A mobile remote sensing method and system for rapidly monitoring atmospheric carbon dioxide column concentration

Abstract

The application discloses a kind of movable ground remote sensing atmospheric carbon dioxide column concentration rapid monitoring method and system, it is related to environmental engineering technical field.The method is combined with portable fourier infrared spectrometer, solar tracker, stepping motor and portable power supply, the accurate measurement of carbon dioxide column concentration in atmosphere is realized.Through multilayer perception machine learning model, the sunlight spectrum data collected by fourier infrared spectrometer is rapidly inversed, and the carbon dioxide column concentration in atmosphere is accurately calculated.The multilayer perception machine learning model is trained based on the training data obtained by atmospheric radiation transmission simulation, the mapping from spectrum data to carbon dioxide column concentration data is realized, so as to ensure the accuracy of inversion.The mobility of platform realizes the continuous monitoring of carbon dioxide concentration in specific geographical position, provides data support for better understanding of regional scale carbon dioxide concentration change, and provides effective measurement means for realizing regional carbon source and carbon sink monitoring.

Description

Technical Field

[0001] This invention relates to the field of environmental engineering technology, and in particular to a mobile remote sensing method and system for rapid monitoring of atmospheric carbon dioxide column concentration. Background Technology

[0002] As the greenhouse gas with the most significant impact on human ecology, atmospheric carbon dioxide (CO2) requires accurate regional measurement for monitoring carbon sources and sinks, and plays an indispensable role in achieving the "dual carbon" goal. One characteristic of atmospheric CO2 is its absorption of solar spectral radiation passing through the atmosphere; the intensity of this absorption is closely related to the atmospheric CO2 concentration. Therefore, measuring the solar spectrum reaching the Earth's surface after atmospheric absorption has become an effective means of obtaining information on atmospheric CO2 concentration.

[0003] Currently, carbon monitoring based on hyperspectral remote sensing mainly relies on space-based satellite observations and fixed-site ground-based observations. However, these methods typically lack the ability to continuously and accurately monitor carbon emissions in a specific area. In contrast, mobile ground-based CO2 monitoring can provide more precise carbon dioxide concentration data for a specific area, offering a reliable basis for revealing emission levels in that region. This is of significant importance for industrial sectors and government agencies in identifying CO2 emission sources and formulating and improving emission reduction policies.

[0004] Mobile ground-based remote sensing monitoring platforms, as an important supplement to space-based satellite remote sensing monitoring, will greatly improve the flexibility and efficiency of regional carbon monitoring. Portable Fourier transform infrared (FTIR) spectrometers, due to their small size and ease of deployment, can be flexibly applied to monitor point source greenhouse gas emissions in a region as a mobile spectral remote sensing observation tool. For inversion algorithms involving high-resolution spectral data and multiple atmospheric parameters, machine learning methods can effectively uncover potential nonlinear relationships within the data. Compared to traditional methods, machine learning requires significantly less computation and can achieve results and application performance comparable to traditional inversion methods. Therefore, using machine learning models for CO2 monitoring can achieve more accurate and rapid atmospheric CO2 concentration inversion.

[0005] Therefore, those skilled in the art are dedicated to developing a mobile remote sensing system for rapid monitoring of atmospheric CO2. By continuously monitoring the solar spectrum using portable FTIR, high-resolution atmospheric absorption spectral information can be obtained. Furthermore, by employing efficient and feasible inversion algorithms, continuous monitoring of CO2 concentrations at specific geographical locations can be achieved, providing an effective measurement method for monitoring regional carbon sources and sinks. Summary of the Invention

[0006] To achieve the above objectives, this invention provides a method for rapid monitoring of atmospheric CO2 column concentration using a mobile ground-based system, comprising the following steps:

[0007] Step 1: Establishment of the CO2 column concentration inversion model. A multi-layer sensing machine learning model is established, using atmospheric radiative transfer simulation to generate training model data. The model is then trained and optimized to ensure the accuracy and reliability of data processing and CO2 concentration inversion.

[0008] Step 2: Assemble the portable Fourier Transform Infrared Spectrometer (FTIR) with the solar tracker, stepper motor, and portable power supply, ensuring stable installation of each component and conducting preliminary testing; input the current platform's geographical location information (latitude and longitude) and time data through the control device for solar tracking positioning; automatically adjust the angle and orientation of the Fourier Transform Infrared Spectrometer using the stepper motor and solar tracker to ensure the optimal angle for receiving sunlight; use FTIR to collect solar spectral data from the atmosphere.

[0009] Step 3: Process the collected spectral data using a pre-trained and tested machine learning inversion model to retrieve the corresponding atmospheric CO2 column concentration; display the retrieved data on the control device's screen and record the data for subsequent analysis.

[0010] Furthermore, the mobile ground-based rapid atmospheric CO2 monitoring platform in step one includes a rapid CO2 inversion monitoring method based on a multilayer perceptron (MLP) machine learning model. This method utilizes machine learning algorithms to invert solar spectral data collected by a Fourier transform infrared spectrometer to quickly and accurately invert the CO2 column concentration. The training of the machine learning model in step one specifically includes:

[0011] (1) Establishment of atmospheric dataset: Based on product data from space-based remote sensing satellites, such as atmospheric gas profiles, temperature, pressure, water vapor concentration profiles, CO2 column concentration, etc., an atmospheric dataset is established.

[0012] (2) Establishment of spectral dataset: Based on the atmospheric radiative transfer model and the established atmospheric dataset, the process of high-resolution solar spectrum being absorbed by the atmosphere, reflected multiple times by the solar tracker, and finally received by FTIR is simulated, and the forward calculation model of atmospheric radiation as shown in Equation (7) is established:

[0013] (7)

[0014] In the formula, It is the intensity of solar spectral radiation at the top of the atmosphere. It is the reflectivity of the mirror. The instrument linearity function of FTIR, Let be the spectral absorption coefficient of the i-th atmospheric layer. Solar radiation detected by FTIR.

[0015] (3) Based on the simulated absorption spectra and their corresponding CO2 concentration column datasets, train and test the multilayer perceptron machine learning model, which must ultimately meet the following requirements:

[0016] (8)

[0017] in, and These are the predicted value and the actual value at the i-th observation point, respectively. and These are the averages of the predicted value and the actual value, respectively. This represents the total number of observation points. Equation (8) measures the degree of linear correlation between the predicted value sequence and the true value sequence. If the result is as described in Equation (8), then the machine learning model is considered to perform well and can meet the inversion requirements of the measured spectral data.

[0018] In a preferred embodiment of the present invention, the MLP machine learning model in step one uses datasets from different geographical locations and different seasons during the training phase to enhance the model's generalization ability and accuracy.

[0019] In a preferred embodiment of the present invention, the solar tracker of the mobile ground-based atmospheric CO2 rapid monitoring platform in step two includes an angle adjustment device for automatically adjusting the angle of the portable Fourier transform infrared spectrometer to ensure that the best sunlight can be received at any time, thereby achieving real-time solar tracking.

[0020] In a preferred embodiment of the present invention, the monitoring method of the mobile ground-based rapid atmospheric CO2 monitoring platform in step three further includes an experimental data preprocessing step, namely, using a normalized data processing method to remove noise and interference in the spectral data, thereby improving the accuracy of the final CO2 column concentration estimation. When the monitoring location needs to be changed, the entire platform is moved to the new location, and then the above steps are repeated to monitor atmospheric CO2 at the new location.

[0021] In another preferred embodiment of the present invention, a portable remote sensing system for rapid monitoring of atmospheric carbon dioxide column concentration includes: a portable Fourier transform infrared spectrometer, a solar tracker, a stepper motor, a portable power supply box, and a laptop computer. The portable power supply box provides power to the infrared spectrometer, the laptop computer, and the stepper motor. The laptop computer is connected to the Fourier transform infrared spectrometer and the stepper motor. The laptop computer calculates the local zenith angle and azimuth angle based on the local latitude, longitude, and time information, and adjusts the solar tracker according to the zenith angle and azimuth angle information to ensure that sunlight enters the Fourier transform infrared spectrometer perpendicularly. The Fourier transform infrared spectrometer continuously and stably obtains solar spectral information and imports the information into the laptop computer. Finally, based on the high-resolution, high signal-to-noise ratio spectral data, the local carbon dioxide gas concentration is rapidly retrieved.

[0022] Compared with traditional monitoring methods, the mobile ground-based remote sensing atmospheric spectral monitoring system significantly improves the flexibility and efficiency of regional carbon monitoring. It enables more accurate and rapid atmospheric CO2 concentration retrieval, effectively monitoring CO2 emissions at specific geographical locations. Simultaneously, machine learning reduces the time consumed in retrieval calculations, maintaining high accuracy and application performance. The following will further explain the concept, specific structure, and resulting technical effects of this invention with reference to the accompanying drawings, to fully understand the purpose, features, and effects of this invention. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the mobile ground-based rapid atmospheric CO2 monitoring platform of the present invention;

[0024] Figure 2 Schematic diagram of the path of the ground-based atmospheric radiative transfer model;

[0025] Figure 3 This is a schematic diagram of the atmospheric forward calculation model of the present invention;

[0026] Figure 4 This is a schematic diagram illustrating the construction of the machine learning training model of the present invention;

[0027] Figure 5 This is a schematic diagram of the specific experimental measurement process of the present invention;

[0028] Figure 6 This is a schematic diagram of the fitting residuals of the forward simulated spectrum and the measured spectrum in this invention;

[0029] Figure 7 This is a schematic diagram illustrating the accuracy of the machine learning model in this invention;

[0030] Figure 8 This is a schematic diagram showing the comparison between the regional CO2 concentration inversion results and satellite data in an embodiment of the present invention. Detailed Implementation

[0031] The following description, with reference to the accompanying drawings, illustrates several preferred embodiments of the present invention to make its technical content clearer and easier to understand. The present invention can be embodied in many different forms, and the scope of protection of the present invention is not limited to the embodiments mentioned herein.

[0032] In the accompanying drawings, components with the same structure are indicated by the same numerical designation, and components with similar structures or functions are indicated by similar numerical designations. The dimensions and thicknesses of each component shown in the drawings are arbitrary, and the present invention does not limit the dimensions and thicknesses of each component. To make the illustrations clearer, the thickness of some components has been appropriately exaggerated in the drawings.

[0033] The purpose of this invention is to provide a mobile, ground-based rapid atmospheric CO2 column concentration monitoring platform. The method includes establishing a forward computational model based on the characteristics of the monitoring platform, creating a simulated spectral dataset and a CO2 column concentration dataset based on the model, and establishing an MLP (Multi-Level Processing) machine learning model. The mobile, ground-based rapid atmospheric CO2 monitoring platform collects high-quality solar spectra, ultimately enabling rapid and high-precision CO2 retrieval.

[0034] like Figure 1 The measurement and monitoring platform shown mainly consists of: 1. a portable Fourier transform infrared spectrometer, 2. a solar tracker, 3. a stepper motor, 4. a portable power supply box, and 5. a laptop computer. The specific implementation is as follows: 1. The portable power supply box powers the infrared spectrometer, laptop computer, and stepper motor; 2. The laptop computer connects to the Fourier transform infrared spectrometer and the stepper motor; 3. The laptop computer calculates the local zenith angle and azimuth angle based on the local latitude, longitude, and time information, and adjusts the solar tracker accordingly to ensure that sunlight enters the Fourier transform infrared spectrometer perpendicularly; 4. The Fourier transform infrared spectrometer continuously and stably acquires solar spectral information and imports the information into the laptop computer; finally, based on the high-resolution, high signal-to-noise ratio spectral data, the local greenhouse gas concentration is quickly retrieved.

[0035] Specifically, the main purpose of the forward calculation model for atmospheric radiative transfer is to simulate the actual solar spectrum received by the ground-based platform. This involves using solar radiation, Earth's atmospheric parameters (temperature, concentration, pressure, altitude, etc.), and relevant sensor instrument parameters as input information. Based on molecular absorption spectroscopy theory, it simulates the interaction between solar radiation and atmospheric molecules, ultimately simulating the process of the spectrum received by the ground-based platform. When a beam of light or broadband light passes through a homogeneous gaseous medium, its intensity will attenuate. This attenuation follows Beer-Lambert law:

[0036]

[0037] in The wave number before incidence is The initial light intensity, The intensity of light after passing through a gaseous medium. Let be the overall absorption coefficient of the gas molecules, representing the absorption coefficient of the gas molecules at wavenumbers of 1. The degree of light absorption depends on the relevant state parameters of the target gas, such as pressure, temperature, and concentration. Let h be the path length through the uniform gas medium and h be the thickness of the atmosphere. Typically, the integral of the absorption coefficient along the path through the gas is called the optical thickness. , can be represented as:

[0038]

[0039] Specifically, an atmospheric radiation transfer model was established for the ground-based experimental system. Figure 2 This invention describes the path of solar radiation transmitted through the atmosphere to the ground-based observation platform. When solar radiation is transmitted through the atmosphere, it can be assumed that the atmosphere is a parallel flat plate of a certain thickness. Solar radiation is first absorbed after interacting with the atmosphere, with the zenith angle... As the solar radiation reaches the observation site, the final solar spectral signal is received by the FTIR sensor. Based on this process, a forward calculation model of atmospheric radiative transfer for this measurement system can be established.

[0040] Assume that the optical path length of a certain layer in the atmosphere is The pressure, temperature, and concentration of CO2 in this atmospheric layer are respectively... , , For wavenumbers of (cm -1 The spectral optical thickness of the solar radiation at point () with perpendicular incidence is shown in equation (3):

[0041]

[0042] The atmospheric vertical optical thickness is:

[0043]

[0044] If the zenith angle of the sun's incidence is Then the single-layer oblique optical path According to the formula, the incident optical thickness can be calculated as follows: By combining the formulas for optical thickness and absorption spectral intensity, the spectral intensity of the solar spectrum at the top of the atmosphere after atmospheric absorption can be calculated as follows:

[0045]

[0046] In the formula Let represent the intensity of the incident solar spectrum. Using this formula, and referring to the weather conditions of the day, we can calculate the solar spectrum information that should be collected at our geographical location on that day.

[0047] Equation (5) does not specifically address the spectral resolution during calculation. To accurately simulate the radiation spectrum measured by the spectrometer, it is generally necessary to first calculate the line-by-line (LBL) spectrum and then convolve it with the instrument line shape function (ILSF) of the FTIR to ensure that the simulated spectrum has the same spectral resolution as the experimentally measured spectrum. During the experiment, the truncation function used by the FTIR in the ground-based measurement platform was the Boxcar function. Its main advantage is that it is a rectangular function in the frequency domain, has a flat frequency response, can preserve the shape and information of the broadband spectral signal, and has good signal smoothing and noise filtering effects, which can reduce the noise level of the measured signal. For this FTIR device, we used the line shape function shown in Equation (6):

[0048]

[0049] Where Res is the nominal resolution of FTIR. The solar radiation spectrum after convolution can be represented by equation (7):

[0050]

[0051] in It refers to the reflectivity of the mirrors on the solar tracker, which uses mirrors with identical emission characteristics on three surfaces.

[0052] Specifically, Figure 3 This paper describes a method for generating a simulated spectral dataset. First, an atmospheric state parameter dataset is constructed using four parameters: water vapor concentration profile, temperature profile, atmospheric pressure profile, and surface pressure within the region. The corresponding atmospheric absorption coefficients are then calculated using an established absorption model. Next, combined with geographic dataset information, an unconvolved simulated spectrum is obtained using an atmospheric transport model. Finally, the convolved model spectral dataset is obtained by analyzing the instrument line shape function of the corresponding Fourier transform infrared spectrometer and the instrument response.

[0053] Specifically, machine learning model building is as follows: Figure 4As shown. In traditional algorithms, obtaining CO2 column concentration typically requires multiple iterative forward calculations. This involves calculating atmospheric simulated spectra using given prior parameters, matching the simulated and measured spectra, determining convergence to identify the direction for parameter optimization, and then performing an optimization solution. This method is extremely computationally intensive and time-consuming for matching high-resolution spectra. However, using machine learning, this time-consuming forward calculation process can be pre-executed, pre-matching the corresponding spectra with the corresponding atmospheric state parameters. Specifically, in the experimental measurement example of this invention, a machine learning model can directly perform multi-parameter matching, significantly improving inversion efficiency. Specifically, this invention uses an MLP-trained model, taking the simulated ground-based spectrum obtained through forward calculation as input and CO2 column concentration information as output, to train the machine learning model. This model is then used for large-scale, high-precision inversion of atmospheric CO2 column concentration. In model training, the Pearson correlation coefficient is commonly used as a criterion for evaluating model performance, serving as a statistical indicator of the strength and direction of the linear relationship between two variables. Based on the absorption spectra and their corresponding CO2 concentration column datasets, a multilayer perceptron machine learning model was trained and debugged. The model was optimized using simulated and measured datasets. The final simulated data needs to meet the following requirements:

[0054]

[0055] In this formula, and These are the predicted value and the actual value at the i-th observation point, respectively. and These are the averages of the predicted and actual values, respectively. This is the total number of observation points. This formula measures the degree of linear correlation between the predicted value sequence and the actual value sequence. Equation (8) indicates that the machine learning model performs well and can meet the inversion requirements of experimental data.

[0056] Figure 5The measurement steps of the mobile ground-based atmospheric CO2 rapid monitoring platform are as follows: (1) After turning on the FTIR and the mobile notebook, determine the current measurement time and local latitude and longitude through the notebook, and store the time and location information as a txt file. (2) Read the txt file through the program and calculate the solar altitude angle and azimuth angle measured at this moment, and drive the motor to rotate the lens to the corresponding angle. Observe the signal-to-noise ratio of the FTIR spectral acquisition software at this time, and determine that the solar beam is perpendicular to the measuring instrument, so that spectral acquisition can be carried out. (3) After completing the spectral acquisition, observe whether the spectrum is stable and whether it meets the spectral signal-to-noise ratio requirements for inversion. If it does not meet the requirements, repeat the above measurement process. (4) If the acquisition results guarantee stability and high signal-to-noise ratio, store the spectral interferogram information and spectral spectrum information in the mobile notebook, and combine them with the pre-trained MLP machine learning model to quickly and accurately obtain the CO2 column concentration at this point and time.

[0057] In a preferred embodiment of the present invention Figure 6 and Figure 7 This demonstrates the accuracy of the forward computation and the precision of the inversion model. In this preferred example, the simulated spectrum calculated by the forward computation process matches the experimentally acquired spectrum well, with a spectral residual of less than 0.1%, indicating the accuracy of the forward simulation model and providing a reliable forward computation model for generating input spectra for machine learning. Simultaneously, the machine learning model trained using the simulated spectrum and the corresponding CO2 column concentration exhibits excellent linear correlation (R > 0.95) and a mean squared error (MSE) < 0.1 ppm, ensuring the reliability of the inversion results.

[0058] Specifically, in a preferred embodiment of the present invention, Figure 8 The results show a comparison between the CO2 inversion results in May at the Lingang Campus of Shanghai Jiao Tong University and the satellite inversion results in the measurement area. Through comprehensive verification and analysis of the results, the rationality and reliability of the results are ensured, demonstrating the feasibility of the mobile remote sensing atmospheric CO2 monitoring platform and its rapid CO2 inversion monitoring method.

[0059] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A mobile remote sensing method for rapid monitoring of atmospheric carbon dioxide column concentration, characterized in that, Includes the following steps: Step 1: Establishment of the carbon dioxide column concentration inversion model. A machine learning model based on multilayer perceptron (MLP) is established. Atmospheric radiative transfer simulation is used to generate training model data. The model is trained and optimized to ensure the accuracy and reliability of data processing and CO2 concentration inversion. Training and optimizing the model specifically includes: (1) Establishment of atmospheric dataset: Based on product data from space-based remote sensing satellites, such as atmospheric gas profiles, temperature, pressure, water vapor concentration profiles, and CO2 column concentration data, an atmospheric dataset is established; (2) Establishment of spectral dataset: Based on the atmospheric radiative transfer model and the established atmospheric dataset, the process of high-resolution solar spectrum being absorbed by the atmosphere, reflected multiple times by the solar tracker, and finally received by FTIR is simulated, and the forward calculation model of atmospheric radiation as shown in Equation (7) is established: （7） In the formula, It is the intensity of solar spectral radiation at the top of the atmosphere. It is the reflectivity of the mirror. The instrument linearity function of FTIR, Let be the spectral absorption coefficient of the i-th atmospheric layer. Solar radiation detected by FTIR; The zenith angle at which the sun is incident. It is a single-layer oblique optical path; (3) Based on the simulated absorption spectra and their corresponding CO2 concentration column datasets, train and test the multilayer perceptron machine learning model, which must ultimately meet the following requirements: （8） in, and These are the predicted value and the actual value at the i-th observation point, respectively. and These are the averages of the predicted value and the actual value, respectively. It is the total number of observation points; Equation (8) measures the degree of linear correlation between the predicted value sequence and the true value sequence; if it is as described in Equation (8), then the machine learning model is considered to perform well and can meet the inversion requirements of the measured spectral data; The multilayer perceptron (MLP) machine learning model uses datasets from different geographical locations and seasons during the training phase to enhance the model's generalization ability and accuracy; Step 2: Set up the portable Fourier transform infrared spectrometer (FTIR) with the solar tracker, stepper motor, and portable power supply, install it securely, and conduct preliminary testing; input the current platform's geographical location information (latitude and longitude) and time data through the control device for solar tracking positioning; Step 3: Process the collected spectral data using a pre-trained and tested machine learning inversion model to retrieve the corresponding atmospheric CO2 column concentration; display the retrieved data on the control device's screen and record the data for subsequent analysis; It also includes an experimental data preprocessing step, which uses a normalized data processing method to remove noise and interference from the spectral data in order to improve the accuracy of the final CO2 column concentration estimation.

2. The method as described in claim 1, characterized in that, The mobile ground-based rapid atmospheric CO2 monitoring platform in step one includes a rapid CO2 concentration inversion monitoring method based on a multilayer perceptron (MLP) machine learning model. This method uses machine learning algorithms to invert solar spectral data collected by a Fourier transform infrared spectrometer to quickly and accurately invert CO2 column concentration.

3. The method as described in claim 1, characterized in that, The solar tracker in step two includes an angle adjustment device for automatically adjusting the angle of the portable Fourier transform infrared spectrometer to ensure that optimal sunlight is received at any time, thus enabling real-time solar tracking.

4. The method as described in claim 1, characterized in that, In step two, a stepper motor and a solar tracker are used to automatically adjust the angle and direction of the Fourier transform infrared spectrometer to ensure the optimal angle for receiving sunlight.

5. The method as described in claim 4, characterized in that, In step two, a portable Fourier transform infrared spectrometer (FTIR) is used to collect solar spectral data in the atmosphere.

6. The method as described in claim 1, characterized in that, After the monitoring in step three is completed, if it is necessary to change the monitoring location, move the entire platform to the new location and repeat the above steps to monitor atmospheric CO2 at the new location.

7. A system employing the mobile remote sensing method for rapid monitoring of atmospheric carbon dioxide column concentration as described in any one of claims 1-6, characterized in that, The system includes a portable Fourier transform infrared spectrometer, a solar tracker, a stepper motor, a portable power supply box, and a laptop computer. The portable power supply box powers the infrared spectrometer, the laptop computer, and the stepper motor. The laptop computer connects to the Fourier transform infrared spectrometer and the stepper motor. The laptop computer calculates the local zenith angle and azimuth angle based on the local latitude, longitude, and time information, and adjusts the solar tracker according to this information to ensure that sunlight enters the Fourier transform infrared spectrometer perpendicularly. The Fourier transform infrared spectrometer continuously and stably acquires solar spectral information and imports the information into the laptop computer. Finally, based on the high-resolution, high signal-to-noise ratio spectral data, the local carbon dioxide gas concentration is rapidly retrieved.

Images