Method and system for monitoring urban co2 plume by integrating observation and simulation

Abstract

The application belongs to the technical field of greenhouse gas emission monitoring, and discloses a method and system for monitoring urban CO2 emission plume by integrating observation and simulation. The method comprises the following steps: collecting and preprocessing multi-source data; constructing a prediction model for atmospheric CO2 column concentration (XCO2) based on machine learning and multi-source satellite observation; simulating long-term artificial CO2 concentration by using an atmospheric transmission model; simulating hourly artificial CO2 concentration data; and constructing an artificial CO2 emission estimation model based on optimal estimation and verifying the results. The application has high automation degree, can automatically identify emission plume on the basis of multi-source data fusion and model simulation, reduces the need for manual intervention, improves monitoring efficiency, and is suitable for large-scale urban emission source monitoring and management.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of greenhouse gas emission monitoring, and particularly relates to a method and system for monitoring urban CO2 plume by integrating observation and simulation. BACKGROUND

[0002] Based on the observation data of atmospheric CO2 concentration, the inversion of anthropogenic CO2 emissions has become an effective means to verify and correct the greenhouse gas emission inventory. Currently, the carbon satellites in orbit include GOSAT / GOSAT-2, OCO-2 / 3 and TanSat, etc. These satellites provide global-scale atmospheric CO2 column concentration (XCO2) inversion data, which has great application potential in monitoring CO2 emissions at city and point source scales, and has the advantages of objectivity, wide range and repeated observation. However, due to the influence of satellite observation mode and cloud coverage, the effective observation times through urban areas are often limited. CO2 in the atmosphere is a long-lived gas, and it is a key step to accurately extract the XCO2 enhancement signal caused by anthropogenic emissions from the high background level. Due to the influence of atmospheric transmission, ecological flux and sensor sensitivity, etc., it is difficult to realize the continuous monitoring and dynamic tracking of urban carbon emissions only by relying on the atmospheric CO2 concentration data observed by carbon satellites.

[0003] Prior art one: a method for identifying point source plume based on carbon satellite data, discloses:

[0004] The threshold segmentation method, Gaussian plume simulation method and cross-sectional flux method are used to identify the plume of atmospheric CO2 column concentration (XCO2) data obtained by satellite observation technology.

[0005] Specifically, the threshold segmentation method relies on the geographical location and spatial range of the emission source, and takes the downwind area less affected by anthropogenic emissions as prior information to accurately divide the anthropogenic emission area and the background area. This method obtains the local XCO2 enhancement value by subtracting the background concentration value from the satellite observation value of the anthropogenic emission area. On this basis, according to the spatial distribution continuity of the observation points, the edge detection and shape constraint of the plume are carried out to ensure the accuracy of the plume boundary identification.

[0006] The Gaussian plume model and the cross-sectional flux method are the mainstream methods for inverting the emission source rate from atmospheric CO2 plume observation. Under the premise of stable wind field, the CO2 concentration distribution near the independent emission point source is taken as a function of wind speed and wind direction. The Gaussian plume assumes that the atmospheric CO2 concentration distribution observed by the satellite conforms to the Gaussian shape, and the satellite observation value of the emission source is fitted. The cross-sectional flux method derives the source rate as the product of the local wind field and the concentration integrated on the plume cross section.

[0007] The disadvantages of prior art one are that these methods have strict conditions for the use of research subjects and satellite observation data, and have certain limitations.

[0008] (1) The location and spatial range of the emission source are predetermined, which makes these methods more suitable for independent large cities or studies of specific point sources of emissions. When it comes to the study of anthropogenic emissions in multiple cities or complex regions, the threshold selection and emission source determination often rely too much on expert knowledge and manual processing, resulting in a lack of objectivity and automation in the identification process, which limits its large-scale application.

[0009] (2) The requirement of spatially continuous pixel distribution of satellite observation data is the core premise of these methods. Current carbon satellites in orbit, such as OCO-2 / OCO-3 satellites, usually observe in a strip or snapshot mode, which is affected by factors such as clouds, aerosols, and weather conditions. The XCO2 data retrieved from satellite observations has a large number of blanks and uneven distribution. These data blanks greatly affect the accurate identification of CO2 plumes and the accurate inversion of emission amounts, increasing the uncertainty of the model.

[0010] Prior art two: anthropogenic carbon dioxide emission quantification method based on collaborative observation of pollution and carbon, discloses:

[0011] Anthropogenic CO2 emissions and conventional pollutant emissions have the same root and origin. NO2 and CO are two major air pollutants and good CO2 tracers. The anthropogenic carbon dioxide emission quantification method based on collaborative observation of pollution and carbon uses satellite remote sensing technology to jointly observe multiple pollutants and greenhouse gases in the atmosphere, analyzes the correlation and spatial and temporal distribution characteristics of pollutants and CO2 emissions from multiple dimensions and time scales, accurately identifies emission sources, and indirectly estimates anthropogenic CO2 emissions. This method integrates CO2 and its accompanying pollutants, such as carbon monoxide (CO), nitrogen oxides (NO x ) and other observation data, uses the spatial and temporal correlation characteristics of pollutants and CO2 emissions, optimizes the identification accuracy of emission sources and the estimation accuracy of emission amounts, and makes up for the limitations of single gas observation methods and the low sensitivity of CO2 to emission changes.

[0012] The disadvantages of the prior art two are that the satellite observation of atmospheric CO2 and pollutant data still has the problems of insufficient spatial coverage and discontinuous data, especially in cloudy areas or complex terrain, which will affect the spatio-temporal accuracy of collaborative analysis and increase the uncertainty of emission inversion. Although pollutants and CO2 emissions have certain correlation in time and space, such relationship is not always linear or simple. The emission proportion of different emission sources, fuel composition, control technology, etc. will affect the degree of collaboration of pollutants and CO2. Therefore, when CO2 emissions are derived based on pollutants, the model assumption may cause deviation. In the area where multi-source emissions overlap, such as large cities or industrial agglomeration areas, the pollutant characteristics of different emission sources may overlap with each other, making it difficult to accurately identify the pollution source. In addition, the emission intensity and timing of different pollution sources may be similar, further increasing the complexity of emission source classification.

[0013] The prior art three is a reconstruction technology of atmospheric CO2 spatio-temporal continuous data set using multi-source data and machine learning, which discloses that by constructing a prediction model with atmospheric CO2 column concentration (XCO2) as the target variable and various parameters related to natural circulation process and human emission as the prediction variables, the complex nonlinear relationship of these variables is trained by using machine learning algorithm, so as to fill in the blank area of satellite observation of XCO2. Through this technical scheme, high spatial resolution, spatio-temporal continuous atmospheric CO2 concentration mapping analysis can be generated, and the problem of discontinuous satellite observation data can be solved.

[0014] Drawbacks of the prior art three: Some SCI use machine learning algorithms to reconstruct atmospheric CO2 datasets, and most studies do not use pollutants homologous to CO2 emissions as predictor variables. Although some studies introduce NO2 as a predictor variable, its contribution to the prediction of XCO2 concentration lacks in-depth discussion, and the impact mechanism of pollutants on XCO2 is not fully analyzed. For example: Guo, K.; Lei, L.; Sheng, M.; Ji, Z.; Song, H. Refining Spatial and Temporal XCO2 Characteristics Observed by Orbiting Carbon Observatory-2 and Orbiting Carbon Observatory-3 Using Sentinel-5P Tropospheric Monitoring Instrument NO2 Observations in China. Remote Sens. 2024, 16, 2456. https: / / doi.org / 10.3390 / rs16132456; and He, Z., Fan, G., Li, X., Gong, F.-Y., Liang, M., Gao, L., Zhou, M., 2024. Spatio-temporal modeling of satellite-observed CO2 columns in China using deep learning. International Journal of Applied Earth Observation and Geoinformation 129, 103859.

[0015] https: / / doi.org / https: / / doi.org / 10.1016 / j.jag.2024.103859; the above 2 papers use NO2, but not CO. In addition, it has not been fully verified whether the reconstructed XCO2 dataset can accurately reproduce the anthropogenic XCO2 enhancement features identified in satellite observations, which affects the application effect of using reconstructed data for emission quantification research. SUMMARY

[0016] In order to overcome the problems in the prior art, the application discloses a kind of observation and simulation integrated city CO2 emission plume monitoring method and system, specifically relates to a kind of observation and simulation integrated city CO2 emission plume monitoring method.The application aims at fully integrated multi-source satellite observation atmospheric CO2 and pollutant concentration, land ecological remote sensing parameter and meteorological data, innovatively developed a kind of CO2 emission plume automatic identification technology combining machine learning algorithm and atmospheric transmission model, can realize the long-term dynamic monitoring of city CO2 emission.

[0017] The technical scheme is as follows: the observation and simulation integrated city CO2 emission plume monitoring method comprises:

[0018] S1, multi-source data collection and preprocessing;

[0019] S2, constructing XCO2 prediction model based on machine learning and multi-source satellite observation;

[0020] S3, simulating hourly anthropogenic CO2 concentration data using anthropogenic CO2 concentration simulation results of atmospheric transmission model;

[0021] S4, verifying anthropogenic CO2 emission estimation model and results based on optimal estimation.

[0022] In step S1, multi-source data collection includes:

[0023] Collecting XCO2 secondary bias correction data sets observed and inverted by greenhouse gas monitoring satellites OCO-2 and OCO-3;

[0024] Collecting multi-source data of atmospheric CO and NO2 pollutant concentration data, vegetation index, and meteorological reanalysis data of the same source of emission; processing the spatial and temporal resolution into monthly values and 0.1° grid resolution, respectively, and normalizing them as input data for the XCO2 prediction model;

[0025] Collecting ground-based atmospheric CO2 concentration, anthropogenic emission inventory, and static geographic data.

[0026] Further, collecting XCO2 secondary bias correction data sets observed and inverted by greenhouse gas monitoring satellites OCO-2 and OCO-3 includes:

[0027] First, quality screening of satellite observed XCO2 data is performed, and observation values with quality flag QF of 0 are retained;

[0028] Then, a curve fitting function is applied to extract the annual growth and seasonal cycle deterministic trend in the XCO2 time series, and the XCO2 time series is decomposed into background XCO2 2,bg and short-term variation XCO2 2,stBy fitting XCO2 data within a moving window of 2° latitude, a spatiotemporal trend model of XCO2 is established, expressed as:

[0029]

[0030] XCO 2,st =XCO2-XCO 2,bg

[0031] In the formula, a0 and a1 are the intercept and slope of the polynomial function, respectively; t is the number in months; i is the number of harmonic functions; ω is the period coefficient, calculated as 2π / 12; β i and γ i All are the coefficients of variation of the harmonic function, δ filter The interannual variation contained in the residuals between observed and fitted values ​​of XCO2; parameters a0, a1, a2, β i ,γ i All were obtained by least squares fitting calculation;

[0032] Finally, the XCO2 observation data of OCO-2 and OCO-3 were analyzed separately. 2,st Gridding is performed at a resolution of 0.1°, monthly averages are calculated, and grid points with more than 5 valid points are retained; the average observations are calculated for the same grid in the same month.

[0033] In step S2, an XCO2 prediction model based on machine learning and multi-source satellite observations is constructed, including:

[0034] S201, utilizing XCO 2,st Using data as the target variable, an XCO2 prediction model was constructed based on machine learning algorithms. Multi-source data, including CO, NO2, vegetation index NDVI, precipitation PRE, surface solar radiation SSR, surface air pressure SurfP, temperature TEM, east-west wind speed U10 and north-south wind speed V10 at 10 meters height, and monthly time labels, were used as prediction variables. The model was trained on data samples consisting of each grid gi and each month ti. The XCO2 prediction model expression is:

[0035]

[0036] In the formula, XCO 2,st[gi,ti] XCO corresponding to grid gi and month ti 2,st Value, CO [gi,ti] The values ​​for CO and NO are given by grid gi and month ti, respectively. 2[gi,ti] The NO2 concentration values ​​corresponding to grid gi and month ti are NDVI. [gi,ti PRES represents the vegetation index corresponding to grid gi and month ti. [gi,ti]P is the precipitation value corresponding to the grid gi and month ti, SSR [gi,ti] SurfP is the surface solar radiation value corresponding to the grid gi and month ti, [gi,ti] TEM is the surface pressure value corresponding to the grid gi and month ti, [gi,ti] U10 is the temperature value corresponding to the grid gi and month ti, [gi,ti] and V10 [gi,ti are the 10-meter height east and north wind speed values corresponding to the grid gi and month ti, respectively, ti is the month, and gi is the grid.

[0037] In step S202, the reconstructed XCO2 spatiotemporal continuous data set is verified and precision evaluated through ten-fold cross-validation, feature importance analysis, and ground observation comparison method.

[0038] In step S3, the hourly anthropogenic CO2 concentration data is simulated using the anthropogenic CO2 concentration simulation results of the atmospheric transmission model, including:

[0039] In step S301, the static geographic data of the terrain and land use of the research area is extracted, and the initial and boundary meteorological data of the meteorological reanalysis data generation model is generated.

[0040] In step S302, the simulation area, time, and physical / chemical scheme are configured, the annual scale anthropogenic emission inventory is used as prior information, and the hourly anthropogenic CO2 concentration data is simulated.

[0041] In step S303, the sensitivity of different physical schemes or emission inventories is tested to evaluate the response of the model to the input.

[0042] In step S4, the anthropogenic CO2 emission estimation model based on optimal estimation and the results are verified, including:

[0043] In step S401, the atmospheric CO2 spatiotemporal continuous data set reconstructed by the machine learning algorithm is compared with the anthropogenic CO2 concentration data simulated by the atmospheric transmission model.

[0044] In step S402, the cost function is constructed, and the emission estimation result when the cost function is lowest is obtained.

[0045] In step S403, the initial conditions of the prior emission inventory are adjusted according to the preliminary simulation results and the feedback of the optimal estimation model.

[0046] In step S404, the atmospheric transmission simulation is performed again based on the updated emission inventory, and the optimized hourly anthropogenic CO2 concentration and corresponding emission estimation result are obtained.

[0047] In step S405, the dynamic change of the urban CO2 plume is monitored through the hourly simulation result data of the research period.

[0048] In step S402, the initial conditions of the prior emission inventory are adjusted, including: using an optimal estimation method, taking the prior emission inventory, wind speed, vegetation index, temperature and emissivity parameters in the surface parameters as state variables, to construct a cost function;

[0049] The minimum value of the cost function is solved by using an iterative method, and the error between the observation and the simulation value is reduced by adjusting the state variable step by step to obtain the emission estimation result when the cost function is lowest.

[0050] In step S403, the initial conditions of the prior emission inventory are adjusted, including: correcting the CO2 emission intensity and the timing distribution.

[0051] In step S405, the dynamic change of the city CO2 plume is monitored by studying the hourly simulation result data of the research period, including:

[0052] The CO2 high-value aggregation area is extracted by using local spatial autocorrelation analysis as a potential CO2 emission plume area; the shape and range of the plume are determined in combination with the geographical location of the city, the atmospheric CO and NO2 concentration; and the diffusion direction of the emission source is identified in combination with the meteorological data of wind speed and wind direction.

[0053] Another object of the present application is to provide an observation and simulation integrated city CO2 emission plume monitoring system, which implements the observation and simulation integrated city CO2 emission plume monitoring method.

[0054] A data collection module is used for multi-source data collection and preprocessing, and the multi-source data includes: OCO-2 and OCO-3 observation inversion XCO2 secondary bias correction data set; multi-source data of atmospheric CO and NO2 pollutant concentration data, vegetation index, and meteorological reanalysis data of the same source of emission; and ground observation atmospheric CO2 concentration, human emission inventory and static geographic data;

[0055] An XCO2 prediction model construction module constructs an XCO2 prediction model based on machine learning and multi-source satellite observation, and verifies and evaluates the accuracy of the reconstructed XCO2 spatiotemporal continuous data set;

[0056] A human CO2 concentration long-term simulation module simulates human CO2 concentration data every hour by using the human CO2 concentration long-term simulation of the atmospheric transmission model.

[0057] A human CO2 emission estimation module based on optimal estimation human CO2 emission estimation model and result verification; by constructing a cost function and obtaining the emission estimation result when the cost function is lowest, and monitoring the dynamic change of the city CO2 plume through long-term data.

[0058] In combination with all the above technical solutions, the present application has the beneficial effects of:

[0059] (1) Multi-source data fusion improves monitoring accuracy: By combining multi-source satellite observation data (including CO2 and other pollutant concentrations, land ecological remote sensing parameters, and meteorological data), the present application comprehensively considers different emission sources and environmental factors, providing more accurate CO2 emission plume monitoring. Compared with single data source methods, this solution can obtain more reliable results in a wider spatial range and complex environmental conditions.

[0060] (2) Introducing machine learning algorithms and atmospheric transport models to improve emission plume identification and enhance method performance: This technical solution not only applies satellite monitoring technology, but also introduces machine learning algorithms and atmospheric transport models into CO2 emission monitoring, which can handle multi-dimensional, nonlinear, and complex input variable relationships, automatically learn the relevance between different data, and track the long-distance transport process of CO2. This advantage enables the technical solution to provide stable and accurate emission prediction and plume identification even when there are gaps in satellite observation data, uneven data, or complex interactions between pollutants and CO2.

[0061] (3) Long-term dynamic monitoring capability: Compared with traditional static monitoring methods, the present application combines satellite observation, machine learning, and atmospheric transport models to achieve long-term dynamic monitoring of urban CO2 emission plumes. This enables the technical solution to continuously track the trends of urban CO2 emissions, supporting more flexible emission control and environmental management decisions.

[0062] (4) Automatic processing and identification: This solution has high automation, can automatically identify emission plumes based on multi-source data fusion and model simulation, reduces the need for manual intervention, improves monitoring efficiency, and is suitable for large-scale urban emission source monitoring and management.

[0063] (5) The present application helps identify emission plumes and estimate emissions by accurately monitoring urban CO2 emissions, providing scientific basis for urban environmental governance and carbon emission reduction; in different cities and regions, it realizes the precise positioning of high-emission areas and resource optimization, which helps to promote the development of cross-city and regional environmental monitoring and governance market, and creates significant commercial value.

[0064] (6) The application has originality in emission plume dynamic monitoring and fine positioning of emission sources, solves the problem that there is no mature technical means for high-precision monitoring of urban CO2 emissions at home and abroad, especially the lack of a scheme combining multi-source satellite observation and transmission simulation, and solves the long-standing problem of real-time and accurate monitoring of urban CO2 emission plume. Through optimization of data fusion and simulation method, effective integration of multi-source data and error minimization are realized, making it possible to monitor CO2 emission dynamics and locate emission sources, and breaking through the monitoring precision that cannot be achieved by traditional technology. Traditional CO2 emission monitoring relies on ground observation or single satellite data, which often underestimates the value of multi-source observation integration. The application overcomes this technical bias, fully explores the synergistic potential of satellite and model data through the integration of multi-source data and advanced optimization algorithms, and realizes higher precision monitoring of urban CO2 emissions, opening up a new way of thinking in the industry. BRIEF DESCRIPTION OF DRAWINGS

[0065] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure;

[0066] Figure 1 is a flow chart of the observation and simulation integrated urban CO2 emission plume monitoring method provided by the embodiments of the application;

[0067] Figure 2 is a schematic diagram of the observation and simulation integrated urban CO2 emission plume monitoring system provided by the embodiments of the application;

[0068] Figure 3 is the monitoring result of urban CO2 emission plume at 13:00 on January 21, 2019, local time in eastern China;

[0069] Figure 4 is the monitoring result of urban CO2 emission plume at 13:00 on January 22, 2019, local time in eastern China;

[0070] Figure 5 is the monitoring result of urban CO2 emission plume at 13:00 on January 23, 2019, local time in eastern China;

[0071] Figure 6 is the monitoring result of urban CO2 emission plume at 13:00 on January 24, 2019, local time in eastern China;

[0072] In the figure: 1, data collection module; 2, XCO2 prediction model construction module; 3, long-term simulation module of anthropogenic CO2 concentration; 4, anthropogenic CO2 emission estimation module. DETAILED DESCRIPTION

[0073] In order to make the above objectives, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application are described in detail below with reference to the drawings. In the following description, a large number of specific details are set forth in order to facilitate a thorough understanding of the present application. However, the present application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the spirit of the present application, so the present application is not limited to the specific implementations disclosed below.

[0074] The innovation of the present application is that the present application improves the identification accuracy of CO2 emission sources by fusing multi-source satellite data and atmospheric transmission model to construct an optimal estimation method. Combined with spatial autocorrelation analysis, the high concentration aggregation area is identified, the shape and diffusion path of the emission plume are determined, and high-precision and dynamic monitoring of urban CO2 emission is realized. The method provides important technical support for urban emission plume positioning and emission reduction policy making.

[0075] The present application provides an automatic identification technology for urban CO2 emission plume by integrating multi-source satellite observation data, fusing machine learning and atmospheric transmission model. Specifically,

[0076] (1) Integration and processing method of multi-source satellite observation data: CO2, NO2 and other pollutant concentration data are combined with meteorological and ecological parameters for CO2 emission monitoring, filling the gap of satellite observation data;

[0077] (2) Emission plume identification algorithm based on machine learning and atmospheric transmission model: the monitoring capability on high resolution and dynamic time scale, the automatic processing process avoiding manual intervention, including continuous acquisition and transmission model of satellite data, long-term operation and updating mechanism, ensuring accurate capture of emission trend, especially the improvement part of atmospheric transmission model in emission simulation underestimation problem, ensuring that the emission simulation result is closer to the actual observation technical scheme.

[0078] In the embodiment 1, as shown in the embodiment 1, the observation and simulation integrated urban CO2 emission plume monitoring method provided by the present application comprises: Figure 1

[0079] S1, multi-source data collection and preprocessing;

[0080] S101, collect OCO-2 and OCO-3 observation inversion XCO2 secondary bias correction data set; first, quality screening is performed on XCO2 data, and only high-quality observation values with quality flag (QF) of 0 are retained; then, a curve fitting function is applied to extract the annual growth and seasonal cycle of the XCO2 time series, and the XCO2 time series is decomposed into two parts: background XCO2 (XCO 2,bg ) and short-term XCO2 change (XCO 2,st ​), a temporal and spatial trend model of XCO2 is established by fitting XCO2 data in a moving window of every 2° latitude;

[0081]

[0082] XCO 2,st = XCO2- XCO 2,bg

[0083] where a0, a1 are the intercept and slope of the polynomial function respectively, t is the number of time unit in month, i is the number of harmonic function, ω is the period coefficient, the calculation formula is 2π / 12, β i and γ i are the variation coefficients of harmonic function, δ filter is the annual variation contained in the residual of XCO2 observation value and fitting value; parameters a0, a1, a2, β i , γ i are calculated by least square fitting.

[0084] Finally, XCO2 observation data (XCO 2,st ) is gridded at 0.1° resolution, and the monthly average value is calculated, only the grid points with effective point number greater than 5 are retained. In order to avoid statistical error caused by different sampling times of different satellites at the same position, OCO-2 and OCO-3 observations are gridded respectively, and then the observation average value of the same grid in the same month is calculated to ensure the consistency of spatial sampling of different satellites;

[0085] S102, collect the concentration data of atmospheric CO and NO2 pollutants of the same source, vegetation index, meteorological reanalysis data and other multi-source data, unify the temporal and spatial resolution to monthly value and 0.1° grid resolution; normalize these data sets respectively as the input data of XCO2 prediction model;

[0086] S103, collect ground observation of atmospheric CO2 concentration, human emission inventory and static geographic data.

[0087] S2, construct XCO2 prediction model based on machine learning and multi-source satellite observation;

[0088] S201, use XCO 2,stData as the target variable, based on machine learning algorithm to build XCO2 prediction model; Using multi-source data, including CO, NO2, NDVI, precipitation PRE, surface solar radiation SSR, surface pressure SurfP, air temperature TEM, 10 meters height of east-west wind speed U10 and north-south wind speed V10, monthly time label as prediction variable, the data sample of each grid gi, each month ti is trained; The expression of XCO2 prediction model is:

[0089]

[0090] In the formula, XCO 2,st[gi,ti] gi, ti corresponding XCO 2,st 2 value, CO [gi,ti] gi, ti corresponding CO concentration value, NO 2[gi,ti] gi, ti corresponding NO2 concentration value, NDVI [gi,ti] gi, ti corresponding vegetation index, PRES [gi,ti] gi, ti corresponding precipitation, SSR [gi,ti] gi, ti corresponding surface solar radiation value, SurfP [gi,ti] gi, ti corresponding surface pressure value, TEM [gi,ti] gi, ti corresponding air temperature value, U10 [gi,ti] And V10 [gi,ti] gi, ti corresponding 10 meters height of east-west wind speed, north-south wind speed, ti is month, gi is grid;

[0091] It can be understood that the application introduces CO as a prediction variable. For the XCO2 prediction model, introducing CO as a prediction variable can assist in identifying CO2 emission sources and plumes. In terms of technology: increase the input variable that can represent carbon emissions and short-term atmospheric transport changes, improve the accuracy of the prediction model; Make up for the lack of single observation data, enhance the robustness and stability of the model, especially important for CO2 data sparse area.

[0092] S202, the reconstructed XCO2 spatiotemporal continuous data set is verified and accuracy evaluated by ten-fold cross-validation, feature importance analysis and ground observation comparison method.

[0093] S3, using the artificial CO2 concentration simulation results of atmospheric transport model, simulate the hourly artificial CO2 concentration data;

[0094] The atmospheric transmission model used in the technical solution is a WRF-Chem model, which is a real-time online coupling simulation based on meteorology and chemistry, combines atmospheric dynamics and chemical reactions, and tracks and predicts the spatial distribution and temporal variation of greenhouse gases and various pollutants. The operation includes preparing input data such as meteorological data, emission data, and static geographic data; compiling and configuring the model, setting the running parameters, running the simulation, outputting and analyzing the results. The model has the ability to couple meteorology and chemistry in real time, can accurately simulate the dynamic processes of CO2 emission, transmission, diffusion, etc., and can provide detailed information on the spatial and temporal distribution, source contribution, diffusion path, etc. of CO2. Its advantages are high spatiotemporal resolution, strong coupling with meteorological conditions, flexible emission source configuration, and suitability for multi-scale CO2 research.

[0095] S301, extracting the terrain and land use static geographic data of the research area, and generating the initial and boundary meteorological data of the model from the meteorological reanalysis data;

[0096] S302, configuring the simulation area, time and physical / chemical scheme, using the annual anthropogenic emission inventory as prior information to simulate hourly anthropogenic CO2 concentration data;

[0097] S303, testing the sensitivity of different physical schemes or emission inventories to evaluate the response of the model to the input.

[0098] Select different physical schemes and emission inventories, set up multiple experimental groups in the same area and time range, and run the WRF-Chem model group by group; compare the output data of each group experiment, including the spatiotemporal distribution characteristics, calculation bias (such as root mean square error, absolute bias), analyze the sensitivity of different physical schemes and emission inventories, determine the optimal configuration scheme of the model, and explore the influence of input conditions on the simulation results of CO2. This sensitivity test can help optimize the model settings and improve the prediction accuracy.

[0099] S4, verifying the anthropogenic CO2 emission estimation model based on the optimal estimate and the results;

[0100] S401, comparative analysis of atmospheric CO2 spatiotemporal continuous data set reconstructed by machine learning algorithm and anthropogenic CO2 concentration data simulated by atmospheric transmission model;

[0101] Compare the results obtained by two different technical means, and compare the two groups of results through spatiotemporal matching, error statistics, and spatiotemporal consistency test;

[0102] S402, constructing a cost function and obtaining the emission estimation result when the cost function is at the lowest;

[0103] Optimization estimation method is adopted, and prior emission inventory, wind speed, vegetation index, and surface parameters (temperature and emissivity) are taken as state variables to construct the cost function. The cost function is constructed by taking the prior emission inventory, wind speed, vegetation index, and surface parameters (such as temperature and emissivity) as state variables, and minimizing the cost function to make the simulation data as close to the observed data as possible. The cost function usually includes two parts: observation term and prior term. The observation term measures the difference between the simulation value and the observed value, and the prior term reflects the deviation of the state variable from the known emission inventory.

[0104] Gradient descent and iterative solution: Gradient descent method or other iterative optimization algorithms are used to optimize the cost function by adjusting the state variables step by step. Each iteration calculates the gradient of the cost function and adjusts the state variables in the direction of reducing the error to minimize the difference between the simulation value and the observed value.

[0105] The iterative method is used to solve the minimum value of the cost function, and the state variables are adjusted step by step to reduce the error between the observation and the simulation value, and the emission estimation result is obtained when the cost function is at the lowest;

[0106]

[0107] In the formula, J(x) is the calculation result of the cost function, n is the number of state variables, T is the matrix transpose, y is the atmospheric CO2 spatiotemporal continuous data reconstructed by the machine learning algorithm, H is the anthropogenic CO2 concentration data simulated by the atmospheric transport model, R is the observation error covariance matrix, which is used to measure the observation error; λ i is the weight parameter, which is used to adjust the relative importance of different state variables in the cost function, x a,i is the prior estimate of the i-th state variable; B i is the prior error covariance matrix of the i-th state variable, which is used to measure the credibility of the prior information; x i is the i-th state variable to be estimated;

[0108] The first term of the formula is the observation bias term, which represents the difference between the observed data and the simulation value, and measures the deviation between the simulation result and the actual observed value. By weighting with the error covariance matrix R, the influence of data points with greater observation impact on the cost function is reduced. The second term is the prior constraint term, which represents the deviation between different state variables (such as emission inventory, wind speed, vegetation index, etc.) and their prior information, and is weighted by the prior error covariance matrix B to ensure that high-credibility prior information has a greater impact on the cost function. By adjusting the importance of different state variables through the weight parameter λ, the model is more flexible to adapt to specific application requirements.

[0109] The optimization algorithm using gradient descent is used to find the minimum value of J(x) by iteratively adjusting the state variables, and the process is as follows:

[0110] (1) Calculate the gradient of the cost function J(x) with respect to x

[0111]

[0112] (2) Update the state variable x by gradient descent method:

[0113]

[0114] where x k+1 is the state variable after the (k+1)th gradient descent update, x k is the state variable at the kth iteration, and a is the learning rate that controls the step size of the update in each iteration. By choosing an appropriate a and the number of iterations k, the value of J(x) can be gradually reduced.

[0115] (3) During the iteration process, when the change of the cost function approaches zero or is less than a pre-set threshold, stop the iteration and consider that J(x) has reached a minimum value, obtaining the optimal state variable x * .

[0116] During the process of gradually adjusting the state variable to reduce the error, the difference between the observed value and the simulated value gradually decreases. In each iteration, the state variable is gradually adjusted to reduce the error between the observed and simulated values by the following steps: (1) initialization: first initialize the state variable x with prior information x a , and calculate the initial value of the cost function J(x); (2) error feedback: in each iteration, adjust the state variable according to the error between the observed value y and the simulated value H, y-H; (3) weight adjustment: in the cost function, the weight λ reflects the importance of each state variable, and by dynamically adjusting different state variables through error feedback, the model can more accurately match the observed value. When the cost function reaches a minimum value, the optimal solution x * of the state variable x is taken as the modified initial condition of the model, and the model is run again.

[0117] S403, according to the preliminary simulation results, adjust the initial conditions of the prior emission inventory in combination with the feedback of the optimal estimation model; including correcting CO2 emission intensity, timing distribution, etc.

[0118] S404, based on the updated emission inventory, re-run the atmospheric transport simulation to obtain the optimized hourly anthropogenic CO2 concentration and corresponding emission estimation results.

[0119] The principle of the model is to input meteorological data, emission data and corresponding static geographic data to obtain artificial CO2 concentration data. After the updated emission inventory data is processed by the model in terms of time and spatial resolution, the emission estimation results are obtained.

[0120] S405, the dynamic changes of the city CO2 plume are monitored by studying the hourly simulation result data of the period.

[0121] The local spatial autocorrelation analysis is used to extract the CO2 high-value aggregation area as a potential CO2 emission plume area; the city geographic location, atmospheric CO and NO2 concentration are combined to further determine the shape and range of the plume; the wind speed and wind direction and other meteorological data are combined to identify the diffusion direction of the emission source; the dynamic changes of the city CO2 plume are monitored through long-term data.

[0122] Local spatial autocorrelation: through the k-neighbor weight method, a row-standardized spatial weight matrix is constructed to describe the adjacency relationship of each spatial unit; according to the attribute value (such as CO2 concentration) of each spatial unit and the values of the surrounding units, the local autocorrelation index is calculated, and the local autocorrelation index calculated by z-score or p-value is subjected to significance test, and the Moran index statistical result will be divided into five types of high-high, low-low, high-low, low-high and non-significant; the spatial units with significant high-high (i.e. spatial autocorrelation is positive and high value is significantly aggregated) are screened out, which represent potential local high-value aggregation areas.

[0123] Plume shape and range: after obtaining the preliminary high-high unit, spatial connectivity analysis is used to further determine the local high-high aggregation area.

[0124] (1) The eight-neighbor (up, down, left and right, and four diagonal neighbors) rule is used to define the connectivity between units;

[0125] (2) Based on the adjacency rule, the connected regions of high-high units are identified. That is, adjacent high-high units form a continuous aggregation area, while isolated high-high units are not considered as aggregation areas;

[0126] Identifying the diffusion direction of the emission source: for the obvious elliptical or rectangular high-high aggregation area, the elliptical fitting method is used to describe the major axis and minor axis; combined with meteorological wind field data, the diffusion direction of the emission source is determined.

[0127] (1) The coordinate points of each spatial unit in the high-high aggregation area are obtained, represented as a set of points (xi, yi);

[0128] (2) Set the principal axis direction as along the wind direction, assuming an angle θ, i.e. the major axis (principal axis) of the ellipse will be parallel to the wind direction: θ = arctan(v y / v x), where v x and v y are the components of the wind field in the x and y directions, respectively;

[0129] (3) Construct the covariance matrix of the spatial unit coordinate points:

[0130]

[0131] where, and are the variances in the x and y directions, respectively, and σ xy is the covariance in the x and y directions.

[0132] (4) Calculate the major and minor axes of the ellipse: Determine the lengths of the major and minor axes of the ellipse through the eigenvalues of the covariance matrix. Let the eigenvalues of the covariance matrix be λ1 and λ2 (λ1 > λ2), then the lengths of the semi-major and semi-minor axes are

[0133] where k is a scaling factor used to control the range of the ellipse coverage, usually k = 2 or 3 to cover 95% or 99% of the data points.

[0134] (5) According to the calculated a, b and θ, draw the ellipse to represent the diffusion shape and range of the high concentration aggregation area. Combined with the direction of the ellipse and the wind field direction, evaluate the diffusion trend and range of the high concentration area, identify the possible pollution source and diffusion path.

[0135] The present application can more accurately capture and analyze the spatio-temporal variation characteristics of CO2 concentration in the atmosphere by constructing a model integrating multi-source data and various technical means, and can accurately identify the pollution source and diffusion path. By combining satellite observation data, meteorological data (such as wind speed, wind direction) and ground monitoring data, etc. multi-source information, the present application can finely monitor the dynamic CO2 emissions in different regions.

[0136] At the same time, with the integration of machine learning algorithms and atmospheric transport models, the present application improves the resolution and inversion accuracy of CO2 emission sources, reduces the errors caused by single data source or model, and significantly enhances the ability to capture the spatio-temporal distribution of CO2 under complex meteorological conditions. The final effect is to provide more accurate and timely CO2 concentration monitoring and emission source identification, providing scientific support for developing emission reduction strategies and environmental management.

[0137] As shown in Figure 2 , the present application provides a kind of observation and simulation integrated urban CO2 emission plume monitoring system, and the system comprises:

[0138] A data collection module 1 is configured to collect and preprocess multi-source data, including OCO-2 and OCO-3 observation inversion XCO2 secondary bias correction data sets, multi-source data of atmospheric CO and NO2 pollutant concentration data, vegetation index, and meteorological reanalysis data of the same source, and multi-source data of ground observation atmospheric CO2 concentration, artificial emission inventory and static geographic data;

[0139] An XCO2 prediction model construction module 2 is configured to construct an XCO2 prediction model based on machine learning and multi-source satellite observation, and verify and evaluate the reconstructed XCO2 spatiotemporal continuous data set;

[0140] An artificial CO2 concentration long-term simulation module 3 is configured to simulate artificial CO2 concentration data per hour by using an atmospheric transmission model for artificial CO2 concentration long-term simulation.

[0141] An artificial CO2 emission estimation module 4 is configured to construct an artificial CO2 emission estimation model based on optimal estimation and verify the results, obtain the emission estimation results when the cost function is lowest, and monitor the dynamic change of the urban CO2 plume through long-term data.

[0142] As can be seen from the above embodiment, the present application provides an automatic identification technology for urban CO2 emission plume, which integrates multi-source satellite observation data, fuses machine learning and atmospheric transmission model, and develops the automatic identification technology for urban CO2 emission plume. Specifically,

[0143] The integration and processing method of multi-source satellite observation data: the concentration data of pollutants such as CO2 and NO2 are combined with meteorological and ecological parameters for CO2 emission monitoring, which fills the blank of satellite observation data;

[0144] The emission plume identification algorithm based on machine learning and atmospheric transmission model: the monitoring capability on high resolution and dynamic time scale, the automatic processing process avoiding artificial intervention, the continuous acquisition and transmission model of satellite data, the long-term running and updating mechanism of the transmission model, the accurate capture of the emission trend, and the improvement of the atmospheric transmission model in the underestimation of the emission amount, which ensures that the emission simulation result is closer to the actual observation.

[0145] Furthermore, the present application focuses on CO2 and pollutant concentration data from multi-source satellite observations, which can achieve large-scale monitoring at global and regional scales by combining machine learning algorithms and atmospheric transport models. Its advantage is wide coverage, which can provide CO2 emission trends across a large area, especially suitable for monitoring the long-term emission evolution of multiple large cities. CO2 emission monitoring based on ground observation networks and near-surface observations using unmanned aerial vehicle or aircraft platforms can more accurately obtain emission data for specific cities or industrial areas, suitable for detailed investigation of short-term or local emission sources. These alternatives can be selected and combined according to specific monitoring needs, regional characteristics, and cost constraints to form a more diverse and flexible urban CO2 emission monitoring system.

[0146] The results of urban CO2 emission plume monitoring in China's eastern region at 13:00 local time on January 21-25, 2019 are shown in FIG. 1. Figures 3-6

[0147] (1) High emission area: Large cities in eastern China usually have high traffic density and industrial activity during the noon period, and the emission plume forms a high CO2 concentration zone in the city and surrounding industrial areas.

[0148] (2) Plume diffusion: Comparing the diffusion paths of emission plumes on different dates, it can be seen that the areas with higher concentrations extend along the wind direction. The temperature in eastern China in winter is relatively low, and the wind speed is relatively small, so the diffusion speed of CO2 is slow, and the emission plume is more concentrated.

[0149] (3) Topographic influence: The geographical environment in eastern China is relatively complex, and there are differences in the diffusion of CO2 emission plumes in coastal cities and inland cities. For example, coastal cities such as Shanghai are affected by sea winds, and the emission plume will diffuse in a certain direction; while inland cities such as Taiyuan and Chifeng are closed due to the terrain, making it difficult for CO2 to diffuse effectively, and forming a high-concentration CO2 accumulation area in the atmosphere.

[0150] The above merely illustrates the preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any modification, equivalent replacement, and improvement made by those skilled in the art within the technical scope disclosed by the present application shall be encompassed within the protection scope of the present application.​

Claims

1. A method for monitoring urban CO2 emission plumes that integrates observation and simulation, characterized in that, The method includes: S1, Multi-source data collection and preprocessing; S2, Construct an XCO2 prediction model based on machine learning and multi-source satellite observations; S3 uses the anthropogenic CO2 concentration simulation results from the atmospheric transport model to simulate hourly anthropogenic CO2 concentration data; S4, Verify the anthropogenic CO2 emission estimation model and results based on the optimal estimate; In step S2, an XCO2 prediction model based on machine learning and multi-source satellite observations is constructed, including: S201, using XCO 2，st Data as the target variable, based on machine learning algorithm to build XCO2 prediction model; using multi-source data, including CO, NO2, vegetation index NDVI, precipitation PRE, surface solar radiation SSR, surface pressure SurfP, air temperature TEM, 10 meters height of east-west wind speed U10 and north-south wind speed V10, monthly value time label as the prediction variable, train the data sample composed of each grid gi and each month ti; XCO2 prediction model expression is: In the formula, XCO 2，st[gi，ti] XCO corresponding to grid gi and month ti 2,st Value, CO [gi，ti] The values ​​for CO and NO are given by grid gi and month ti, respectively. 2[gi，ti] The NO2 concentration values ​​corresponding to grid gi and month ti are NDVI. [gi，ti] PRES represents the vegetation index corresponding to grid gi and month ti. [gi，ti] For the precipitation corresponding to grid gi and month ti, SSR [gi，ti] SurfP represents the surface solar radiation value corresponding to grid gi and month ti. [gi，ti] TEM represents the surface air pressure value corresponding to grid gi and month ti. [gi，ti] U10 represents the temperature value corresponding to grid gi and month ti. [gi，ti] and U10 [gi，ti] These represent the east-west and north-south wind speeds at 10 meters above the grid (gi) and month (ti), respectively, where ti is the month and gi is the grid. S202 uses ten-fold cross-validation, feature importance analysis, and ground observation comparison methods to validate and evaluate the accuracy of the reconstructed XCO2 spatiotemporal continuous dataset.

2. The method for monitoring urban CO2 emission plumes that integrates observation and simulation according to claim 1, characterized in that, In step S1, multi-source data collection includes: Collect the second-order bias-corrected XCO2 dataset retrieved from observations inverted by greenhouse gas monitoring satellites OCO-2 and OCO-3; Multi-source data were collected, including atmospheric CO and NO2 pollutant concentration data, vegetation index, and meteorological reanalysis data from sources of emission. The spatiotemporal resolution was processed to monthly values ​​and 0.1° grid resolution, and then normalized to serve as input data for the XCO2 prediction model. Collect ground-based observations of atmospheric CO2 concentrations, anthropogenic emission inventories, and static geographic data.

3. The method for monitoring urban CO2 emission plumes that integrates observation and simulation according to claim 2, characterized in that, Collect the second-order bias-corrected XCO2 dataset retrieved from observations inverted by the greenhouse gas monitoring satellites OCO-2 and OCO-3, including: First, the XCO2 data observed by satellite were screened for quality, and observations with a quality flag QF of 0 were retained. Then, a curve fitting function is applied to extract the deterministic trends of annual growth and seasonal cycles in the XCO2 time series, decomposing the XCO2 time series into background XCO2. 2,bg and short-term changes XCO 2,st By fitting XCO2 data within a moving window of 2° latitude, a spatiotemporal trend model of XCO2 is established, expressed as: XCO 2，st =XCO2-XCO 2，bg In the formula, a0 and a1 are the intercept and slope of the polynomial function, respectively; t is the number in months; i is the number of harmonic functions; ω is the period coefficient, calculated as 2π / 12; β i and γ i All are the coefficients of variation of the harmonic function, δ filter The interannual variation contained in the residuals between observed and fitted values ​​of XCO2; parameters a0, a1, a2, β i γ i All were obtained by least squares fitting calculation; Finally, the XCO2 observation data of OCO-2 and OCO-3 were analyzed separately. 2,st Gridding is performed at a resolution of 0.1°, monthly averages are calculated, and grid points with more than 5 valid points are retained; the average observations are calculated for the same grid in the same month.

4. The method for monitoring urban CO2 emission plumes that integrates observation and simulation according to claim 1, characterized in that, In step S3, using the anthropogenic CO2 concentration simulation results from the atmospheric transport model, hourly anthropogenic CO2 concentration data are simulated, including: S301, extract static geographic data of topography and land use in the study area, and generate initial and boundary meteorological data for the model using meteorological reanalysis data; S302, configured with simulation area, time and physical / chemical scheme, uses an annual scale anthropogenic emission inventory as prior information to simulate hourly anthropogenic CO2 concentration data; S303 tests the sensitivity of different physical scenarios or emission inventories to evaluate the model’s response to inputs.

5. The method for monitoring urban CO2 emission plumes that integrates observation and simulation according to claim 1, characterized in that, In step S4, the anthropogenic CO2 emission estimation model and results based on the optimal estimate are verified, including: S401, Comparative analysis of the spatiotemporal continuous dataset of atmospheric CO2 reconstructed by machine learning algorithm and the anthropogenic CO2 concentration data simulated by atmospheric transport model; S402, Construct the cost function and obtain the emission estimation result when the cost function is minimized; S403. Based on the preliminary simulation results and the feedback from the optimal estimation model, the initial conditions of the prior emission inventory are adjusted. S404, based on the updated emission inventory, atmospheric transport simulation was performed again to obtain optimized hourly anthropogenic CO2 concentration and corresponding emission estimates; S405 monitors the dynamic changes of urban CO2 plumes using hourly simulation data during the study period.

6. The method for monitoring urban CO2 emission plumes that integrates observation and simulation according to claim 5, characterized in that, In step S402, the initial conditions of the prior emission inventory are adjusted, including: using an optimization estimation method, a cost function is constructed with the prior emission inventory, wind speed, vegetation index, and temperature and emissivity parameters in the surface parameters as state variables; The minimum cost function is found by using an iterative method. The error between the observed and simulated values ​​is reduced by gradually adjusting the state variables, and the emission estimation result when the cost function is at its minimum is obtained.

7. The method for monitoring urban CO2 emission plumes that integrates observation and simulation according to claim 5, characterized in that, In step S403, the initial conditions of the prior emission inventory are adjusted, including: correcting the CO2 emission intensity and time series distribution.

8. The method for monitoring urban CO2 emission plumes that integrates observation and simulation according to claim 5, characterized in that, In step S405, the dynamic changes of the urban CO2 plume are monitored using hourly simulation results data during the study period, including: Local spatial autocorrelation analysis was used to extract high-value CO2 accumulation areas as potential CO2 emission plume regions; the shape and range of the plume were determined by combining urban geographical location and atmospheric CO and NO2 concentrations; and the diffusion direction of emission sources was identified by combining meteorological data of wind speed and wind direction.

9. A city CO2 emission plume monitoring system integrating observation and simulation, characterized in that, This system is implemented using the integrated observation and simulation method for monitoring urban CO2 emission plumes as described in any one of claims 1-8. The system comprises: The data collection module (1) is used for multi-source data collection and preprocessing. The multi-source data includes: XCO2 second-order bias correction dataset retrieved from OCO-2 and OCO-3 observations; multi-source data of atmospheric CO and NO2 pollutant concentrations, vegetation index, and meteorological reanalysis data from the same emission source; and atmospheric CO2 concentrations, anthropogenic emission inventories, and static geographic data observed on the ground. XCO2 prediction model construction module (2) constructs an XCO2 prediction model based on machine learning and multi-source satellite observation, and verifies and evaluates the accuracy of the reconstructed XCO2 spatiotemporal continuous dataset. The long-term simulation module for anthropogenic CO2 concentration (3) uses an atmospheric transport model to simulate the long-term anthropogenic CO2 concentration and generate hourly anthropogenic CO2 concentration data. The anthropogenic CO2 emission estimation module (4) is based on the optimal estimation model for anthropogenic CO2 emission estimation and result verification; by constructing a cost function and obtaining the emission estimation result when the cost function is at its lowest, and by monitoring the dynamic changes of urban CO2 plume through long-term data.

Images