Construction method of atmospheric CO2 satellite remote sensing data error correction model

By constructing an error correction model for remote sensing data atmospheric CO2 satellites, using full physical inversion and posterior covariance decomposition technology, combined with foundation observation and machine learning algorithms, the problem of error decomposition of satellite remote sensing data is solved, and monitoring accuracy and reliability are improved. on the global carbon cycle and climate change assessment.

CN119069020AActive Publication Date: 2024-12-03廊坊空间信息技术研发服务中心 +1

Patent Information

Application Number
CN202411260769.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-10
Publication Date
2024-12-03
Estimated Expiration
2044-09-10

AI Technical Summary

Technical Problem

Traditional atmospheric CO2 concentration monitoring methods have problems of insufficient accuracy and uneven data in global observations. The errors of satellite remote sensing data are difficult to quantitatively decompose, affecting the accuracy of global carbon cycle and climate change assessment.

Method used

By collecting satellite remote sensing data for preprocessing, inputting the full physical inversion model for inversion, combining posterior covariance decomposition and spatiotemporal matching of the total foundation carbon observation network, error correction factors are obtained, and an error correction model for atmospheric CO2 satellite remote sensing data is constructed. , optimize the model using random forests, decision threshold classification and machine learning algorithms.

Benefits of technology

The accuracy and accuracy of the error correction model of the atmospheric CO2 satellite remote sensing data is improved, and intelligent correction and real-time decomposition of errors is realized, adapting to different needs and being universal.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119069020A_ABST
    Figure CN119069020A_ABST
Patent Text Reader

Abstract

The invention discloses a construction method of an atmospheric CO2 satellite remote sensing data error correction model, which comprises the following steps: acquiring remote sensing data of an atmospheric CO2 satellite, and preprocessing the remote sensing data; inputting the remote sensing data into a full-physical inversion model to obtain inversion data, and performing posterior covariance decomposition on the inversion data to obtain a posterior error; performing space-time matching on the inversion data according to a foundation total carbon quantity observation network to obtain comparison data, and performing difference correction analysis on the comparison data to obtain an error correction factor; and constructing an atmospheric CO2 satellite remote sensing data error correction model according to the posterior error and the error correction factor, optimizing the atmospheric CO2 satellite remote sensing data error correction model, and outputting a target model. The method not only can improve the error correction precision based on the atmospheric CO2 satellite remote sensing data, but also has good interpretability, and can be directly applied to an atmospheric CO2 satellite remote sensing data error correction model system.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of error correction, and in particular to a method for constructing an error correction model for atmospheric CO2 satellite remote sensing data. Background Art

[0002] As one of the most important greenhouse gases, changes in atmospheric CO2 concentration will have a significant impact on global climate change, human living environment and even the structure of the entire social-economic system. Therefore, monitoring of atmospheric CO2 concentration has always been a hot topic in current scientific research. Traditional atmospheric CO2 concentration monitoring is mainly based on ground-based observations, which have high observation accuracy, but there are few observation sites around the world and they are unevenly distributed, making it difficult to accurately characterize the distribution of CO2 concentration. Satellite monitoring can provide global CO2 concentration observation data over a large range. Atmospheric CO2 remote sensing monitoring is to analyze the spectral data of CO2 specific absorption bands captured by satellites to estimate the concentration of CO2 in the atmosphere. This process is affected by many factors, including atmospheric conditions, instrument noise, surface reflection, etc.

[0003] The posterior covariance is a measure of the uncertainty in parameter estimation after considering the observed data in the inversion process. It can quantitatively analyze the contribution of different parameters to the uncertainty of the inversion results, but it uses mathematical statistical methods and its value is 2-3 times lower than the true error. Ground-measured data provide an opportunity for independent verification and can be used to evaluate the accuracy of satellite observation data, but their quantity is limited and can only estimate the total error, making it difficult to quantitatively decompose the error. Therefore, the error decomposition method for atmospheric CO2 satellite remote sensing observations based on posterior covariance and ground-measured atmospheric CO2 is aimed at quantitatively analyzing the error sources and structure, and further optimizing the accuracy and reliability of atmospheric CO2 remote sensing monitoring, which is crucial for understanding the global carbon cycle and assessing the impact of climate change. Summary of the invention

[0004] The purpose of the present invention is to provide a method for constructing an atmospheric CO2 satellite remote sensing data error correction model.

[0005] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0006] The present invention comprises the following steps:

[0007] Collecting atmospheric CO2 satellite remote sensing data and preprocessing the remote sensing data;

[0008] Inputting the remote sensing data into a full physical inversion model to obtain inversion data, and performing a posteriori covariance decomposition on the inversion data to obtain a posteriori error;

[0009] Performing spatiotemporal matching on the inversion data according to the ground-based carbon total observation network to obtain comparison data, and performing difference correction analysis on the comparison data to obtain an error correction factor;

[0010] An atmospheric CO2 satellite remote sensing data error correction model is constructed according to the a posteriori error and the error correction factor, the atmospheric CO2 satellite remote sensing data error correction model is optimized, and a target model is output.

[0011] Furthermore, the method of the full physical inversion model includes:

[0012] Determine the absorption of the characteristic spectrum of carbon dioxide, the expression is:

[0013]

[0014] The optical attenuation coefficient of the characteristic spectrum λ is L(λ), the optical path is F, and the incident radiation of the characteristic spectrum λ is The outgoing radiation of the characteristic spectrum λ through the optical path F is The concentration of the ath wavelength absorbing gas is k a , the high-frequency narrow-band absorption cross section that changes rapidly with wavelength is β a1 , the low-frequency broadband absorption cross section that changes slowly with wavelength is β a2 , the Rayleigh scattering effect of the characteristic spectrum λ is δ R (λ), the Mie scattering effect of the characteristic spectrum λ is δ M (λ);

[0015] Calculate the spectral intensity of the undifferentiated absorption using the expression:

[0016]

[0017] The undifferentiated absorption spectrum intensity of the characteristic spectrum λ is Calculate the differential optical depth:

[0018]

[0019] Where the differential optical depth is Z, calculate the integral of the carbon dioxide gas concentration on the optical path:

[0020]

[0021] The integral of the carbon dioxide gas concentration on the optical path is The weight function of the gas column is used as the influence coefficient of the differential spectrum;

[0022] Improve the formula, the expression is:

[0023]

[0024] The actual column concentration of carbon dioxide at time s is y s , the model column concentration of carbon dioxide at time s is The inversion column concentration of carbon dioxide at time s is The change parameter of the ath wavelength absorbing gas is q a , the actual parameter at the sth moment is q s , the b-th model spectral intensity of the atmospheric state and surface reflectivity is The number of wavelength absorbing gases is N, and the bth term that compensates for the high-frequency spectral response is B b , the weighted average spectral intensity of the bth model is

[0025] The CO2 column concentration was obtained by adjusting the model strength of the above equation;

[0026] Calculate the error of the inter-layer weight vector in the inversion:

[0027] γ(Q U+1 )=γ(Q U )+D(Q U+1 -Q U )

[0028] The number of iterations is U, and the vector of the weight Q of the Uth iteration is Q U , the Jacobian matrix is ​​D, and the vector of the weight Q for the U+1th iteration is Q U+1 , vector Q U The error is γ(Q U ), vector Q U+1 The error is γ(Q U+1 );

[0029] Derivative the weight vector error, the expression is:

[0030] Q U+1 =Q U -(D T D) -1 D T γ(Q U )

[0031] The transpose of the Jacobian matrix is ​​D T , calculate the inversion error:

[0032]

[0033] The inversion error is ε, and the predicted carbon dioxide column concentration is The actual carbon dioxide column concentration is x, and the control factor is λ;

[0034] The influence coefficient of the differential spectrum was adjusted according to the inversion error until the inversion error was less than 0.20.

[0035] Furthermore, the method of performing a posteriori covariance decomposition on the inversion data to obtain a posteriori error includes:

[0036] Calculate the total error:

[0037]

[0038] The pressure weight function is g, and the posterior covariance matrix is Transpose to T and calculate the posterior covariance:

[0039]

[0040] The Jacobian matrix is ​​D and the measurement error covariance matrix is ​​S τ , the prior covariance matrix is ​​S ι , the state variable is u, and the function of the state variable is H(u);

[0041] Calculation error:

[0042]

[0043] The kth contribution function is G k , the instrument error is The error of the carbon dioxide profile is The error of other factors is The average kernel function is A, the identity matrix is ​​E, and the average kernel function of carbon dioxide is A CO2 , the prior covariance matrix of carbon dioxide is S ι,CO2 , the subset of the prior covariance matrix is ​​S ιe , the subset of the average kernel matrix is ​​A ue , the jth state variable is u j , the i-th posterior state variable

[0044] Compute the mean kernel function for CO2, a subset of the mean kernel matrix, the prior covariance matrix for CO2, and a subset of the prior covariance matrix:

[0045] A CO2 =A(1:nz,1:nz)

[0046] A ue =A(1:nz,nz+1:n)

[0047] S ιe =S ι (1:nz,1:nz)

[0048] S ι,CO2 =S ι(nz+1:nz+1:nz)

[0049] Where the profile layer of CO2 is nz, the number of inversion state parameters is n, and the prior covariance matrix and a subset of the prior covariance matrix of carbon dioxide are calculated:

[0050] Calculation error percentage:

[0051]

[0052] The posterior error of the i-th parameter is

[0053] Furthermore, the method of performing difference correction analysis on the comparison data to obtain an error correction factor comprises:

[0054] Sort the comparison data by time, and construct a recursive relationship about the time series based on the comparison data. The expression is:

[0055]

[0056] The output data at the s+1th moment is The error between the remote sensing data and the comparison data at the s+1th moment is d(s+1), the comparison data at the s+1th moment is W(s+1), transposed to T, and the impact factor is ζ;

[0057] Using the least squares algorithm for recursion, the expression is:

[0058]

[0059] The gain matrix at the sth moment is The covariance matrix at the sth moment is The forgetting factor is ρ, and the covariance matrix at the s+1th moment is

[0060] Update comparison data, the expression is:

[0061]

[0062] The updated comparison data at time s+1 is The penalty weight vector is Calculate the error ratio parameter:

[0063]

[0064] The error ratio parameter at the s+1th moment is h(s+1), and the determinant operation is |·|;

[0065] The comparison data with error ratio parameter greater than 0.121 are taken as problem data, and the problem data are input into the atmospheric CO2 satellite remote sensing data prediction model to obtain the predicted carbon dioxide concentration;

[0066] Calculate the error correction factor based on the predicted carbon dioxide concentration:

[0067]

[0068] The error correction factor is x, and the problem data The number of problem data is m. The error ratio parameter is The optimal control coefficient is ξ.

[0069] Furthermore, a method for constructing an atmospheric CO2 satellite remote sensing data error correction model based on the a posteriori error and the error correction factor includes:

[0070] The a posteriori error is corrected according to the error correction factor to obtain the corrected error, and the atmospheric CO2 satellite remote sensing data error correction model is constructed according to the corrected error;

[0071] The error correction models for atmospheric CO2 satellite remote sensing data include random forest algorithm, decision threshold classification algorithm, and machine learning algorithm;

[0072] The random forest algorithm randomly divides the corrected error into training data and test data in a ratio of 7:2;

[0073] The decision threshold classification algorithm selects the best decision threshold by analyzing the decision function output and actual label of the atmospheric CO2 satellite remote sensing data prediction model, and classifies the training data according to the decision threshold to obtain classified data;

[0074] The machine learning algorithm obtains the error characteristics by learning the classified data and corrects the input data according to the error characteristics;

[0075] The test data was input into the atmospheric CO2 satellite remote sensing data error correction model to obtain the corrected carbon dioxide concentration. The learning rate of the atmospheric CO2 satellite remote sensing data error correction model was adjusted according to the mean square error between the corrected carbon dioxide concentration and the actual carbon dioxide concentration until the mean square error was lower than 0.132, then the adjustment was stopped.

[0076] Furthermore, the method for optimizing the atmospheric CO2 satellite remote sensing data error correction model comprises:

[0077] Calculate the fitness value of the search sub-population and select the optimal population and the optimal individual;

[0078] Calculate the transition coefficient:

[0079]

[0080] The transition washing is The random number between 0 and 1 is r1, the current number of iterations is t, and the maximum number of iterations is t max ;

[0081] When the transition coefficient is greater than a certain parameter, update the position and number of searchers:

[0082]

[0083]

[0084] The dimensional constant is υ, the random number generated by levy is R(υ), the normal distribution random number is β, the random numbers from 0 to 1 are r2 and r3, and the c-th searcher of the t-th iteration is The random searcher for the t+1th iteration is The empty position of the c-th searcher in the t+1th iteration is The empty initial update position of the c-th searcher in the t+1th iteration is The merged empty update position of the c-th searcher in the t+1th iteration is The number of searchers is N, the fitness value ranking function is ε(·), and the air update position of the c-th searcher in the t+1th iteration is The individual function with the best fitness value is new;

[0085] When the transition coefficient is less than a certain parameter, update the position and number of searchers:

[0086]

[0087] The random numbers from 1 to N-1 are r4, r5, r6, the random numbers from 0 to 1 are r7, η, and the c-th searcher of the t-th iteration is The random search sub-iterations of the tth iteration are The water position of the c-th searcher in the t+1th iteration is The initial updated position of the c-th searcher in the t+1th iteration is The merged empty update position of the c-th searcher in the t+1th iteration is The updated position of the c-th searcher in the water at the t+1th iteration is The cooperation factor is

[0088] Calculate and compare the fitness values ​​of the populations, update the optimal value, and stop the iteration if the maximum number of iterations is reached, otherwise reselect the optimal population and the optimal individual.

[0089] The beneficial effects of the present invention are:

[0090] The present invention is a method for constructing an atmospheric CO2 satellite remote sensing data error correction model. Compared with the prior art, the present invention has the following technical effects:

[0091] The present invention can improve the accuracy of the atmospheric CO2 satellite remote sensing data error correction model through preprocessing, full physical inversion, posterior covariance decomposition, time-space matching, difference correction analysis, model construction and model optimization steps, thereby improving the precision of the atmospheric CO2 satellite remote sensing data error correction model, optimizing the atmospheric CO2 satellite remote sensing data error correction model, greatly saving resources, and improving work efficiency, and can realize intelligent correction of atmospheric CO2 satellite remote sensing data errors, and perform error decomposition and model optimization on the atmospheric CO2 satellite remote sensing data error correction model in real time, which is of great significance to multiple atmospheric CO2 satellite remote sensing data error correction models, can adapt to atmospheric CO2 satellite remote sensing data error correction models of different standards and different atmospheric CO2 satellite remote sensing data error correction model requirements, and has certain universality. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] Figure 1 The present invention is a flowchart of the steps of a method for constructing an atmospheric CO2 satellite remote sensing data error correction model. DETAILED DESCRIPTION

[0093] The present invention is further described below by means of specific embodiments. The illustrative embodiments and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.

[0094] The present invention provides a method for constructing an atmospheric CO2 satellite remote sensing data error correction model, comprising the following steps:

[0095] like Figure 1 As shown, in this embodiment, the following steps are included:

[0096] Collecting atmospheric CO2 satellite remote sensing data and preprocessing the remote sensing data;

[0097] In the actual evaluation, the suburban area of ​​XXX was taken as the research object, and remote sensing data for the past three years were obtained. The remote sensing data included the concentration distribution information of carbon dioxide in the atmosphere, time variation data, carbon dioxide concentration data, flow and exchange data of carbon dioxide in the atmosphere, satellite data, and hyperspectral observation satellite observation data;

[0098] Inputting the remote sensing data into a full physical inversion model to obtain inversion data, and performing a posteriori covariance decomposition on the inversion data to obtain a posteriori error;

[0099] In the actual evaluation, the inversion data at 6.30am and 3.50pm on July 15, 22 were 319ppm and 578ppm respectively;

[0100] Performing spatiotemporal matching on the inversion data according to the ground-based carbon total observation network to obtain comparison data, and performing difference correction analysis on the comparison data to obtain an error correction factor;

[0101] In the actual evaluation, the error correction factor is 0.3173;

[0102] An atmospheric CO2 satellite remote sensing data error correction model is constructed according to the a posteriori error and the error correction factor, the atmospheric CO2 satellite remote sensing data error correction model is optimized, and a target model is output.

[0103] In this embodiment, the method of the full physical inversion model includes:

[0104] Determine the absorption of the characteristic spectrum of carbon dioxide, the expression is:

[0105]

[0106] The optical attenuation coefficient of the characteristic spectrum λ is L(λ), the optical path is F, and the incident radiation of the characteristic spectrum λ is The outgoing radiation of the characteristic spectrum λ through the optical path F is The concentration of the ath wavelength absorbing gas is k a , the high-frequency narrow-band absorption cross section that changes rapidly with wavelength is β a1 , the low-frequency broadband absorption cross section that changes slowly with wavelength is β a2 , the Rayleigh scattering effect of the characteristic spectrum λ is δ R (λ), the Mie scattering effect of the characteristic spectrum λ is δ M (λ);

[0107] Calculate the spectral intensity of the undifferentiated absorption using the expression:

[0108]

[0109] The undifferentiated absorption spectrum intensity of the characteristic spectrum λ is Calculate the differential optical depth:

[0110]

[0111] Where the differential optical depth is Z, calculate the integral of the carbon dioxide gas concentration on the optical path:

[0112]

[0113] The integral of the carbon dioxide gas concentration on the optical path is The weight function of the gas column is used as the influence coefficient of the differential spectrum;

[0114] Improve the formula, the expression is:

[0115]

[0116] The actual column concentration of carbon dioxide at time s is y s , the model column concentration of carbon dioxide at time s is The inversion column concentration of carbon dioxide at time s is The change parameter of the ath wavelength absorbing gas is q a , the actual parameter at the sth moment is q s , the b-th model spectral intensity of the atmospheric state and surface reflectivity is The number of wavelength absorbing gases is N, and the bth term that compensates for the high-frequency spectral response is B b , the weighted average spectral intensity of the bth model is

[0117] The CO2 column concentration was obtained by adjusting the model strength of the above equation;

[0118] Calculate the error of the inter-layer weight vector in the inversion:

[0119] γ(Q U+1 )=γ(Q U )+D(Q U+1 -Q U )

[0120] The number of iterations is U, and the vector of the weight Q of the Uth iteration is Q U , the Jacobian matrix is ​​D, and the vector of the weight Q for the U+1th iteration is Q U+1 , vector Q U The error is γ(Q U ), vector Q U+1 The error is γ(Q U+1 );

[0121] Derivative the weight vector error, the expression is:

[0122] Q U+1 =Q U -(D T D) -1 D T γ(Q U )

[0123] The transpose of the Jacobian matrix is ​​D T , calculate the inversion error:

[0124]

[0125] The inversion error is ε, and the predicted carbon dioxide column concentration is The actual carbon dioxide column concentration is x, and the control factor is λ;

[0126] The influence coefficient of the differential spectrum was adjusted according to the inversion error until the inversion error was less than 0.20.

[0127] In this embodiment, the method of performing a posteriori covariance decomposition on the inversion data to obtain a posteriori error includes:

[0128] Calculate the total error:

[0129]

[0130] The pressure weight function is g, and the posterior covariance matrix is Transpose to T and calculate the posterior covariance:

[0131]

[0132] The Jacobian matrix is ​​D and the measurement error covariance matrix is ​​S τ , the prior covariance matrix is ​​S ι , the state variable is u, and the function of the state variable is H(u);

[0133] Calculation error:

[0134]

[0135]

[0136] The kth contribution function is G k , the instrument error is The error of the carbon dioxide profile is The error of other factors is The average kernel function is A, the identity matrix is ​​E, and the average kernel function of carbon dioxide is A CO2 , the prior covariance matrix of carbon dioxide is S ι,CO2 , the subset of the prior covariance matrix is ​​S ιe , the subset of the average kernel matrix is ​​A ue , the jth state variable is u j , the i-th posterior state variable

[0137] Compute the mean kernel function for CO2, a subset of the mean kernel matrix, the prior covariance matrix for CO2, and a subset of the prior covariance matrix:

[0138] A CO2 =A(1:nz,1:nz)

[0139] A ue =A(1:nz,nz+1:n)

[0140] S ιe =S ι (1:nz,1:nz)

[0141] S ι,CO2 =S ι (nz+1:nz+1:nz)

[0142] Where the profile layer of CO2 is nz, the number of inversion state parameters is n, and the prior covariance matrix and a subset of the prior covariance matrix of carbon dioxide are calculated:

[0143] Calculation error percentage:

[0144]

[0145] The posterior error of the i-th parameter is

[0146] In this embodiment, the method of performing difference correction analysis on the comparison data to obtain an error correction factor includes:

[0147] Sort the comparison data by time, and construct a recursive relationship about the time series based on the comparison data. The expression is:

[0148]

[0149] The output data at the s+1th moment is The error between the remote sensing data and the comparison data at the s+1th moment is d(s+1), the comparison data at the s+1th moment is W(s+1), transposed to T, and the impact factor is ζ;

[0150] Using the least squares algorithm for recursion, the expression is:

[0151]

[0152] The gain matrix at the sth moment is The covariance matrix at the sth moment is The forgetting factor is ρ, and the covariance matrix at the s+1th moment is

[0153] Update comparison data, the expression is:

[0154]

[0155] The updated comparison data at time s+1 is The penalty weight vector is Calculate the error ratio parameter:

[0156]

[0157] The error ratio parameter at the s+1th moment is h(s+1), and the determinant operation is |·|;

[0158] The comparison data with error ratio parameter greater than 0.121 are taken as problem data, and the problem data are input into the atmospheric CO2 satellite remote sensing data prediction model to obtain the predicted carbon dioxide concentration;

[0159] Calculate the error correction factor based on the predicted carbon dioxide concentration:

[0160]

[0161] The error correction factor is x, and the problem data The number of problem data is m. The error ratio parameter is h g , the optimal control coefficient is ξ.

[0162] In this embodiment, the method for constructing an atmospheric CO2 satellite remote sensing data error correction model based on the a posteriori error and the error correction factor includes:

[0163] The a posteriori error is corrected according to the error correction factor to obtain the corrected error, and the atmospheric CO2 satellite remote sensing data error correction model is constructed according to the corrected error;

[0164] The error correction models for atmospheric CO2 satellite remote sensing data include random forest algorithm, decision threshold classification algorithm, and machine learning algorithm;

[0165] The random forest algorithm randomly divides the corrected error into training data and test data in a ratio of 7:2;

[0166] The decision threshold classification algorithm selects the best decision threshold by analyzing the decision function output and actual label of the atmospheric CO2 satellite remote sensing data prediction model, and classifies the training data according to the decision threshold to obtain classified data;

[0167] The machine learning algorithm obtains the error characteristics by learning the classified data and corrects the input data according to the error characteristics;

[0168] The test data was input into the atmospheric CO2 satellite remote sensing data error correction model to obtain the corrected carbon dioxide concentration. The learning rate of the atmospheric CO2 satellite remote sensing data error correction model was adjusted according to the mean square error between the corrected carbon dioxide concentration and the actual carbon dioxide concentration until the mean square error was lower than 0.132, then the adjustment was stopped.

[0169] In this embodiment, the method for optimizing the atmospheric CO2 satellite remote sensing data error correction model includes:

[0170] Calculate the fitness value of the search sub-population and select the optimal population and the optimal individual;

[0171] Calculate the transition coefficient:

[0172]

[0173] The transition washing is The random number between 0 and 1 is r1, the current number of iterations is t, and the maximum number of iterations is t max ;

[0174] When the transition coefficient is greater than a certain parameter, update the position and number of searchers:

[0175]

[0176] The dimensional constant is υ, the random number generated by levy is R(υ), the normal distribution random number is β, the random numbers from 0 to 1 are r2 and r3, and the c-th searcher of the t-th iteration is The random searcher for the t+1th iteration is The empty position of the c-th searcher in the t+1th iteration is The empty initial update position of the c-th searcher in the t+1th iteration is The merged empty update position of the c-th searcher in the t+1th iteration is The number of searchers is N, the fitness value ranking function is ε(·), and the air update position of the c-th searcher in the t+1th iteration is The individual function with the best fitness value is new;

[0177] When the transition coefficient is less than a certain parameter, update the position and number of searchers:

[0178]

[0179] The random numbers from 1 to N-1 are r4, r5, r6, the random numbers from 0 to 1 are r7, η, and the c-th searcher of the t-th iteration is The random search sub-iterations of the tth iteration are The water position of the c-th searcher in the t+1th iteration is The initial updated position of the c-th searcher in the t+1th iteration is The merged empty update position of the c-th searcher in the t+1th iteration is The updated position of the c-th searcher in the water at the t+1th iteration is The cooperation factor is

[0180] Calculate and compare the fitness values ​​of the populations, update the optimal value, and stop the iteration if the maximum number of iterations is reached, otherwise reselect the optimal population and the optimal individual.

[0181] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for constructing an error correction model for atmospheric CO2 satellite remote sensing data, characterized in that: The following steps are involved: Collecting atmospheric CO2 satellite remote sensing data and preprocessing the remote sensing data; Inputting the remote sensing data into a full physical inversion model to obtain inversion data, and performing a posteriori covariance decomposition on the inversion data to obtain a posteriori error; Performing spatiotemporal matching on the inversion data according to the ground-based carbon total observation network to obtain comparison data, and performing difference correction analysis on the comparison data to obtain an error correction factor; An atmospheric CO2 satellite remote sensing data error correction model is constructed according to the a posteriori error and the error correction factor, the atmospheric CO2 satellite remote sensing data error correction model is optimized, and a target model is output.

2. The method for constructing an atmospheric CO2 satellite remote sensing data error correction model according to claim 1, characterized in that: The method of the full physical inversion model comprises: Determine the absorption of the characteristic spectrum of carbon dioxide, the expression is: The optical attenuation coefficient of the characteristic spectrum λ is L(λ), the optical path is F, and the incident radiation of the characteristic spectrum λ is v o (λ), the outgoing radiation of the characteristic spectrum λ through the optical path F is v(λ), and the concentration of the a-th wavelength absorbing gas is k a , the high-frequency narrow-band absorption cross section that changes rapidly with wavelength is β a1 , the low-frequency broadband absorption cross section that changes slowly with wavelength is β a2 , the Rayleigh scattering effect of the characteristic spectrum λ is δ R (λ), the Mie scattering effect of the characteristic spectrum λ is δ M (λ); Calculate the spectral intensity of the undifferentiated absorption using the expression: The undifferentiated absorption spectrum intensity of the characteristic spectrum λ is Calculate the differential optical depth: Where the differential optical depth is Z, calculate the integral of the carbon dioxide gas concentration on the optical path: The integral of the carbon dioxide gas concentration on the optical path is The weight function of the gas column is used as the influence coefficient of the differential spectrum; Improve the formula, the expression is: The actual column concentration of carbon dioxide at time s is y s , the model column concentration of carbon dioxide at time s is The inversion column concentration of carbon dioxide at time s is The change parameter of the ath wavelength absorbing gas is q a , the actual parameter at the sth moment is q s , the b-th model spectral intensity of the atmospheric state and surface reflectivity is The number of wavelength absorbing gases is N, and the bth term that compensates for the high-frequency spectral response is B b , the weighted average spectral intensity of the bth model is The CO2 column concentration was obtained by adjusting the model strength of the above equation; Calculate the error of the inter-layer weight vector in the inversion: γ(Q U+1 )=γ(Q U )+D(Q U+1 -Q U ) The number of iterations is U, and the vector of the weight Q of the Uth iteration is Q U , the Jacobian matrix is ​​D, and the vector of the weight Q for the U+1th iteration is Q U+1 , vector Q U The error is γ(Q U ), vector Q U+1 The error is γ(Q U+1 ); Derivative the weight vector error, the expression is: Q U+1 =Q U -(D T D) -1 D T γ(Q U ) The transpose of the Jacobian matrix is ​​D T , calculate the inversion error: The inversion error is ε, and the predicted carbon dioxide column concentration is The actual carbon dioxide column concentration is x, and the control factor is λ; The influence coefficient of the differential spectrum was adjusted according to the inversion error until the inversion error was less than 0.

20.

3. The method for constructing an atmospheric CO2 satellite remote sensing data error correction model according to claim 1, characterized in that: The method of performing a posteriori covariance decomposition on the inversion data to obtain a posteriori error comprises: Calculate the total error: The pressure weight function is g, and the posterior covariance matrix is Transpose to T and calculate the posterior covariance: The Jacobian matrix is ​​D and the measurement error covariance matrix is ​​S τ , the prior covariance matrix is ​​S ι , the state variable is u, and the function of the state variable is H(u); Calculation error: The kth contribution function is G k , the instrument error is The error of the carbon dioxide profile is The error of other factors is The average kernel function is A, the identity matrix is ​​E, and the average kernel function of carbon dioxide is A CO2 , the prior covariance matrix of carbon dioxide is S ι,CO2 , the subset of the prior covariance matrix is ​​S ιe , the subset of the average kernel matrix is ​​A ue , the jth state variable is u j , the i-th posterior state variable Compute the mean kernel function for CO2, a subset of the mean kernel matrix, the prior covariance matrix for CO2, and a subset of the prior covariance matrix: TO CO2 =A(1:nz,1:nz) TO ue =A(1:nz,nz+1:n) S ιe =S ι (1:nz,1:nz) S ι,CO2 =S ι (nz+1:nz+1:nz) Where the profile layer of CO2 is nz, the number of inversion state parameters is n, and the prior covariance matrix and a subset of the prior covariance matrix of carbon dioxide are calculated: Calculation error percentage: The posterior error of the i-th parameter is 4. The method for constructing an atmospheric CO2 satellite remote sensing data error correction model according to claim 1, characterized in that: The method of performing difference correction analysis on the comparison data to obtain an error correction factor comprises: Sort the comparison data by time, and construct a recursive relationship about the time series based on the comparison data. The expression is: θ(s+1)=e -ζ ·W T (s+1)+d 2 (s+1) The output data at the s+1th moment is θ(s+1), the error between the remote sensing data and the comparison data at the s+1th moment is d(s+1), the comparison data at the s+1th moment is W(s+1), transposed to T, and the impact factor is ζ; Using the least squares algorithm for recursion, the expression is: The gain matrix at the sth moment is The covariance matrix at the sth moment is The forgetting factor is ρ, and the covariance matrix at the s+1th moment is Update comparison data, the expression is: The updated comparison data at time s+1 is The penalty weight vector is Calculate the error ratio parameter: The error ratio parameter at the s+1th moment is h(s+1), and the determinant operation is |·|; The comparison data with error ratio parameter greater than 0.121 are taken as problem data, and the problem data are input into the atmospheric CO2 satellite remote sensing data prediction model to obtain the predicted carbon dioxide concentration; Calculate the error correction factor based on the predicted carbon dioxide concentration: The error correction factor is The number of problem data θ is The error ratio parameter of the problem data θ is h θ , the optimal control coefficient is ξ.

5. The method for constructing an atmospheric CO2 satellite remote sensing data error correction model according to claim 1, characterized in that: The method for constructing an atmospheric CO2 satellite remote sensing data error correction model according to the a posteriori error and the error correction factor comprises: The a posteriori error is corrected according to the error correction factor to obtain the corrected error, and the atmospheric CO2 satellite remote sensing data error correction model is constructed according to the corrected error; The error correction models for atmospheric CO2 satellite remote sensing data include random forest algorithm, decision threshold classification algorithm, and machine learning algorithm; The random forest algorithm randomly divides the corrected error into training data and test data in a ratio of 7:2; The decision threshold classification algorithm selects the best decision threshold by analyzing the decision function output and actual label of the atmospheric CO2 satellite remote sensing data prediction model, and classifies the training data according to the decision threshold to obtain classified data; The machine learning algorithm obtains the error characteristics by learning the classified data and corrects the input data according to the error characteristics; The test data was input into the atmospheric CO2 satellite remote sensing data error correction model to obtain the corrected carbon dioxide concentration. The learning rate of the atmospheric CO2 satellite remote sensing data error correction model was adjusted according to the mean square error between the corrected carbon dioxide concentration and the actual carbon dioxide concentration until the mean square error was lower than 0.132, then the adjustment was stopped.

6. The method for constructing an atmospheric CO2 satellite remote sensing data error correction model according to claim 1, characterized in that: The method for optimizing the atmospheric CO2 satellite remote sensing data error correction model comprises: Calculate the fitness value of the search sub-population and select the optimal population and the optimal individual; Calculate the transition coefficient: The transition washing is The random number between 0 and 1 is r1, the current number of iterations is t, and the maximum number of iterations is t max ; When the transition coefficient is greater than a certain parameter, update the position and number of searchers: The dimensional constant is υ, the random number generated by levy is R(υ), the normal distribution random number is β, the random numbers from 0 to 1 are r2 and r3, and the c-th searcher of the t-th iteration is The random searcher for the t+1th iteration is The empty position of the c-th searcher in the t+1th iteration is The empty initial update position of the c-th searcher in the t+1th iteration is The merged empty update position of the c-th searcher in the t+1th iteration is The number of searchers is N, the fitness value ranking function is ε(·), and the air update position of the c-th searcher in the t+1th iteration is The individual function with the best fitness value is new; When the transition coefficient is less than a certain parameter, update the position and number of searchers: The random numbers from 1 to N-1 are r4, r5, r6, the random numbers from 0 to 1 are r7, η, and the c-th searcher of the t-th iteration is The random search sub-iterations of the tth iteration are The water position of the c-th searcher in the t+1th iteration is The initial updated position of the c-th searcher in the t+1th iteration is The merged empty update position of the c-th searcher in the t+1th iteration is The updated position of the c-th searcher in the water at the t+1th iteration is The cooperation factor is Calculate and compare the fitness values ​​of the populations, update the optimal value, and stop the iteration if the maximum number of iterations is reached, otherwise reselect the optimal population and the optimal individual.

Citation Information

Patent Citations

  • Deviation correction method and system of XCO2 inversion data of TanSat satellite

    CN112161943A

  • Satellite carbon dioxide high-precision remote sensing inversion method

    CN116008226A

  • High-spatial-resolution satellite load methane column concentration remote sensing inversion method and system

    CN116804621A

  • DSM-integrated satellite remote sensing surface temperature intelligent inversion method and system

    CN117828969A

  • Flat-field-domain atmospheric correction method for hyperspectral satellite, electronic equipment and medium

    CN118464188A

Cited By

  • Automatic remote sensing inversion method and system, electronic equipment and program product

    CN120430039A

  • Fine particle modal ratio correction method and device based on ground and satellite fusion

    CN120721581A