A hyperspectral atmospheric methane concentration retrieval method and system for high-aerosol scenarios

Abstract

The present application relates to the technical field of atmospheric remote sensing monitoring, and discloses a hyperspectral atmospheric methane concentration inversion method and system for high aerosol scenes, which solves the problems of low inversion accuracy and poor adaptability of hyperspectral imagers in high aerosol and heterogeneous surfaces in the prior art. The method comprises the following steps: obtaining atmospheric parameters and multi-band surface albedo data and processing, constructing a radiation transfer model to output a forward simulation vector; constructing a lookup table of methane and interference gas optical absorption coefficients containing different temperature and pressure based on CLARS-GFIT; correcting the forward simulation model using GIFT3 to quantify the influence of aerosols, and generating a forward simulation spectrum combined with a spectral response function; adjusting the state variables to convergence through an iterative maximum posterior algorithm to obtain the methane concentration. The present application improves the inversion accuracy of high aerosol scenes, adapts to heterogeneous surfaces such as coal mine areas, matches the characteristics of AHSI, and takes into account efficiency, which can provide support for methane point source monitoring.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of atmospheric remote sensing monitoring, in particular to a hyperspectral atmospheric methane concentration inversion method and system for a high aerosol scene. BACKGROUND

[0002] Methane is the second largest anthropogenic greenhouse gas after carbon dioxide. Methane leakage in the process of underground coal mining and processing, oil and chemical production and transportation can easily form high-concentration point source plume emissions, which not only causes energy waste and environmental pollution, but also exacerbates the greenhouse effect. Therefore, accurate detection and quantification of such methane point source emissions are of great significance to climate change response, environmental protection and sustainable development.

[0003] In the prior art, the methane concentration inversion accuracy is still low, and the following problems exist:

[0004] The current mainstream methane point source inversion relies on statistical methods such as matched filtering and iterative lognormal matched filtering. Such methods assume that the background spectrum is uniform and the methane emission is sparse, and can achieve effective inversion in areas with uniform ground surface and stable atmosphere. However, in coal mine areas and other heterogeneous ground surfaces, the background spectrum is highly heterogeneous, and the background uniformity assumption of statistical methods is invalid, resulting in a significant increase in inversion error.

[0005] In addition, the differential absorption spectroscopy technology is a physical inversion method based on the Lambert-Beer law. Although it has been widely used in medium-resolution satellite and airborne hyperspectral observations, the spectral resolution and signal-to-noise ratio of AHSI (Atmospheric Hyperspectral Imager) are significantly lower than those of specialized greenhouse gas monitoring instruments. The DOAS (Differential Optical Absorption Spectroscopy) method cannot effectively separate the weak absorption signal of methane from the ground surface radiation background, resulting in large concentration inversion deviation when directly applied, which cannot meet the precise quantification requirements of point sources in coal mine areas. The full-physical inversion method can reduce background interference from the mechanism by modeling the atmospheric radiation transmission of each pixel, and has shown high precision advantage in medium-resolution satellite methane column concentration inversion. However, for AHSI (Atmospheric Hyperspectral Imager) and other hyperspectral satellites, the full-physical method still faces technical challenges: first, the reduced spectral resolution of AHSI (Atmospheric Hyperspectral Imager) leads to a decrease in the calculation accuracy of gas absorption cross section, and a higher resolution absorption coefficient database needs to be constructed; second, the large amount of hyperspectral data makes the iterative optimization calculation of traditional full-physical methods inefficient, which is difficult to adapt to the business inversion of large-scale coal mine areas; third, the state variable system is not optimized for high aerosol scenes, and the variability of un-inverted atmospheric parameters is not included in the error matrix, resulting in insufficient stability of the inversion results.

[0006] Therefore, in order to solve the problems in the prior art, the present application provides a hyperspectral atmospheric methane concentration retrieval method and system for a high aerosol scene. SUMMARY

[0007] In view of the problems in the prior art, the present application aims to provide a hyperspectral atmospheric methane concentration retrieval method and system for a high aerosol scene.

[0008] To achieve the above object, the present application provides the following technical solutions.

[0009] A hyperspectral atmospheric methane concentration retrieval method for a high aerosol scene, comprising:

[0010] A prior data acquisition step acquires atmospheric parameter data and multi-band ground surface albedo data of a target area, performs uniform pressure level adaptation processing on the atmospheric parameter data, performs methane short-wave infrared absorption spectrum interpolation processing on the ground surface albedo data, constructs a radiation transfer model through the processed data, and outputs a forward simulation vector;

[0011] A look-up table construction step constructs a short-wave infrared hyperspectral optical absorption coefficient look-up table including methane and interfering gases and containing different temperature and pressure conditions based on CLARS-GFIT;

[0012] A forward simulation step constructs a hyperspectral imaging forward simulation model according to the forward simulation vector, corrects the forward simulation model through a GFIT3 model to quantify the influence of high aerosol on radiation transfer, and combines a spectral response function of a hyperspectral sensor to perform convolution processing on a forward simulation spectrum vector, thereby generating a forward simulation spectrum;

[0013] An iterative maximum posteriori retrieval step determines state variables including methane concentration, interfering gas concentration, and spectral parameters, adjusts the state variables through an iterative maximum posteriori algorithm, minimizes a cost function of a measured spectrum and the forward simulation spectrum, simplifies the state variables, and incorporates the influence of un-retrieved atmospheric parameters into an error matrix, iteratively solves until convergence, and obtains a methane concentration retrieval result.

[0014] Preferably, in the prior data acquisition step, the atmospheric parameter data includes air temperature, atmospheric humidity, trace gas concentration, ground surface temperature, and ground surface pressure, the uniform pressure level adaptation processing maps the atmospheric parameter data to a pressure level preset by a radiation transfer model to ensure that the atmospheric parameter of each layer is matched with the vertical resolution of the model, and the interpolation processing of the multi-band ground surface albedo data maps discrete band albedo to the methane short-wave infrared absorption spectrum to ensure that the pre-processed data is directly adapted to the radiation transfer model.

[0015] Preferably, in the look-up table constructing step, the interference gas is a gas affecting short-wave infrared spectrum absorption, and in the look-up table calling step, the optical absorption coefficients of methane and the interference gas under corresponding conditions are obtained through interpolation calculation according to real-time temperature and pressure data of the target region, and are input into the spectrum absorption characteristic model constructing as basic data in the forward simulation.

[0016] Preferably, the forward simulation step comprises: the GFIT3 model modifies the radiation transfer equation of the forward simulation model, introduces parameters representing aerosol scattering and absorption characteristics, corrects the transmission process of short-wave infrared radiation in a high aerosol scene, and constructs the spectrum response function based on the imaging principle of the hyperspectral sensor, and convolves the high-resolution spectrum vector generated by the forward simulation, so that the characteristics of the forward simulation spectrum and the measured spectrum of the sensor are consistent.

[0017] Preferably, in the iterative maximum posterior inversion step, the state variables further comprise polynomial fitting coefficients and spectrum offset terms, the cost function in the iteration is determined by the observation errors of the measured and simulated spectra and the prior errors of the state variables, the state variables simplify and retain the core parameters of methane concentration and surface albedo, and the convergence condition is set to be that the change amount of the state variables in adjacent iterations is less than a set threshold.

[0018] Preferably, the radiation transfer model is configured with:

[0019] ;

[0020] wherein, represents a high-resolution backscattering simulation radiation amount at a wavelength , is a solar incident radiation intensity, is an air mass factor of a vertical atmosphere =[1,72] layer, is a default optical thickness of each trace gas element from the U.S. standard atmosphere, wherein the initial value of atmospheric CH4 is set to 2000ppb, is a scaling factor applied to the optical depth, is a polynomial coefficient representing a low-frequency spectrum feature, wherein represents state variables, including gas concentrations, atmospheric and surface parameters, is a model input parameter, and k is a preset number.

[0021] Preferably, the cost function of the iterative maximum posterior algorithm is configured as:

[0022] ;

[0023] Wherein, J(x) is a cost function, y is an AHSI (hyperspectral imager) observed true spectrum, F is a forward model simulated AHSI (hyperspectral imager) imaging spectrum, b is a model input parameter, x is a state variable, x a is a state variable prior value, S a is an error covariance matrix of the state variable prior value, S ɛ is an observed spectrum error covariance matrix.

[0024] Preferably, the maximum posterior algorithm iteration is configured with:

[0025] ;

[0026] Wherein, x i+1 represents a state variable vector to be inverted at the i+1th iteration, x i represents a state variable vector to be inverted at the ith iteration, K represents a Jacobian matrix, that is, a partial derivative matrix of the forward simulation model with respect to the state variable x, represents the influence degree of a small change in the state variable on the simulated spectrum, and γ is an iteration adjustment parameter, used to dynamically adjust the trust region of the iteration step size, c represents a bias term, x a represents a state variable prior value vector, that is, an initial reference value of the state variable.

[0027] A hyperspectral atmospheric methane concentration inversion system for a high aerosol scene, comprising:

[0028] An a priori data acquisition module acquires atmospheric parameter data and multi-band ground surface albedo data of a target area, performs uniform pressure level adaptation processing on the atmospheric parameter data, and performs methane short-wave infrared absorption spectrum interpolation processing on the ground surface albedo data, constructs a radiation transfer model through the processed data, and outputs a forward simulation vector;

[0029] A lookup table construction module constructs a short-wave infrared hyperspectral optical absorption coefficient lookup table including methane and interfering gases and containing different temperature and pressure conditions based on CLARS-GFIT;

[0030] A forward simulation module constructs a hyperspectral imaging forward simulation model according to the forward simulation vector, corrects the forward simulation model through a GFIT3 model to quantify the influence of high aerosols on radiation transfer, and combines a spectral response function of a hyperspectral sensor to perform convolution processing on the forward simulation spectrum vector, thereby generating a forward simulation spectrum;

[0031] The iterative maximum posterior inversion module determines state variables including methane concentration, interfering gas concentration and spectral parameters, adjusts the state variables through an iterative maximum posterior algorithm and minimizes a cost function of the measured spectrum and the forward simulation spectrum, simplifies the state variables and includes the influence of un-inverted atmospheric parameters into an error matrix, iteratively solves to convergence, and obtains a methane concentration inversion result.

[0032] The present application has the following advantages:

[0033] (1) Significantly improve the inversion accuracy of high aerosol scenes, solve the problem of radiation transmission distortion, the present application corrects the radiation transmission equation of the forward simulation model through the GFIT3 model, introduces parameters representing the scattering and absorption characteristics of aerosols, quantifies the attenuation effect of high aerosols on short-wave infrared radiation transmission, directly addresses the radiation interference pain point of aerosol dense areas such as coal mine areas, and avoids the deviation between simulated spectrum and measured spectrum caused by ignoring the effect of aerosols in traditional radiation transmission model.

[0034] (2) Adapt to heterogeneous terrain, reduce the uncertainty of inversion caused by background spectrum interference, the present application breaks through the limitation of the existing statistical method background uniformity assumption, and adapts to heterogeneous terrain through two technical designs: first, in the data preprocessing stage, the multi-band surface albedo data is interpolated to the methane short-wave infrared absorption spectrum segment, ensuring that the spatial heterogeneity of the surface albedo is included in the radiation transmission modeling; second, in the inversion stage, the state variables are simplified to methane column concentration and surface albedo core parameters, directly quantifying the spectral interference of heterogeneous underlying surface such as coal mine area photovoltaic panel, mountain shadow and water body, effectively solving the core problem of confusion between methane absorption signal and surface interference signal under heterogeneous terrain, and improving the stability of methane point source detection in complex areas such as coal mine areas.

[0035] (3) Accurately match the characteristics of AHSI (hyperspectral imager) sensor, solve the problem of hyperspectral data inversion adaptability, the present application constructs a CLARS-GFIT method based on GFIT3 model, and the spectral resolution is 0.01cm -1The gas optical absorption coefficient lookup table is used for making up for the low spectral resolution of AHSI (hyperspectral imager) than that of a medium resolution satellite, ensuring accurate calculation of the absorption characteristics of methane and interfering gases, and constructing a Gaussian spectral response function based on the half-width full-width parameter of AHSI (hyperspectral imager), and the convolution processing is performed on the high-resolution spectral vector generated by forward simulation, so that the wavelength dimension and signal characteristics of the forward simulation spectrum and the measured spectrum of AHSI (hyperspectral imager) are completely matched, the inversion deviation caused by poor sensor adaptability of the traditional DOAS (Differential Optical Absorption Spectroscopy, Differential Optical Absorption Spectroscopy) method is avoided, the sensor adaptability is further strengthened through the binding of the radiation transmission model parameter and the observation angle and the spectral range of AHSI (hyperspectral imager), and it is ensured that the inversion result can be directly connected to the measured data of AHSI (hyperspectral imager).

[0036] (4) Considering the inversion accuracy and calculation efficiency, meeting the demand of business application, under the full physical inversion framework, the efficiency is improved through state variable simplification and error matrix integration, the initial 11-dimensional state variable is simplified, the iterative calculation amount is greatly reduced, the variability of the atmospheric parameters not participating in the inversion is calculated through the measured and simulated spectral residual under the background concentration scene, and is included in the observation error covariance matrix, so that the redundant calculation of non-core parameters is avoided. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 It is a flow chart of a hyperspectral atmospheric methane concentration inversion method for a high aerosol scene according to the present application;

[0038] Figure 2 It is a spectral response function schematic diagram of an embodiment of the present application;

[0039] Figure 3 It is an imaging simulation spectrum schematic diagram of an embodiment of the present application;

[0040] Figure 4 It is a methane concentration random enhancement spectral simulation scatter diagram of an embodiment of the present application;

[0041] Figure 5 It is a WRF-LES simulation point source plume concentration enhancement spectral simulation schematic diagram of an embodiment of the present application;

[0042] Figure 6 It is a WRF-LES point source plume spectral simulation under a satellite real spectrum of an embodiment of the present application. DETAILED DESCRIPTION

[0043] An embodiment of the present application proposes a hyperspectral atmospheric methane concentration inversion method for a high aerosol scene, comprising:

[0044] The prior data acquisition step acquires atmospheric parameter data and multi-band ground surface albedo data of a target area, performs uniform pressure level adaptation processing on the atmospheric parameter data, performs methane short-wave infrared absorption spectrum interpolation processing on the ground surface albedo data, constructs a radiation transmission model through the processed data, and outputs a forward simulation vector;

[0045] The look-up table construction step constructs a short-wave infrared high-spectral optical absorption coefficient look-up table including methane and interfering gases and containing different temperature and pressure conditions based on CLARS-GFIT.

[0046] The forward simulation step constructs a hyperspectral imaging forward simulation model according to the forward simulation vector, corrects the forward simulation model through a GFIT3 model to quantify the influence of high aerosols on radiation transmission, and generates a forward simulation spectrum through convolution processing of the forward simulation spectrum vector in combination with a spectral response function of a hyperspectral sensor.

[0047] The iterative maximum posterior inversion step determines state variables including methane concentration, interfering gas concentration, and spectral parameters, adjusts the state variables through an iterative maximum posterior algorithm, minimizes a cost function of a measured spectrum and the forward simulation spectrum, simplifies the state variables, and includes the influence of un-inverted atmospheric parameters into an error matrix, iteratively solves to convergence, and obtains a methane concentration inversion result.

[0048] Atmospheric and ground surface parameters are important prior data for atmospheric radiation transmission modeling. This embodiment intends to use the ECMWF (European Centre for Medium-Range Weather Forecasts) simulation and MODIS (Moderate Resolution Imaging Spectroradiometer) inversion product data. The atmospheric parameter data of the research area will be collected according to the observation time (about 11-12 o'clock Beijing time) of GF-5BAHSI (hyperspectral imager). In addition, all the atmospheric parameter data will be interpolated to the uniform pressure layer of the radiation transmission model. The MODIS ground surface albedo product only provides discrete band (0.648, 0.858, 0.470, 0.555, 1.240, 1.640 and 2.13 μm) data, and this embodiment will obtain the ground surface albedo of the CH4 short-wave infrared absorption spectrum through an interpolation method.

[0049] Since the absorption intensity of CH4 and other gas molecules in the atmosphere is affected by both the gas temperature and the gas pressure, this embodiment will use the CLARS-GFIT method to construct a look-up table database of short-wave infrared high-spectral resolution gas optical absorption coefficients including different temperatures and pressures. Thus, in the subsequent calculation of the optical absorption coefficients of atmospheric CH4 and other gases, the results can be quickly obtained through interpolation according to the temperature and pressure.

[0050] Preferably, in the prior data acquisition step, the atmospheric parameter data includes air temperature, atmospheric humidity, trace gas concentration, surface temperature and surface pressure, the unified atmospheric pressure level adaptation processing maps the atmospheric parameter data to the atmospheric pressure level preset by the radiation transfer model, ensuring that each atmospheric parameter matches the model vertical resolution, and the interpolation processing of the multi-band surface albedo data maps the discrete waveband albedo to the methane shortwave infrared absorption spectrum, ensuring that the preprocessed data is directly adapted to the radiation transfer model.

[0051] Specifically, in the lookup table construction step, the interference gas is a gas that affects shortwave infrared spectrum absorption, and when the lookup table is called, the optical absorption coefficients of methane and the interference gas under corresponding conditions are obtained through interpolation calculation according to real-time temperature and pressure data of the target area, and are input into the spectrum absorption characteristic model construction as basic data in the forward simulation.

[0052] Specifically, the forward simulation step includes that the GFIT3 model modifies the radiation transfer equation of the forward simulation model, introduces parameters representing the scattering and absorption characteristics of aerosols, corrects the transmission process of shortwave infrared radiation in a high aerosol scene, the spectrum response function is constructed based on the imaging principle of the hyperspectral sensor, and the high-resolution spectrum vector generated by the forward simulation is convolved to make the characteristics of the forward simulation spectrum consistent with the characteristics of the measured spectrum of the sensor.

[0053] In the SWIR spectrum absorption inversion algorithm of atmospheric CH4, the radiation transfer of solar radiation along the light path to the sensor needs to be modeled, and the radiation changes caused by different concentrations of atmospheric CH4 are also separated and simulated. The classical differential absorption spectrum technology (DOAS (Differential Optical Absorption Spectroscopy)) is based on the Beer-Lambert law, which describes the relationship between the incident radiation intensity of a vertical column ( ) and the measured radiation intensity after passing through a light path containing an absorber ( ). ) between the incident radiation intensity of a vertical column ( ) and the measured radiation intensity after passing through a light path containing an absorber (

[0054] ). ；

[0055] In the formula, represents the absorption cross section, represents the concentration of CH4 gas molecules. And the absorption cross section is a function of wavelength ( ), air temperature (T) and atmospheric pressure (p).

[0056] Based on the AHSI (hyperspectral imager) sensor can distinguish the atmospheric CH4 absorption characteristics, and the point source plume inversion region only exists short-wave infrared weak absorption gas premise assumption. The embodiment adopts a radiation transfer model, and is applied to AHSI (hyperspectral imager) imaging spectrum simulation. For the case of higher aerosol, the embodiment adopts GFIT3 to optimize the above radiation transfer model.

[0057] Specifically, in the iterative maximum posterior inversion step, the state variable further includes a polynomial fitting coefficient and a spectral offset term, and a cost function in iteration is determined by observation errors of measured and simulated spectra and prior errors of the state variable, the state variable simplifies and retains core parameters of methane concentration and surface albedo, and a convergence condition is set as a state variable change amount of adjacent iterations being less than a set threshold.

[0058] Specifically, the radiation transfer model is configured with:

[0059] ;

[0060] wherein, represents a high-resolution backscattering simulation radiation amount at a wavelength , is a solar incident radiation intensity, is an air mass factor of a vertical atmosphere =[1,72] layer, is a default optical thickness of each trace gas element from the U.S. standard atmosphere, wherein an initial value of atmospheric CH4 is set as 2000ppb, is a scaling factor applied to the optical depth, is a polynomial coefficient representing a low-frequency spectral feature, wherein represents a state variable, including gas concentration, atmospheric and surface parameters, is a model input parameter, and k is a preset number.

[0061] The embodiment simulates in a spectral range of 2076-2510nm (around 2300nm) with a spectral resolution of 0.02nm, and forms a forward simulation vector . Then, through a spectral response function, as shown in Figure 2 , a convolution calculation based on a Gaussian distribution modeling with a mean value of 0 and a full width at half maximum (FWHM) parameter is performed, to realize a high-resolution simulation vector and an imaging simulation spectrum after convolution of the spectral response function, as shown in Figure 3 . Then, sampling of the simulation spectrum at known AHSI (hyperspectral imager) wavelength positions can obtain a forward simulation spectrum F (x) representing satellite imaging.

[0062] Due to the significantly lower spectral resolution and signal-to-noise ratio of AHSI (Airborne Hyperspectral Imager) than TROPOMI (Tropospheric Monitoring Instrument) and GMI (Greenhouse Gas Monitoring Instrument), the atmospheric CH4 retrieval based on AHSI (Airborne Hyperspectral Imager) cannot directly use the DOAS (Differential Optical Absorption Spectroscopy) retrieval method. The embodiment adopts an iterative maximum posterior algorithm, adjusts the state variable to be inverted, makes the difference between the imaging spectrum simulated by the forward model and the AHSI (Airborne Hyperspectral Imager) measured spectrum, and statistically optimizes the solution on the basis of considering the AHSI (Airborne Hyperspectral Imager) and the forward simulation spectrum error.

[0063] ;

[0064] In the formula, y is the AHSI (Airborne Hyperspectral Imager) observation real spectrum; F is the forward model simulation AHSI (Airborne Hyperspectral Imager) imaging spectrum; b is the model input parameter, for example, the optical absorption coefficient of atmospheric molecules, etc., x is the state variable, and the embodiment specifically includes 11 parameters including target gas (CH4) and interfering gas (CO2, H2O and NO2) concentration, 5 polynomial fitting coefficients (atmospheric, surface and aerosol parameters, etc.) and 2 spectral offset items; is an error term. In actual calculation, the state variable to be inverted x is calculated by solving the minimum value of the one-dimensional variation cost function.

[0065] Specifically, the cost function of the iterative maximum posterior algorithm is configured as:

[0066] ;

[0067] In the formula, J(x) is the cost function, y is the AHSI (Airborne Hyperspectral Imager) observation real spectrum, F is the forward model simulation AHSI (Airborne Hyperspectral Imager) imaging spectrum, b is the model input parameter, x is the state variable, x a is the state variable prior value, S a is the error covariance matrix of the state variable prior value, S ɛ is the observation spectrum error covariance matrix. The matrix is calculated by comparing the difference between the AHSI (Airborne Hyperspectral Imager) measured spectrum and the forward simulation spectrum under the CH4 background concentration scene.

[0068] In order to reduce the calculation complexity and improve the efficiency, the following simplification is made to the problem. Since many atmospheric parameters (such as other gases or clouds) are not directly related to CH4 retrieval, they are not included in the state vector. CH4 is usually stable in the atmosphere (unless in the fossil energy point source emission area), so attention can be focused on a few key parameters. Simplification method:

[0069] ;

[0070] Since the function is nonlinear, the solving process needs to be obtained by iteration. For the state variable i second iteration , the Jacobian matrix is calculated.

[0071] ;

[0072] Meanwhile, the embodiment obtains the state variable x that minimizes the cost function by Gauss-Newton iteration.

[0073] Specifically, the maximum posterior algorithm iteration is configured with:

[0074] ;

[0075] Where x i+1 represents the state variable vector to be inverted at the i+1th iteration, x i represents the state variable vector to be inverted at the ith iteration, K represents the Jacobian matrix, that is, the partial derivative matrix of the forward simulation model with respect to the state variable x, represents the degree of influence of a small change in the state variable on the simulated spectrum, γ is an iteration adjustment parameter, used to dynamically adjust the trust region of the iteration step, c represents the bias term, x a represents the state variable prior value vector, that is, the initial reference value of the state variable.

[0076] Parameter is updated in each iteration to minimize the cost function, and is calculated according to the following ratio: R

[0077] ；

[0078] Where is the value of the cost function corresponding to the forward model and the state variable in the i+1th iteration; and is the value of the cost function of under the linear assumption. From this definition, it can be seen that R represents the influence of the nonlinear characteristics of the forward model on the cost function. When the forward model is linear, R=1. In specific calculations, R is initialized to 10. In each iteration, when R>0.75, R is reduced by half; otherwise, R remains unchanged. When the state variable satisfies the following relationship, the solution converges:

[0079] ；

[0080] ​The change of the representative state variable in iteration is less than the posterior error, wherein n represents the number of elements of the state variable vector; is the posterior estimation error covariance matrix of the final state variable estimation value , which can be expressed as:

[0081] In the formula, represents the posterior inversion error and the estimation error correlation between different state variable parameters. This is an important indicator of the inversion result error estimation.

[0082] The embodiment of the present application also provides a hyperspectral atmospheric methane concentration inversion system for a high aerosol scene, comprising:

[0083] A prior data acquisition module acquires atmospheric parameter data and multi-band ground surface albedo data of a target area, performs unified pressure level adaptation processing on the atmospheric parameter data, performs methane short-wave infrared absorption spectrum interpolation processing on the ground surface albedo data, constructs a radiation transfer model through the processed data, and outputs a forward simulation vector;

[0084] A lookup table construction module constructs a short-wave infrared hyperspectral optical absorption coefficient lookup table including methane and interfering gases and containing different temperature and pressure conditions based on CLARS-GFIT;

[0085] A forward simulation module constructs a hyperspectral imaging forward simulation model according to the forward simulation vector, corrects the forward simulation model through a GFIT3 model to quantify the influence of high aerosol on radiation transfer, and generates a forward simulation spectrum through convolution processing of the forward simulation spectrum vector combined with a spectral response function of a hyperspectral sensor.

[0086] An iterative maximum posterior inversion module determines state variables including methane concentration, interfering gas concentration and spectral parameters, adjusts the state variables through an iterative maximum posterior algorithm and minimizes a cost function of a measured spectrum and the forward simulation spectrum, simplifies the state variables and includes the influence of un-inverted atmospheric parameters into an error matrix, iteratively solves to convergence, and obtains a methane concentration inversion result.

[0087] The following shows and describes a methane concentration random enhancement spectrum simulation of the present application:

[0088] (1) Optionally, satellite data of one scene of GF-5B / AHSI (hyperspectral imager) is acquired to obtain Wavelength and FWHM (full width at half maximum), and Wavelength and FWHM within 2110-2455nm are obtained;

[0089] ​(2) Generate a 100x100 image matrix (Concentration: 100x100; Spectrum: 100x100x41; Albedo: 100x100), 98% of which is background concentration (1900ppb), 2% of which is randomly increased concentration, and the random increase in concentration value is (1-1500ppb), to construct the simulated XCH4 concentration data (100x100) of the background concentration+increased concentration;

[0090] (3) Randomly generate the Albedo matrix (0.06-0.20), set other gas molecules to the initial value (1.0), set the spectral offset to 0, based on the concentration matrix and the Albedo matrix, use the forward model to simulate the clean high-resolution spectrum pixel by pixel, and then use the FWHM and Gaussian function convolution of the satellite to generate the imaging spectrum (1x1x41);

[0091] (4) Take 0.002% to 0.02% of the maximum value of each imaging spectrum as the standard deviation (corresponding to the signal-to-noise ratio 500 to 50), and the mean value is 0, add Gaussian noise to each spectral segment, and finally return the imaging spectrum data after adding noise;

[0092] (5) Use the MF, ILMF and COBRA methods in turn to invert the simulated spectrum. Basic characteristics, principles and advantages of the present application.

[0093] As shown in Figure 4 , the scatter plot compares the true CH4 enhancement value and the CH4 enhancement value obtained by randomly simulating the enhancement value (ranging from 0 to 1500ppb). (a) to (c) represent MF, ILMF and COBRA respectively. Only the enhanced pixels are shown in the figure. The color of the points is distinguished according to the true noise ratio. The red solid line is the linear regression line of these scatter data. The red dotted line is a straight line with a slope of 1.

[0094] The present application also proposes a WRF-LES simulation point source plume concentration enhancement spectrum simulation, as shown in Figure 5, optionally satellite data of one scene GF-5B / AHSI (hyperspectral imager), obtaining Wavelength and FWHM, obtaining Wavelength and FWHM within the wavelength 2110-2455nm; generating 200x333 image matrix (concentration: 200x333; spectrum: 200x333x41; Albedo: 200x333), the plume enhanced concentration of WRF-LES simulation (1.0t·h-1, 1.0m / s) is the growth value, and the rest is the background value, the simulated XCH4 concentration data (200x333) of the background concentration+growth concentration is constructed; the Albedo matrix is randomly generated (0.06-0.30), other gas molecules are set to the initial value (1.0), the spectral offset is set to 0, based on the concentration matrix and the Albedo matrix, the clean high-resolution spectrum is simulated by using the forward model per pixel, then the imaging spectrum (1x1x41) is generated by using the FWHM of the satellite and the Gaussian function convolution; each imaging spectrum is added with Gaussian noise according to 0.2%to2%of the maximum value as the standard deviation, and the mean value is 0, Gaussian noise is added to each spectral segment, and finally the imaging spectrum data after adding noise is returned; the simulated spectrum is inversed by using the MF, ILMF and COBRA method in turn. The background concentration used for solving ΔXCH4 in the full physical inversion is the mean value of all background concentration inversions.

[0095] The embodiment of the present application also proposes WRF-LES point source plume spectrum simulation under satellite real spectrum, as shown in Figure 6 Satellite data of one scene GF-5B / AHSI (hyperspectral imager) without point source plume is selected, Wavelength and FWHM are obtained, Wavelength and FWHM within the wavelength 2110-2455nm are obtained, and satellite spectrum data are obtained; the simulated XCH4 concentration data (200x333) of the background concentration+growth concentration is constructed based on the WRF-LES simulated plume concentration plus the background concentration (1900ppb), the transmittance is calculated per pixel, the simulated spectrum is simulated, and the simulated spectrum is inversed by using the MF, ILMF and COBRA method in turn.

[0096] It should be pointed out that the present application is not limited by the above-mentioned embodiments, and is only a part of the embodiments, and a number of improvements and supplements made without departing from the spirit and scope of the present application are regarded as the protection scope of the present application.

Claims

1. A hyperspectral method for retrieving atmospheric methane concentration in high aerosol scenarios, characterized in that, include: The prior data acquisition steps involve acquiring atmospheric parameter data and multi-band surface albedo data for the target area, performing unified pressure level adaptation processing on the atmospheric parameter data, interpolating the surface albedo data using the methane shortwave infrared absorption spectrum, constructing a radiative transfer model based on the processed data, and outputting a forward simulation vector. The lookup table construction steps involve constructing a short-wave infrared hyperspectral optical absorption coefficient lookup table based on CLARS-GFIT, including methane and interfering gases, and covering different temperature and pressure conditions. The forward simulation step involves constructing a hyperspectral imaging forward simulation model based on the forward simulation vector, correcting the forward simulation model using the GFIT3 model to quantify the influence of high aerosols on radiative transfer, and convolving the forward simulation spectral vector with the spectral response function of the hyperspectral sensor to generate a forward simulation spectrum. The iterative maximum a posteriori inversion step determines the state variables including methane concentration, interfering gas concentration, and spectral parameters. The state variables are adjusted by the iterative maximum a posteriori algorithm and the cost function between the measured spectrum and the forward simulated spectrum is minimized. The state variables are simplified and the influence of unretrieved atmospheric parameters is included in the error matrix. The solution is iteratively solved until convergence to obtain the methane concentration inversion result. The forward simulation step includes modifying the radiative transfer equation of the forward simulation model, introducing parameters characterizing aerosol scattering and absorption characteristics, and correcting the short-wave infrared radiation transmission process in high aerosol scenarios. The spectral response function is constructed based on the imaging principle of the hyperspectral sensor and convolves the high-resolution spectral vector generated by the forward simulation to make the characteristics of the forward simulation spectrum consistent with those of the sensor's measured spectrum.

2. The method for inverting hyperspectral atmospheric methane concentration in high aerosol scenarios according to claim 1, characterized in that, In the prior data acquisition step, the atmospheric parameter data includes air temperature, atmospheric humidity, trace gas concentration, surface temperature, and surface pressure. The unified pressure level adaptation process maps the atmospheric parameter data to the pressure level preset by the radiative transfer model, ensuring that the atmospheric parameters of each layer match the vertical resolution of the model. The interpolation processing of multi-band surface albedo data maps the discrete band albedo to the methane shortwave infrared absorption spectrum, ensuring that the preprocessed data is directly adapted to the radiative transfer model.

3. The method for inverting hyperspectral atmospheric methane concentration in high aerosol scenarios according to claim 1, characterized in that, In the lookup table construction step, the interfering gas is a gas that affects the absorption of short-wave infrared spectrum. When calling the lookup table, the optical absorption coefficients of methane and interfering gas under the corresponding conditions are calculated by interpolation based on the real-time temperature and pressure data of the target area, and input into the spectral absorption characteristic model in the forward simulation as the basic data.

4. The method for inverting hyperspectral atmospheric methane concentration in high aerosol scenarios according to claim 1, characterized in that, In the iterative maximum a posteriori inversion step, the state variables also include polynomial fitting coefficients and spectral offset terms. The cost function in the iteration is determined simultaneously by the observation error of the measured and simulated spectra and the prior error of the state variables. After simplification, the state variables retain the core parameters of methane concentration and surface albedo. The convergence condition is set to the state variable change in adjacent iterations being less than a set threshold.

5. The method for inverting hyperspectral atmospheric methane concentration in high aerosol scenarios according to claim 3, characterized in that, The radiative transfer model is configured with: ； in, Indicates at wavelength High-resolution backscattering simulation of radiation at that location, It is the intensity of solar incident radiation. It is the vertical atmosphere =Air mass factor of layer [1,72] This is the default optical thickness for each trace gas element from the US standard atmosphere, where the initial value for atmospheric CH4 is set to 2000 ppb. It is a scaling factor applied to optical depth. These are polynomial coefficients, representing low-frequency spectral characteristics, where... These represent state variables, including gas concentration, atmospheric and surface parameters. The input parameters for the model are k, which is the preset number of iterations.

6. The method for inverting hyperspectral atmospheric methane concentration in high aerosol scenarios according to claim 4, characterized in that, The cost function of the iterative maximum a posteriori algorithm is configured as follows: ； Where J(x) is the cost function, y is the actual AHSI observation spectrum, F is the imaging spectrum of the forward model simulating AHSI, b is the model input parameter, and x is the state variable. a S is the prior value of the state variable. a S is the error covariance matrix of the prior values ​​of the state variables. ɛ This is the covariance matrix of the observed spectral error.

7. The hyperspectral atmospheric methane concentration inversion method for high aerosol scenarios according to claim 6, characterized in that, The maximum a posteriori algorithm iterative configuration includes: ； Where, x i+1 Let x represent the vector of state variables to be inverted at the (i+1)th iteration. i Let represent the vector of state variables to be inverted at the i-th iteration, K represent the Jacobian matrix, i.e., the matrix of partial derivatives of the forward simulation model with respect to the state variable x, representing the degree of influence of small changes in the state variable on the simulated spectrum, γ is the iteration adjustment parameter used to dynamically adjust the confidence region of the iteration step size, c represents the bias term, and x a This represents the prior value vector of the state variables, i.e., the initial reference values ​​of the state variables.

8. A hyperspectral atmospheric methane concentration inversion system for high aerosol scenarios, applicable to the hyperspectral atmospheric methane concentration inversion method for high aerosol scenarios as described in any one of claims 1 to 7, characterized in that, include: The prior data acquisition module acquires atmospheric parameter data and multi-band surface albedo data of the target area, performs unified pressure level adaptation processing on the atmospheric parameter data, performs methane shortwave infrared absorption spectrum interpolation processing on the surface albedo data, constructs a radiative transfer model based on the processed data and outputs a forward simulation vector. The lookup table construction module, based on CLARS-GFIT, constructs a lookup table for short-wave infrared hyperspectral optical absorption coefficients, including methane and interfering gases, and covering different temperature and pressure conditions. The forward simulation module constructs a hyperspectral imaging forward simulation model based on the forward simulation vector, corrects the forward simulation model using the GFIT3 model to quantify the influence of high aerosols on radiative transfer, and convolves the forward simulation spectral vector with the spectral response function of the hyperspectral sensor to generate a forward simulation spectrum. The iterative maximum a posteriori inversion module determines the state variables including methane concentration, interfering gas concentration, and spectral parameters. The iterative maximum a posteriori algorithm is used to adjust the state variables and minimize the cost function between the measured spectrum and the forward simulated spectrum. The state variables are simplified and the influence of unretrieved atmospheric parameters is included in the error matrix. The solution is iteratively solved until convergence to obtain the methane concentration inversion result. The forward simulation step includes modifying the radiative transfer equation of the forward simulation model, introducing parameters characterizing aerosol scattering and absorption characteristics, and correcting the short-wave infrared radiation transmission process in high aerosol scenarios. The spectral response function is constructed based on the imaging principle of the hyperspectral sensor and convolves the high-resolution spectral vector generated by the forward simulation to make the characteristics of the forward simulation spectrum consistent with those of the sensor's measured spectrum.

Images