Method for constructing atmospheric CO2 high-resolution vertical profile based on machine learning

Abstract

The invention discloses a method for constructing an atmospheric CO2 high-resolution vertical profile based on machine learning, and belongs to the technical field of atmospheric environment monitoring, and the method comprises the following steps: (1) monitoring low-precision atmospheric data and high-precision atmospheric data at a ground observation point at the same time, and obtaining the low-precision atmospheric data and the high-precision atmospheric data which are continuously compared in the same period; the data serve as original data of machine learning; specifically, the low-precision atmospheric data is low-precision atmospheric CO2 concentration continuous monitoring data of a ground observation point; the high-precision atmosphere data comprises in-situ air pressure continuous monitoring data, in-situ temperature continuous monitoring data and high-precision atmosphere CO2 concentration continuous monitoring data of a ground observation point; (2) importing the original data into a machine learning model for machine learning to obtain a corrected vertical observation data regression model of the atmospheric CO2 concentration; (3) monitoring atmospheric data at the monitoring points in the vertical direction of the area to be measured, including low-precision atmospheric CO2 concentration vertical continuous monitoring data, vertical in-situ air pressure continuous monitoring data, vertical in-situ temperature continuous monitoring data and high-precision atmospheric CO2 concentration vertical monitoring data, obtaining the atmospheric data of the monitoring points in the vertical direction, and importing the obtained atmosphere data into the regression model, and outputting to obtain a high-resolution atmosphere CO2 vertical profile. The problems that an existing vertical profile is complex in construction, low in precision and low in practicability are solved.

Description

Technical Field

[0001] The present invention relates to the field of atmospheric environment monitoring technology, and in particular to a method for constructing a high-resolution vertical profile of atmospheric CO2 based on machine learning. Background Art

[0002] Since the Industrial Revolution, with the large-scale exploitation and use of fossil fuels, atmospheric CO2, SO2, NO X The concentrations of SO2 and NO are much higher than before the Industrial Revolution. The increase in atmospheric CO2 concentration is closely related to global warming and climate change. X It is the main precursor of acid rain and, combined with particulate matter in the atmosphere, can easily cause respiratory diseases. Therefore, it is necessary to monitor atmospheric CO2, SO2, NO X The concentration and its changing trend are of great significance.

[0003] In recent years, monitoring of atmospheric CO2 concentrations has primarily focused on the ground, using high-precision monitoring instruments to monitor concentration changes in real time. However, high-resolution and high-precision monitoring of concentration changes along vertical gradients or at a specific altitude is limited by the heavy, bulky, and expensive nature of these instruments, making them difficult to transport to high altitudes. Research on this topic is relatively limited. While observation towers can be used to bring atmospheric air from different altitudes to the ground using Teflon tubing for high-precision monitoring, the limited height, high construction costs, and high maintenance requirements, coupled with the need for infrastructure (such as a computer room and a stable power supply), have significantly limited research. Low-precision miniature atmospheric CO2 probes offer advantages such as light weight, low cost, ease of use, and portability. However, their measurement accuracy is significantly affected by changes in atmospheric pressure. When observing atmospheric CO2 concentration vertically, atmospheric pressure decreases with increasing altitude, leading to significant measurement offsets and difficulties in maintaining accuracy. This significantly impacts high-precision and high-resolution vertical atmospheric CO2 concentration monitoring. The Chinese patent "Three-dimensional simulation method and system of atmospheric CO2 profile based on fusion of multi-source data from the sky and the ground" also discloses a simulation method of the atmospheric CO2 profile. The patent focuses on the integration of multi-source observations (ground observation data, monitoring data from balloons or aircraft, and carbon dioxide concentration data monitored by carbon satellites), and uses models such as carbontracker for prediction. The obtained profile data is combined with humidity profiles, temperature profiles, etc. At the same time, the profile calibration lacks the calibration of high-precision test values of atmospheric samples. It is a pure simulation result, the construction method is complex, and the accuracy is poor.

[0004] Based on the above background, how to use low-cost, low-precision miniature atmospheric CO2 probes to construct high-resolution vertical profiles of CO2 to serve greenhouse gas scientific research and the formulation of carbon emission reduction policies is an urgent problem that needs to be solved in this invention. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a method for constructing a high-resolution vertical profile of atmospheric CO2 based on machine learning, so as to solve the problems of complex vertical profile construction, low precision and poor practicality.

[0006] The present invention solves the above technical problems through the following technical solutions:

[0007] A method for constructing a high-resolution vertical profile of atmospheric CO2 based on machine learning, the method is as follows:

[0008] (1) Simultaneously monitor low-precision atmospheric data and high-precision atmospheric data at ground observation points to obtain continuous low-precision atmospheric data and high-precision atmospheric data for comparison over the same period as raw data for machine learning;

[0009] Specifically, low-precision atmospheric data include: low-precision atmospheric CO2 concentration continuous monitoring data at ground observation points; high-precision atmospheric data include: in-situ air pressure continuous monitoring data, in-situ temperature continuous monitoring data, and high-precision atmospheric CO2 concentration continuous monitoring data at ground observation points;

[0010] (2) Importing the original data into the machine learning model for machine learning to obtain the vertical observation data regression model of the corrected atmospheric CO2 concentration;

[0011] (3) Monitor atmospheric data at monitoring points in the vertical direction of the area to be measured, including low-precision vertical continuous monitoring data of atmospheric CO2 concentration, vertical in-situ continuous monitoring data of air pressure, vertical in-situ continuous monitoring data of temperature, and high-precision vertical monitoring data of atmospheric CO2 concentration. Obtain atmospheric data from monitoring points in the vertical direction, import the acquired atmospheric data into the regression model, and output a high-resolution vertical profile of atmospheric CO2.

[0012] When constructing a high-resolution atmospheric CO2 vertical profile, a high-precision atmospheric CO2 monitor can be used to monitor the vertical change of atmospheric CO2 concentration in the vertical direction. However, due to the large size of the high-precision atmospheric CO2 monitor and the required voltage reaching 220V, it is difficult to actually use it in the vertical high altitude. Although the low-precision miniature atmospheric CO2 probe can continuously monitor the atmospheric CO2 concentration change data, its accuracy is low and the constructed vertical profile is offset. Therefore, the present invention first collects low-precision atmospheric data and high-precision atmospheric data from ground observation points to construct a vertical observation data regression model. On the basis of this regression model, the low-precision atmospheric CO2 concentration vertical continuous monitoring data, vertical in-situ pressure continuous monitoring data, vertical in-situ temperature continuous monitoring data, and high-precision atmospheric CO2 concentration vertical monitoring data of the monitoring points in the vertical direction of the measured area are imported into the model to obtain a high-resolution atmospheric CO2 vertical profile.

[0013] Furthermore, in (1) and (3), the low-precision atmospheric CO2 concentration continuous monitoring data and the low-precision atmospheric CO2 concentration vertical continuous monitoring data are obtained by monitoring using a low-precision miniature atmospheric CO2 probe, and the in-situ air pressure continuous monitoring data, in-situ temperature continuous monitoring data and vertical in-situ air pressure continuous monitoring data, and vertical in-situ temperature continuous monitoring data are obtained by monitoring using a miniature meteorological instrument.

[0014] Furthermore, in (1), the high-precision continuous monitoring data of atmospheric CO2 concentration at the ground observation point is obtained by continuous monitoring using a high-precision atmospheric CO2 monitor during ground monitoring, and in (3), the high-precision vertical monitoring data of atmospheric CO2 concentration at the monitoring point in the vertical direction is obtained by collecting atmospheric samples using an atmospheric sample collection system during high-altitude monitoring and testing using a high-precision atmospheric CO2 monitor when returning to the ground.

[0015] Low-precision miniature atmospheric CO2 probes are small and portable, and can be used on high-altitude observation platforms. They typically continuously monitor atmospheric CO2 concentrations at ground or vertical observation points at 15-second intervals, providing low-precision continuous atmospheric CO2 concentration monitoring data. Miniature weather meters can also be used on high-altitude observation platforms to continuously monitor in-situ air pressure and temperature at ground or vertical observation points at 15-second intervals, providing continuous in-situ air pressure and temperature monitoring data at different altitudes.

[0016] High-precision atmospheric CO2 monitors can directly obtain continuous high-precision atmospheric CO2 concentration data at ground observation points, but it is difficult to carry out monitoring activities at high altitudes in the vertical direction. Therefore, in (3), an atmospheric sample collection system is used to collect atmospheric samples at monitoring points at different altitudes in the vertical direction. When returning to the ground, a high-precision atmospheric CO2 monitor is used to test the atmospheric CO2 concentration, and high-precision atmospheric CO2 concentration vertical monitoring data at different altitudes are obtained.

[0017] Furthermore, in (1), at least two ground observation points are selected to monitor low-precision atmospheric data and high-precision atmospheric data, and continuous monitoring comparative data for the same period of at least one week are obtained respectively.

[0018] When collecting raw data, it is necessary to select at least two different ground observation points to monitor relevant data, and at the same time, each location must obtain at least one week of continuous monitoring comparison data for the same period.

[0019] Furthermore, in (3), the difference in altitude between the selected monitoring point and the ground of the area to be measured in the vertical direction is 20-2000 meters.

[0020] Furthermore, in (3), at least seven monitoring points in the vertical direction are selected to obtain high-precision vertical monitoring data of atmospheric CO2 concentration at at least seven altitude points in the vertical direction.

[0021] Furthermore, based on the altitude of the ground observation point, seven vertical monitoring points are selected, and the vertical altitude difference between the monitoring point and the ground is preferably 20 meters, 100 meters, 200 meters, 500 meters, 1000 meters, 1500 meters and 1700 meters.

[0022] When constructing a vertical profile of atmospheric CO2 in the area to be measured, it is necessary to obtain atmospheric CO2 concentration values at at least seven vertical altitude points to construct high-precision vertical atmospheric CO2 concentration monitoring data. Specifically, atmospheric samples are collected at each altitude point, and upon return, high-precision atmospheric CO2 concentration data is measured at that altitude point using a high-precision atmospheric CO2 monitor. This high-precision vertical atmospheric CO2 concentration monitoring data is then constructed.

[0023] Furthermore, the machine learning model in (2) can adopt one of multivariate linear regression, random forest, support vector regression, gradient boosting regression tree, ridge regression, elastic network regression, and multilayer perceptron regression.

[0024] Furthermore, the machine learning model selected the random forest model, and after testing, its R 2 =0.9595; RMSE=2.1424.

[0025] Furthermore, at the ground observation point, the micro-meteorological instrument and the high-precision atmospheric CO2 monitor monitor the gas data at the low-precision micro-atmospheric CO2 probe; when monitoring the point in the vertical direction, a high-altitude observation platform is used to monitor the atmospheric data of the monitoring point in the vertical direction, and the micro-meteorological instrument and the atmospheric sample sampling system carried by the high-altitude observation platform respectively monitor and collect the gas data and gas at the low-precision micro-atmospheric CO2 probe.

[0026] The layout of low-precision miniature atmospheric CO2 probes and miniature meteorological instruments at ground observation points and vertical observation points must be completely consistent. The low-precision atmospheric CO2 concentration, in-situ air pressure, and in-situ temperature monitored are required to be at the same altitude to prevent the atmospheric CO2 concentration, in-situ air pressure, and in-situ temperature from being at data at points other than the same altitude, thereby eliminating errors caused by human factors.

[0027] Furthermore, high-altitude observation platforms can choose drones or retained balloons.

[0028] The present invention discloses an application of a method for constructing a high-resolution vertical profile of atmospheric CO2 based on machine learning, characterized in that the method is also suitable for constructing high-precision and high-resolution vertical profiles of gaseous pollutants in field or sudden atmospheric environmental pollution environments, and the gaseous pollutants include SO2, CO, and NO2.

[0029] Of course, the method of the present invention is suitable for constructing high-resolution vertical profiles of atmospheric CO2, and can also be extended to construct vertical profiles of CO, NO2, and SO2.

[0030] Beneficial effects:

[0031] (1) The present invention uses the vertical observation data of low-cost miniature low-precision atmospheric CO2 probes to construct a high-resolution vertical profile of atmospheric CO2 concentration, overcoming the problem that the measurement accuracy of low-precision miniature atmospheric CO2 probes decreases as the ambient atmospheric pressure changes, and can serve greenhouse gas research and carbon emission reduction policy formulation.

[0032] (2) The method involved in the present invention can obtain high-resolution and high-precision profiles of atmospheric CO2 concentration or gas pollutant concentration in the wild or under sudden atmospheric pollution, serving scientific research or atmospheric pollution control.

[0033] (3) The method involved in the present invention has the advantages of low cost, high precision, simple process, easy operation and easy promotion. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 :A method for constructing high-resolution vertical profiles of atmospheric CO2;

[0035] Figure 2 : Comparison of vertical profile results constructed by different methods. DETAILED DESCRIPTION

[0036] The present invention will be described in detail below with reference to specific embodiments and accompanying drawings:

[0037] Example 1: Regression model construction

[0038] (1) Simultaneously monitor low-precision atmospheric data and high-precision atmospheric data at ground observation points to obtain continuous low-precision atmospheric data and high-precision atmospheric data for comparison over the same period as the raw data for machine learning; specifically:

[0039] Low-precision atmospheric data include: low-precision atmospheric CO2 concentration continuous monitoring data from ground observation points;

[0040] High-precision atmospheric data include: in-situ continuous monitoring data of air pressure, in-situ continuous monitoring data of temperature, and high-precision continuous monitoring data of atmospheric CO2 concentration at ground observation points;

[0041] When selecting observation points, it is necessary to select at least two different ground observation points to monitor low-precision atmospheric data and high-precision atmospheric data, and obtain at least one week of continuous monitoring comparison data for the same period at each ground observation point.

[0042] (2) The obtained raw data, including low-precision atmospheric data and high-precision atmospheric data, are imported into the machine learning model to obtain a vertical observation data regression model of the corrected atmospheric CO2 concentration.

[0043] (3) Selecting vertical monitoring points in the area to be measured to monitor atmospheric data. When selecting, the vertical monitoring points should have an altitude difference of 20-2000 meters from the ground of the area to be measured, and at least seven vertical monitoring points should be selected. The monitoring points are preferably selected at seven monitoring points with an altitude difference of 20 meters, 100 meters, 200 meters, 500 meters, 1000 meters, 1500 meters, and 1700 meters from the ground.

[0044] Obtain vertical atmospheric data from monitoring points in the vertical direction: low-precision vertical continuous monitoring data of atmospheric CO2 concentration, vertical in-situ continuous monitoring data of air pressure, vertical in-situ continuous monitoring data of temperature, and high-precision vertical monitoring data of atmospheric CO2 concentration;

[0045] When obtaining ground data in step (1), specifically when monitoring at ground observation points, low-precision atmospheric CO2 concentration data is obtained by monitoring using a low-precision micro-atmospheric CO2 probe, in-situ continuous air pressure monitoring data and in-situ continuous temperature monitoring data are obtained by monitoring using a micro-meteorological instrument, and high-precision continuous atmospheric CO2 concentration monitoring data is obtained by directly monitoring using a high-precision atmospheric CO2 monitor.

[0046] When acquiring data in the vertical direction in step (3), specifically, a high-altitude observation platform (such as a drone, a retained balloon, etc.) is equipped with a low-precision micro-atmospheric CO2 probe to monitor low-precision vertical continuous monitoring data of atmospheric CO2 concentration, vertical in-situ continuous pressure monitoring data and vertical in-situ continuous temperature monitoring data are monitored using a micro-meteorological instrument, and an atmospheric sample collection system is also equipped to collect atmospheric samples of monitoring points in the vertical direction. When collecting atmospheric samples of monitoring points in the vertical direction, atmospheric samples are collected from seven monitoring points with altitude differences of 20 meters, 100 meters, 200 meters, 500 meters, 1000 meters, 1500 meters and 1700 meters from the ground, respectively. Then, when the high-altitude observation platform returns to the ground, a high-precision atmospheric CO2 monitor is used to monitor the atmospheric samples collected from each monitoring point to obtain the atmospheric CO2 concentration of each monitoring point in the vertical direction, and finally obtain high-precision vertical monitoring data of atmospheric CO2 concentration.

[0047] The atmospheric data of the monitoring points in the vertical direction are imported into the regression model, and the high-resolution vertical profile of atmospheric CO2 is output.

[0048] In addition, the layout of the low-precision micro-atmospheric CO2 probes and micro-meteorological instruments at the ground observation points and the vertical monitoring points must be completely consistent, requiring the monitored low-precision atmospheric CO2 concentration, in-situ air pressure, and in-situ temperature to be at the same height. Specifically, the micro-meteorological instruments at the ground observation points and the vertical monitoring points monitor the in-situ air pressure and in-situ temperature at the low-precision micro-atmospheric CO2 probes. The high-precision atmospheric CO2 monitor also monitors the high-precision atmospheric CO2 concentration at the low-precision micro-atmospheric CO2 probes at the ground observation points, while the vertical monitoring points collect atmospheric samples at the low-precision micro-atmospheric CO2 probes at the monitoring points, and then monitor the high-precision atmospheric CO2 concentration values after monitoring the atmospheric samples.

[0049] Example 2: Selection of regression model

[0050] Experimental area: Xi'an and the hinterland of Qinling Mountains

[0051] Specific ground observation points: Institute of Earth Environment, Chinese Academy of Sciences (500m) and Heihe Forest Park (1700m).

[0052] Observation time: 1 consecutive week.

[0053] Instruments: Low-precision micro atmospheric CO2 probe, micro weather meter (aligned with the probe of low-precision micro atmospheric CO2 tester), and high-precision atmospheric CO2 monitor should be arranged at the same level, and each instrument collects data at a frequency of every 15 seconds.

[0054] The data was obtained based on the method of Example 1, and the original data was obtained after removing the abnormal observation data.

[0055] The raw data were fed into the following machine learning models: multiple linear regression (MLR), random forest (RF), support vector regression (SVR), gradient boosted regression tree (GBRT), ridge regression (RR), elastic network regression (ElasticNet), and multilayer perceptron regression (MLP regressor). The raw data were then subjected to regression modeling and analysis using the scikit-learn library, yielding the results in Table 1.

[0056] The root mean square error (RMSE) and the coefficient of determination (R 2 ) is a performance indicator for evaluating machine learning models. The RMSE and R obtained by each machine learning model 2 As shown in Table 1.

[0057] RMSE is an indicator used to measure the prediction accuracy of a prediction model on continuous data. It measures the root mean square difference between the predicted value and the true value, indicating the average degree of deviation between the predicted value and the true value. The smaller the RMSE value, the better. If the RMSE is close to 0, it means that there is almost no difference between the predicted value and the actual value. 2 It measures the ratio between the variation of the dependent variable predicted by the regression model and the total variation of the dependent variable. Its value range is between 0 and 1. The closer it is to 1, the stronger the explanatory power of the regression model for the dependent variable.

[0058]

[0059] where y i，pred is the i-th predicted value of the model, y i is the i-th observation value of the reference instrument, is the mean value of the reference instrument observation set.

[0060] Table 1

[0061]

[0062] Analysis of the data in Table 1 shows that among the seven regression models above, the performance of the random forest (RF) model (R 2 =0.9895; RMSE=2.1424) are better than the other six machine learning models, so the random forest model is selected as the machine learning model of the present invention.

[0063] Example 3: Construction of a high-resolution vertical profile of atmospheric CO2 concentration

[0064] Based on Examples 1 and 2, a high-altitude observation platform with an unmanned aerial vehicle as the carrier, carrying a low-precision miniature atmospheric CO2 probe, a miniature meteorological instrument and an atmospheric sample collection system, was used to observe the vertical distribution of high-altitude atmospheric CO2 concentration within an altitude difference of 2000 meters on November 26, 2024, to verify the feasibility and accuracy of the method.

[0065] Area to be measured: Vertical profile of high-altitude atmospheric CO2 at the Qinling Observatory at an altitude of 500 meters.

[0066] Vertical monitoring points: vertical monitoring point 1 at an altitude of 520 meters (altitude difference of 20 meters), vertical monitoring point 2 at an altitude of 600 meters (altitude difference of 100 meters), vertical monitoring point 3 at an altitude of 700 meters (altitude difference of 200 meters), vertical monitoring point 4 at an altitude of 1000 meters (altitude difference of 500 meters), vertical monitoring point 5 at an altitude of 1500 meters (altitude difference of 1000 meters), vertical monitoring point 6 at an altitude of 2000 meters (altitude difference of 1500 meters), and vertical monitoring point 7 at an altitude of 2200 meters (altitude difference of 1700 meters).

[0067] Unmanned aerial vehicles (UAVs) are used to monitor data from vertical monitoring points. Low-precision micro-atmospheric CO2 probes, micro-meteorological instruments, and atmospheric sample collection systems (air bags) are deployed on the UAVs, with all instruments positioned at the same level. This provides low-precision vertical continuous monitoring data for atmospheric CO2 concentration, vertical in-situ continuous pressure monitoring data, vertical in-situ continuous temperature monitoring data, and atmospheric samples collected at each vertical monitoring point. Upon returning to the ground, high-precision atmospheric CO2 monitors are used to monitor each atmospheric sample, obtaining high-precision atmospheric CO2 concentrations at each monitoring point. This, in turn, provides high-precision vertical monitoring data for atmospheric CO2 concentrations at each vertical monitoring point.

[0068] Among them, the high-precision atmospheric CO2 concentration data at 7 vertical monitoring points are shown in Table 2.

[0069] Table 2

[0070]

[0071] The low-precision vertical continuous monitoring data of atmospheric CO2 concentration, vertical in-situ continuous monitoring data of atmospheric pressure, vertical in-situ continuous monitoring data of temperature, and high-precision vertical monitoring data of atmospheric CO2 concentration at vertical monitoring points are introduced into the vertical observation data regression model of atmospheric CO2 concentration corrected by the present invention. The data deviation is shown in Table 3 and the constructed vertical profile is shown in Figure 2 .

[0072] Table 3

[0073]

[0074] From the results we can see that:

[0075] As shown in Table 3, Figure 2 As shown in the figure, the vertical profile directly constructed by using a low-precision micro atmospheric CO2 probe (probe original value) has a large deviation (47.7±38.4ppm) from the profile constructed by the high-precision test value (high-precision value) of the atmospheric sample, while the vertical profile constructed by using the method of the present invention (probe correction value) is basically consistent with the vertical profile formed by the high-precision test value, and the deviation (0.2±4.6ppm) achieves an ideal high-resolution effect.

[0076] The above embodiments are intended only to illustrate the technical solutions of the present invention and are not intended to limit the scope of the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art will appreciate that the technical solutions of the present invention may be modified or replaced with equivalents without departing from the spirit and scope of the technical solutions of the present invention, and such modifications or equivalents shall be encompassed by the claims of the present invention. Any techniques, shapes, and structures not described in detail herein are well known.

Claims

1. A method for constructing a high-resolution vertical profile of atmospheric CO2 based on machine learning, characterized in that: The method is as follows: (1) Simultaneously monitor low-precision atmospheric data and high-precision atmospheric data at ground observation points to obtain continuous low-precision atmospheric data and high-precision atmospheric data for comparison over the same period as raw data for machine learning; Specifically, low-precision atmospheric data include: low-precision atmospheric CO2 concentration continuous monitoring data at ground observation points; high-precision atmospheric data include: in-situ air pressure continuous monitoring data, in-situ temperature continuous monitoring data, and high-precision atmospheric CO2 concentration continuous monitoring data at ground observation points; (2) Importing the original data into the machine learning model for machine learning to obtain the vertical observation data regression model of the corrected atmospheric CO2 concentration; (3) Monitor atmospheric data at monitoring points in the vertical direction of the area to be measured, including: low-precision vertical continuous monitoring data of atmospheric CO2 concentration, vertical in-situ continuous monitoring data of air pressure, vertical in-situ continuous monitoring data of temperature, and high-precision vertical monitoring data of atmospheric CO2 concentration. Obtain atmospheric data at the monitoring points in the vertical direction, import the obtained atmospheric data into the regression model, and output a high-resolution vertical profile of atmospheric CO2.

2. The method for constructing a high-resolution vertical profile of atmospheric CO2 based on machine learning according to claim 1, characterized in that: In (1) and (3), the low-precision atmospheric CO2 concentration continuous monitoring data and the low-precision atmospheric CO2 concentration vertical continuous monitoring data are obtained by monitoring using a low-precision miniature atmospheric CO2 probe, and the in-situ air pressure continuous monitoring data, in-situ temperature continuous monitoring data and vertical in-situ air pressure continuous monitoring data, and vertical in-situ temperature continuous monitoring data are obtained by monitoring using a miniature meteorological instrument.

3. The method for constructing a high-resolution vertical profile of atmospheric CO2 based on machine learning according to claim 2, characterized in that: In (1), the high-precision continuous monitoring data of atmospheric CO2 concentration at the ground observation point is obtained by continuous monitoring using a high-precision atmospheric CO2 monitor during ground monitoring. In (3), the high-precision vertical monitoring data of atmospheric CO2 concentration at the monitoring point in the vertical direction is obtained by collecting atmospheric samples using an atmospheric sample collection system during high-altitude monitoring and testing using a high-precision atmospheric CO2 monitor when returning to the ground.

4. The method for constructing a high-resolution vertical profile of atmospheric CO2 based on machine learning according to claim 3, characterized in that: In (1), at least two ground observation points are selected to monitor low-precision atmospheric data and high-precision atmospheric data, and continuous monitoring comparison data for the same period of at least one week are obtained respectively.

5. The method for constructing a high-resolution vertical profile of atmospheric CO2 based on machine learning according to claim 4, characterized in that: In (3), the difference in altitude between the selected monitoring point and the ground of the area to be measured in the vertical direction is 20-2000 meters.

6. The method for constructing a high-resolution vertical profile of atmospheric CO2 based on machine learning according to claim 5, characterized in that: In (3), at least 7 monitoring points in the vertical direction are selected to obtain high-precision vertical monitoring data of atmospheric CO2 concentration at at least 7 altitude points in the vertical direction.

7. The method for constructing a high-resolution vertical profile of atmospheric CO2 based on machine learning according to claim 6, characterized in that: The machine learning model in (2) can be one of multiple linear regression, random forest, support vector regression, gradient boosting regression tree, ridge regression, elastic network regression, and multilayer perceptron regression.

8. The method for constructing a high-resolution vertical profile of atmospheric CO2 based on machine learning according to claim 7, characterized in that: The machine learning model selected is a random forest model.

9. The method for constructing a high-resolution vertical profile of atmospheric CO2 based on machine learning according to claim 1, characterized in that: At the ground observation point, the micro-meteorological instrument and the high-precision atmospheric CO2 monitor monitor the gas data at the low-precision micro-atmospheric CO2 probe; when monitoring in the vertical direction, a high-altitude observation platform is used to monitor the atmospheric data of the monitoring points in the vertical direction, and the micro-meteorological instrument and the atmospheric sample sampling system carried by the high-altitude observation platform respectively monitor and collect the gas data and gas at the low-precision micro-atmospheric CO2 probe.

10. Application of the method for constructing a high-resolution vertical profile of atmospheric CO2 based on machine learning according to any one of claims 1 to 9, characterized in that: The method is also applicable to constructing high-precision and high-resolution vertical profiles of gaseous pollutants in outdoor or sudden atmospheric pollution environments, wherein the gaseous pollutants include CO, NO2, and SO2.

Images