiment of the present invention, the method includes constructing an absorption coefficient lookup table, further comprising: in step S241, based on a row-by-row radiative transfer model, combining spectral parameters from a high-resolution molecular absorption transmission database and spectral files generated by a line-by-line absorption coefficient calculation program, coupling a preset version of a continuous absorption model for Moeller-Tobin water vapor and other gases, constructing an absorption optical thickness lookup table with a preset resolution; in step S242, using the absorption optical thickness lookup table to simulate the continuous absorption effect of CO2, O3, and N2O gases in the thermal infrared band; in step S243, using the absorption optical thickness lookup table to calculate the water vapor absorption coefficient of H2O under preset dry air conditions and preset humid air conditions, and estimating the water vapor absorption value of any concentration by linear interpolation.

[0065] Specifically, step S241 lays a precise physical and data foundation for constructing the absorption coefficient lookup table, ensuring that the lookup table can efficiently provide reliable absorption optical thickness data. The line-by-line radiative transfer model possesses high-precision radiative transfer calculation capabilities, serving as the basis for ensuring the accuracy of absorption effect calculations. Its line-by-line calculation characteristic can accurately reconstruct the absorption process of gas molecules. The spectral parameters of the high-resolution molecular absorption transmission database contain key information such as the position and intensity of absorption lines of gas molecules at different wavenumbers. The line-by-line absorption coefficient calculation program generates detailed spectral files based on these parameters, providing data support for accurate calculation of gas absorption effects. A preset version of the continuous absorption model for water vapor and other gases, MT_CKD_3.4, is used to compensate for the shortcomings of discrete spectral line calculations in describing the continuous absorption effect of gases, fully reconstructing the absorption characteristics of water vapor and other gases in the thermal infrared band. By integrating the above elements, an absorption optical thickness lookup table with a preset resolution of 0.01 cm is constructed. -1 The resolution setting balances computational accuracy and query efficiency, ensuring that the accuracy requirements of radiation transmission simulation are met while avoiding problems such as excessively large lookup table size and long query time caused by excessively high resolution.

[0066] Step S242 utilizes the pre-constructed absorption optical thickness lookup table to efficiently simulate the continuous absorption effect of key gases, providing input for the radiative transfer model. H2O, CO2, O3, and N2O are gases that significantly affect radiative transfer in the thermal infrared band. Their continuous absorption effect directly alters the transmission path and intensity of radiation in the atmosphere, thus affecting the accuracy of the simulated spectrum. The absorption optical thickness lookup table pre-stores absorption optical thickness data for these gases under different conditions. This eliminates the need for complex line-by-line calculations in real-time during the radiative transfer simulation; absorption effect data under corresponding conditions can be quickly obtained through a lookup and matching process, significantly improving the computational efficiency of the forward model and meeting the needs of operational batch processing. Simultaneously, the high precision of the lookup table ensures that the simulated absorption effect closely matches actual atmospheric conditions, guaranteeing the accuracy of subsequent simulated spectra.

[0067] Step S243 focuses on water vapor as a key influencing factor, achieving rapid acquisition of the absorption coefficient for arbitrary water vapor concentrations through precise calculations and interpolation. Water vapor exhibits significant absorption in the thermal infrared band with a wide concentration variation range; therefore, accurate acquisition of its absorption coefficient is crucial for radiative transfer simulation. Preset dry air and preset humid air conditions provide clear boundary scenarios for the calculation of the water vapor absorption coefficient. The preset dry air condition corresponds to an H2O volume fraction of 1 ppm, and the preset humid air condition corresponds to an H2O volume fraction of 4 × 10⁻⁶. 4 Based on these two boundary conditions, the corresponding water vapor absorption coefficient can be directly obtained using an absorption optical thickness lookup table. The linear interpolation method, on the other hand, establishes a continuous correspondence between concentration and absorption coefficient based on the absorption coefficient data under the boundary conditions. This allows for rapid estimation of the absorption value at any water vapor concentration, ensuring computational accuracy while avoiding the tedious process of calculating each concentration individually. This approach enables the absorption coefficient lookup table to adapt to different atmospheric humidity scenarios, further enhancing its versatility and practicality, and providing strong support for the fast radiative transfer forward model to cope with complex and changing atmospheric environments.

[0068] The entire construction process first establishes a basic lookup table by integrating high-precision models, authoritative data, and practical algorithms. Then, the lookup table is used to efficiently simulate the continuous absorption effects of key gases. Finally, boundary condition calculations and interpolation methods are used to expand the applicability of the water vapor absorption coefficient, gradually improving the functionality of the absorption coefficient lookup table. This lookup table effectively solves the problems of excessive time consumption and insufficient accuracy in real-time calculation of gas absorption effects in existing forward models, providing support for the efficient and stable operation of rapid radiative transfer forward models, thereby ensuring the accuracy and timeliness of atmospheric ammonia concentration inversion.

[0069] Figure 9 This is a flowchart illustrating the method for setting the NH3 prior profile and covariance matrix provided by this invention. Figure 9As shown, according to an embodiment of the present invention, the method includes setting an NH3 prior profile and a covariance matrix, and further includes: in step S244, the NH3 prior profile is derived from simulation data of the Earth system composition prediction model, and profile data corresponding to the pressure layer below the first preset pressure is extracted. The profile data is subjected to outlier removal, regional clustering analysis based on geographical and meteorological characteristics, and preset processing of vertical interpolation at preset distances to generate a standardized NH3 prior profile adapted to the target observation area; in step S245, the covariance matrix is ​​constructed based on the standardized NH3 prior profile, and the error variance of the prior profile is amplified by a preset scaling factor. A preset vertical correlation distance is set to describe the error correlation in the vertical direction of the profile. The preset scaling factor and the preset vertical correlation distance are determined according to the tropospheric distribution characteristics of NH3 and the convergence requirements of the inversion algorithm.

[0070] Specifically, step S244 generates a standardized NH3 prior profile adapted to the target region through precise data source selection and multi-step data processing, providing reasonable initial physical constraints for inversion. The simulation data from the Earth system composition prediction model possesses advantages in global coverage and spatiotemporal continuity, providing basic information on the vertical distribution of ammonia in different regions and time periods. As a data source for the prior profile, this ensures the physical rationality of the initial constraints. The first preset pressure is set to 200 hPa, a value that aligns with the atmospheric distribution characteristics of ammonia, which is mainly concentrated in the troposphere below 200 hPa. Extracting profile data below this pressure allows focus on the main distribution areas, avoiding interference from irrelevant upper-level data. Outlier removal targets data points that may exceed the normal concentration range in the model simulation. By setting reasonable screening thresholds, outliers are removed, ensuring the reliability of the profile data. Regional clustering analysis based on geographical and meteorological characteristics groups regions with similar geographical conditions (such as topography and land use types) and meteorological backgrounds (such as circulation patterns and humidity conditions) into one category, enabling the prior profile to conform to the ammonia emission and diffusion patterns of different regions. Preset equal-interval vertical interpolation unifies the vertical resolution of the profile. The interpolation interval is 1km, ensuring that the data of each pressure layer is evenly distributed, which is convenient for subsequent inversion algorithm processing. The standardized NH3 prior profile generated after a series of processing meets the inversion requirements in terms of format, resolution and physical rationality, and can provide accurate initial guidance for the inversion process.

[0071] Step S245 constructs a covariance matrix by scientifically setting parameters to quantify the uncertainty of the prior profile, providing a quantitative basis for the prior constraints in the cost function. The covariance matrix describes the error distribution and correlation of the prior profile data at each altitude level. Its construction is based on the standardized NH3 prior profile, ensuring a high correlation between uncertainty quantification and the prior data itself. A preset scaling factor is used to amplify the error variance of the prior profile. This preset scaling factor is set to 350%, because the simulation data of the Earth system composition prediction model has inherent errors. Appropriately amplifying the error variance can avoid excessive constraints on the inversion results from prior information, reserving sufficient correction space for the observation data. A preset vertical correlation distance is used to describe the error correlation in the vertical direction of the profile. The maximum threshold of this preset vertical correlation distance is 3km, which represents the correlation degree between errors at adjacent altitude levels. This parameter setting aligns with the vertical distribution pattern of ammonia in the troposphere. Ammonia concentration exhibits continuous variation in the vertical direction, and errors at adjacent altitude levels often have a certain correlation. The vertical correlation distance can accurately characterize this correlation characteristic. The determination of the preset scaling factor and preset vertical correlation distance is not arbitrary, but fully combines the tropospheric distribution characteristics of NH3 and the convergence requirements of the inversion algorithm. This ensures that the inversion results will not deviate from the physical reality due to improper error quantification, and that the inversion algorithm will converge quickly to the optimal solution. This makes the constructed covariance matrix both physically reasonable and adaptable to the actual needs of the inversion process.

[0072] The entire setup process, through the standardization of prior profiles and the precise construction of the covariance matrix, forms a complete prior information system. This system provides scientific initial constraints and uncertainty quantification for optimal inversion, effectively avoiding the multiple solutions problem that may occur during the inversion process, guiding the inversion results to converge to a physically reasonable range, while taking into account the convergence efficiency and stability of the inversion algorithm. It addresses the pain point of insufficient stability in existing inversion algorithms and provides crucial prior support for the accurate inversion of atmospheric ammonia concentration.

[0073] This invention also provides a multi-platform satellite thermal infrared hyperspectral atmospheric ammonia monitoring system 100. Figure 10 This is a block diagram of the multi-platform satellite thermal infrared hyperspectral atmospheric ammonia monitoring system provided by the present invention. Figure 10As shown, the system includes: a data collection module 110, used to collect thermal infrared hyperspectral data, atmospheric state parameters, surface parameters, and instrument characteristic parameters from geostationary or polar-orbiting satellites; a fast radiative transfer forward model module 120, used to receive preprocessed normalized parameters based on the physical mechanism of thermal infrared radiative transfer and input them into a fast radiative transfer forward model with radiative component decomposition, cross-satellite adaptation, and comparative benchmark functions to obtain simulated spectra consistent with the measured data formats of geostationary and polar-orbiting satellites; an optimal estimation and inversion module 130, used to construct a cost function containing observation error constraints and NH3 prior information constraints based on the simulated spectrum and the measured spectrum of the satellite, and based on the optimal estimation theory, to solve for the minimum value of the cost function using the Levenberg-Marquardt iterative algorithm, and invert the atmospheric ammonia concentration profile; and a post-processing and error assessment module 140, used to perform quality control and column concentration conversion on the inverted atmospheric ammonia column concentration, and to verify the accuracy by combining multi-source observation data to obtain an atmospheric ammonia concentration dataset.

[0074] The Fengyun satellite thermal infrared hyperspectral atmospheric ammonia concentration monitoring system provided by this invention has each functional module corresponding one-to-one with the process of the aforementioned monitoring method. The module function implementation logic, technical details and corresponding steps in the method are completely consistent and can be directly referenced without further explanation.

[0075] The specific correspondences are as follows: the data collection module corresponds to the step of "collecting thermal infrared hyperspectral data, atmospheric state parameters, surface parameters, and instrument characteristic parameters from geostationary or polar-orbiting satellites" in the method; the fast radiative transfer forward model module corresponds to the relevant process of "constructing a fast radiative transfer forward model, generating simulated spectra, and establishing a benchmark" in the method; the optimization estimation and inversion module corresponds to the step of "constructing a cost function and using the Levenberg-Marquardt iterative algorithm to invert the atmospheric ammonia concentration profile" in the method; and the post-processing and error assessment module corresponds to the entire process of "quality control and column concentration conversion of atmospheric ammonia column concentration, verifying accuracy by combining multi-source observation data, and obtaining an atmospheric ammonia concentration dataset" in the method.

[0076] The specific implementation methods, technical parameters, and constraints of each module of the system are consistent with the detailed description in the method section. For relevant details, please refer directly to the corresponding content of the aforementioned method.

[0077] This invention systematically addresses the shortcomings of existing technologies, such as insufficient targeting of domestically produced satellite sensors, poor data compatibility across different platforms, and limited inversion accuracy and stability, through a standardized preprocessing workflow, a cross-satellite-adaptive rapid radiative transfer forward model, a dual-constraint optimization inversion algorithm, and a multi-source independent data verification system. The technical solution fully leverages the observational potential of the Fengyun satellite's GIIRS and HIRAS dual thermal infrared hyperspectral sensors, achieving collaborative utilization and efficient conversion of multi-satellite data. Furthermore, rigorous physical mechanism modeling and full-process quality control ensure the physical rationality and numerical reliability of atmospheric ammonia concentration monitoring results. This method and system provide independent and controllable technical support and high-quality data products for precise atmospheric environmental monitoring, emission source inventory optimization, and global nitrogen cycle research in my country, possessing significant practical value and broad application prospects.

[0078] 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.

Claims

1. A method for monitoring atmospheric ammonia by multi-platform satellite thermal infrared hyperspectral, characterized in that, The method, applicable to thermal infrared hyperspectral sensors for various geostationary and polar-orbiting satellites, includes: Collect thermal infrared hyperspectral data, atmospheric state parameters, surface parameters, and instrument characteristic parameters of the geostationary or polar-orbiting satellite; Based on the physical mechanism of thermal infrared radiative transfer, the preprocessed normalized parameters are input into the fast radiative transfer forward model to obtain a simulated spectrum consistent with the measured data format of the geostationary satellite and the polar-orbiting satellite; wherein, the construction of the fast radiative transfer forward model includes: Based on the principle of thermal infrared radiation transmission, the observation geometric parameters in the standardized atmospheric profile, surface parameters and instrument characteristic parameters obtained after standardized preprocessing are used as dynamic inputs, and the absorption coefficient lookup table is used as auxiliary calculation parameters. The uplink radiation received by the satellite is decomposed into three components: the component of radiation emitted from the Earth's surface after being transmitted through the atmosphere, the component of radiation emitted from the atmosphere, and the component of radiation emitted from the atmosphere after being reflected from the Earth's surface and then transmitted through the atmosphere. By forward mapping from atmospheric state variables to channel radiance or radiance spectrum, a simulated spectrum consistent with the format of satellite measured data is generated. Based on the simulated spectrum and the measured spectrum of the satellite, and based on the optimal estimation theory, a cost function is constructed that includes observation error constraints and NH3 prior information constraints, including: Based on the theory of optimal estimation, the construction basis is established with the deviation between the simulated spectrum and the satellite measured spectrum as the center; We introduce observation error constraints and NH3 prior information constraints, and then perform weighted processing on observation bias and prior bias by combining the corresponding spectral error covariance matrix and NH3 prior covariance matrix respectively. By combining the weighted observation bias term with the prior bias term, a quadratic cost function that combines observation error constraints and prior information constraints is constructed. Define the prior profile and covariance matrix of NH3, including: The NH3 prior profile is derived from simulation data of the Earth system composition prediction model. Profile data corresponding to the pressure layer below the first preset pressure is extracted. The profile data is subjected to outlier removal, regional clustering analysis based on geographical and meteorological characteristics, and preset processing of vertical interpolation at preset distances to generate a standardized NH3 prior profile adapted to the target observation area. The covariance matrix is ​​constructed based on the standardized NH3 prior profile. A preset scaling factor is used to amplify the error variance of the prior profile, and a preset vertical correlation distance is set to describe the error correlation in the vertical direction of the profile. The preset scaling factor and the preset vertical correlation distance are determined according to the tropospheric distribution characteristics of NH3 and the convergence requirements of the inversion algorithm. The Levenberg-Marquardt iterative algorithm is used to solve for the minimum value of the cost function, and the atmospheric ammonia concentration profile is obtained by inversion. The atmospheric ammonia column concentration obtained by inversion was subjected to quality control and column concentration conversion. The accuracy was verified by combining multi-source observation data to obtain an atmospheric ammonia concentration dataset.

2. The method of claim 1, wherein, The atmospheric ammonia column concentration obtained by inversion is subjected to quality control and column concentration conversion, and the accuracy is verified by combining multi-source observation data to obtain an atmospheric ammonia concentration dataset, including: Based on the reliability indicators, inversion error thresholds, and effective pixel criteria of the inversion process, the atmospheric ammonia concentration profile obtained by inversion is subjected to multi-dimensional quality screening and column concentration conversion to eliminate invalid and abnormal inversion results. Accuracy verification was conducted using multi-source independent data matching the spatiotemporal range of satellite observations. This multi-source independent data included high-precision ground and airborne observation data, mature polar-orbiting satellite inversion data, and atmospheric chemical transport model simulation data. The high-precision ground and airborne observation data originated from field measurements using solar-tracking Fourier transform infrared spectroscopy, tunable laser absorption spectroscopy, or observations by airborne hyperspectral sensors. The mature polar-orbiting satellite inversion data came from industry-verified satellite ammonia inversion products. The atmospheric chemical transport model simulation data came from the output of global or regional-scale atmospheric chemical transport models. The consistency between the inversion results and the multi-source verification data was quantitatively evaluated by calculating three statistical indicators: correlation coefficient, root mean square error, and spatiotemporal distribution similarity. Consistency was deemed satisfactory when the correlation coefficient was greater than a preset correlation coefficient, the root mean square error was less than a preset root mean square error, and the spatiotemporal distribution similarity was greater than a preset spatiotemporal distribution similarity. Retain ammonia concentration data that meets the preset consistency criteria and is of acceptable quality. Supplement the data with metadata such as observation time, spatial coordinates, and inversion error, and integrate them to form a standardized atmospheric ammonia concentration dataset.

3. The method according to claim 1, characterized in that, The fast radiative transfer forward model generates a benchmark, including: The preprocessed standardized thermal infrared hyperspectral radiance spectral data, standardized atmospheric state parameters, and standardized surface parameters, combined with the constructed absorption coefficient lookup table, NH3 prior profile, and satellite instrument characteristic parameters, are input into the fast radiative transfer forward model. The fast radiative transfer forward model calculates the absorption effect of atmospheric gases in the thermal infrared band based on an absorption coefficient lookup table, couples atmospheric radiative transfer with the surface radiative emission process, and simulates the generated thermal infrared hyperspectral radiance spectrum of the top atmosphere. The simulated thermal infrared hyperspectral radiance spectrum was resampled for spectral resolution and matched with bands to generate a radiance comparison spectrum that perfectly matches the detection band, spectral resolution, and observation geometry of the actual satellite observation spectrum. This comparison spectrum was used as a benchmark for comparing the deviation with the actual satellite observation spectrum during the ammonia concentration inversion process.

4. The method according to claim 1, characterized in that, The method includes preprocessing the collected data, specifically the thermal infrared hyperspectral data: Perform radiometric calibration to convert the raw satellite data into radiance values ​​for the top of the atmosphere; Perform cloud masking and clear sky pixel determination, and remove invalid pixels; Complete the removal of bad pixels and the screening of effective detection bands to generate standardized thermal infrared hyperspectral radiance spectral data.

5. The method according to claim 1, characterized in that, The method includes preprocessing the collected data, specifically the atmospheric state parameters and the surface parameters: Conduct multi-source data fusion and spatiotemporal registration to unify the spatiotemporal scale of the data and ensure that the two types of parameters are accurately matched with the spatiotemporal location of satellite thermal infrared hyperspectral observations; Perform quality control, remove abnormal data and invalid values, and verify the rationality of parameters in conjunction with the observation scenario; Standardized format conversion and parameter optimization are performed to generate standardized atmospheric state parameters and standardized surface parameters that are adapted to the input requirements of the fast radiative transfer forward model.

6. The method of claim 1, wherein, The method includes constructing an absorption coefficient lookup table, and further includes: Based on the line-by-line radiative transfer model, combined with the spectral parameters of the high-resolution molecular absorption transmission database and the spectral file generated by the line-by-line absorption coefficient calculation program, a preset version of the continuous absorption model of Moeller-Tobin water vapor and other gases is coupled to construct an absorption optical thickness lookup table with a preset resolution. The absorption optical thickness lookup table is used to simulate the continuous absorption effect of CO2, O3 and N2O gases in the thermal infrared band; The water vapor absorption coefficient of H2O under preset dry air conditions and preset humid air conditions is calculated using the absorption optical thickness lookup table, and the water vapor absorption value of any concentration is estimated by linear interpolation.

7. A multi-platform satellite thermal infrared hyperspectral atmospheric ammonia monitoring system, characterized in that, include: The data collection module is used to collect thermal infrared hyperspectral data, atmospheric state parameters, surface parameters, and instrument characteristic parameters from geostationary or polar-orbiting satellites. The fast radiative transfer forward model module is used to receive preprocessed normalized parameters based on the physical mechanism of thermal infrared radiative transfer and input them into a fast radiative transfer forward model that has radiative component decomposition capabilities, cross-satellite adaptation capabilities, and generates a comparison benchmark, thereby obtaining a simulated spectrum consistent with the measured data format of the geostationary satellite and the polar-orbiting satellite; wherein, constructing the fast radiative transfer forward model includes: Based on the principle of thermal infrared radiation transmission, the observation geometric parameters in the standardized atmospheric profile, surface parameters and instrument characteristic parameters obtained after standardized preprocessing are used as dynamic inputs, and the absorption coefficient lookup table is used as auxiliary calculation parameters. The uplink radiation received by the satellite is decomposed into three components: the component of radiation emitted from the Earth's surface after being transmitted through the atmosphere, the component of radiation emitted from the atmosphere, and the component of radiation emitted from the atmosphere after being reflected from the Earth's surface and then transmitted through the atmosphere. By forward mapping from atmospheric state variables to channel radiance or radiance spectrum, a simulated spectrum consistent with the format of satellite measured data is generated. The optimal estimation inversion module is used to construct a cost function based on the simulated spectrum and the satellite's measured spectrum, and based on optimal estimation theory, which includes observation error constraints and NH3 prior information constraints. This function includes: Based on the theory of optimal estimation, the construction basis is established with the deviation between the simulated spectrum and the satellite measured spectrum as the center; We introduce observation error constraints and NH3 prior information constraints, and then perform weighted processing on observation bias and prior bias by combining the corresponding spectral error covariance matrix and NH3 prior covariance matrix respectively. By combining the weighted observation bias term with the prior bias term, a quadratic cost function that combines observation error constraints and prior information constraints is constructed. Define the prior profile and covariance matrix of NH3, including: The NH3 prior profile is derived from simulation data of the Earth system composition prediction model. Profile data corresponding to the pressure layer below the first preset pressure is extracted. The profile data is subjected to outlier removal, regional clustering analysis based on geographical and meteorological characteristics, and preset processing of vertical interpolation at preset distances to generate a standardized NH3 prior profile adapted to the target observation area. The covariance matrix is ​​constructed based on the standardized NH3 prior profile. A preset scaling factor is used to amplify the error variance of the prior profile, and a preset vertical correlation distance is set to describe the error correlation in the vertical direction of the profile. The preset scaling factor and the preset vertical correlation distance are determined according to the tropospheric distribution characteristics of NH3 and the convergence requirements of the inversion algorithm. The Levenberg-Marquardt iterative algorithm is used to solve for the minimum value of the cost function, and the atmospheric ammonia concentration profile is obtained by inversion. The post-processing and error assessment module is used to perform quality control and column concentration conversion on the retrieved atmospheric ammonia column concentration, and to verify the accuracy by combining multi-source observation data to obtain an atmospheric ammonia concentration dataset.

Images