An atmospheric XCO based on multi-source remote sensing data 2 Spatialization Methods

Abstract

The present invention discloses a method for spatialization of atmospheric XCO2 based on multi-source remote sensing data, including: fusing atmospheric XCO2 remote sensing data of multiple greenhouse gas observation satellites; obtaining spatial independent variable data including atmospheric NO2 data; performing Gaussian filtering and smoothing processing on atmospheric NO2 data, selecting the smoothed result of atmospheric NO2 data with the highest correlation coefficient as the input variable of an atmospheric XCO2 remote sensing estimation model; using atmospheric XCO2 remote sensing data as dependent variables, atmospheric NO2 data, vegetation index, elevation data and temperature data as independent variables, constructing a geographically weighted regression model, obtaining an atmospheric XCO2 remote sensing estimation model; and obtaining a spatial continuous distribution of atmospheric XCO2. The present invention makes up for the defect of a large number of blank areas in satellite remote sensing atmospheric XCO2 data, reflects the spatial continuous distribution pattern of atmospheric XCO2, and provides technical support for atmospheric XCO2 monitoring.

Description

Technical Field

[0001] The invention belongs to the technical field of atmospheric environment monitoring, and in particular relates to an atmospheric XCO2 spatialization method of multi-source remote sensing data. Background Art

[0002] Since the industrial age, the CO2 concentration in the atmosphere has increased dramatically due to the massive use of fossil fuels and the reduction of natural vegetation such as forests, leading to global warming, which in turn has caused extreme weather events such as heat waves, droughts, tidal floods, etc., which seriously endanger the human living environment. Monitoring the spatiotemporal changes in atmospheric CO2 concentration can effectively evaluate the effectiveness of carbon emission reduction and provide important auxiliary decision-making information for promoting the dual carbon goals.

[0003] There are two main ways to monitor atmospheric CO2: ground observation and satellite observation. Ground stations can obtain accurate point-scale CO2 observation data, but due to the high cost of establishing and maintaining a monitoring network, the number of monitoring stations is very limited and the distribution is severely uneven, making it impossible to obtain information on the spatiotemporal changes in atmospheric CO2 concentrations over a large range. Satellite observations have the advantages of long time series and wide spatial range, which make up for the shortcomings of ground observations and become an important means of monitoring global and regional atmospheric CO2 concentrations. However, due to the narrow width of carbon satellites and the influence of cloud cover, there are a large number of monitoring blank areas in the satellite-inverted atmospheric carbon dioxide column average dry air mixing ratio (XCO2) data, which seriously affects the practical application of the data. At present, the methods for obtaining spatially continuous and fully covered XCO2 based on satellite observation data mainly include spatial interpolation and statistical model methods, among which the spatial interpolation method is more commonly used. Ordinary Kriging is the most classic spatial interpolation method and is widely used. However, this method only relies on the XCO2 observations themselves for interpolation, which is greatly affected by the location and distribution characteristics of the station. Co-kriging uses land use, land cover information and emission estimates as auxiliary information to interpolate the XCO2 distribution, which has improved accuracy compared with ordinary Kriging. Spatiotemporal Kriging takes into account the spatial and temporal correlation of the observation data at the same time, improving the spatiotemporal continuity of the interpolation results, but the amount of calculation is large. The spatial interpolation method is limited by the sparseness of satellite XCO2 observations. The spatial distribution of the interpolation results often presents a strip distribution characteristic related to the satellite orbit. In addition to the limitation of the amount of calculation, it can usually only output a lower spatial resolution (mostly between 0.25 and 1°) in large-scale study areas. In order to improve the accuracy and resolution of generated data, a few studies have used statistical models to generate spatially continuous XCO2 data, mainly using satellite-observed XCO2 as the dependent variable and environmental factors such as vegetation, meteorology, and topography as independent variables, and using linear regression models, XGBoost, extreme random trees and other statistical models to build atmospheric XCO2 remote sensing estimation models. However, existing studies only consider natural environmental factors such as vegetation, meteorology, and topography in the selection of spatial independent variables, and do not consider pollutants such as NO2 that are strongly correlated with anthropogenic CO2 emissions and factors such as the intensity of human activities.

[0004] The invention with patent publication number CN116796534A discloses a method and system for estimating greenhouse gas emissions from point sources based on satellite detection, but the invention is only used for estimating carbon dioxide emissions from point sources. The invention with patent publication number CN116167202A discloses a satellite remote sensing carbon dioxide emission inversion method based on multi-source data, which is also a carbon dioxide emission inversion method for point sources (each factory).

[0005] The invention with patent publication number CN116189796A discloses a satellite-borne shortwave infrared CO2 column concentration estimation method based on machine learning, which performs feature screening on 9 weak_CO2 bands and 6 O2 bands data, NDVI normalized vegetation index, SR surface reflectance data, DEM elevation terrain data, ERA5 atmospheric data, AOD aerosol data, and TCCON station observation data, and retains the first 31 features selected according to importance; performs correlation analysis on the first 31 selected features through heat maps to find out the features with strong and weak correlation with CO2 column concentration; merges the features with strong and weak correlation with CO2 column concentration, inputs them into five integrated learning regression models, and uses the model with the highest prediction accuracy to estimate the shortwave infrared CO2 column concentration. However, the invention only uses the pixel values ​​of the OCO-2 satellite band data to estimate the atmospheric XCO2. Since greenhouse gas satellites such as OCO-2 can only observe the earth with a narrow width and discrete footprint, and the influence of cloud cover leads to a large number of blank monitoring areas, their estimation results can only cover a very small spatial range and cannot well characterize the spatial detail differences of CO2.

[0006] The invention with patent publication number CN115630870A discloses a method for analyzing the spatiotemporal characteristics and influencing factors of atmospheric CO2 in geological carbon storage areas. The best combination of explanatory variables is selected as independent variables, and the atmospheric CO2 concentration in the geological carbon storage area is used as the dependent variable to construct a geographically weighted regression model. However, the purpose of this invention is to analyze the impact of underground CO2 storage on the surface of the storage area under the influence of influencing factors. Therefore, the leakage influencing factors selected include the fault and crack distribution and wellbore distribution data in the geological carbon storage area. Other influencing factors include surface topography undulation, wind speed and direction, and vegetation coverage data. It cannot be promoted and applied to the spatialization scenario of atmospheric XCO2 in conventional areas.

[0007] The invention with patent publication number CN114330146B discloses a satellite gas data completion method and system, which respectively uses two satellites to obtain the first gas information (CO2 data) and the second gas information (NO2 data), establishes a correlation between the first gas information and the second gas information of the same geographic coordinates to train a machine learning model through a data set to complete the missing first gas information, and then applies a decision tree module to obtain a satellite data model of the completed first gas information from the second gas information. Similarly, the invention only uses one satellite to obtain CO2 data, and fails to make full use of the existing satellite atmospheric XCO2 data of multiple greenhouse gases to improve the spatial coverage and observation period of satellite data. In addition, the patent only directly completes based on the relationship between NO2 and CO2, without considering that the two greenhouse gases have very different life cycles in the atmosphere, resulting in a large difference in their spatial distribution patterns, and the difference in surface characteristics also has a non-negligible impact on the correlation between the two gases. Therefore, there is a large uncertainty in directly and simply using the relationship between NO2 and CO2 for interpolation, and the reliability of the completed data is insufficient. Summary of the invention

[0008] Technical problems to be solved: Among the existing atmospheric CO2 observation methods, ground-based observations cannot obtain large-scale observation data due to the very limited number of observation stations and their severely uneven distribution. Satellite observations have a large number of monitoring blank areas in the XCO2 data due to the narrow width of carbon satellites and the influence of cloud cover. Although satellite observations have a larger observation data range than ground-based observations, they still cannot obtain spatially continuous atmospheric XCO2 distribution. Aiming at the technical problems of low spatialization precision of atmospheric XCO2 in a large area, large amount of calculation, failure to effectively consider pollutants such as NO2 and human activity intensity factors that are highly correlated with anthropogenic CO2 emissions, the present invention discloses a spatialization method of atmospheric XCO2 of multi-source remote sensing data, data fusion is performed for atmospheric XCO2 remote sensing data of multi-source satellites such as GOSAT, OCO-2, and OCO-3, and an atmospheric XCO2 remote sensing estimation model is constructed based on atmospheric XCO2 fusion data and spatial independent variables such as vegetation index, temperature data, terrain data (elevation data), and atmospheric NO2 data. Compared with atmospheric XCO2 remote sensing data using a single satellite, the fusion of multi-source atmospheric XCO2 remote sensing data improves the spatial and temporal coverage of XCO2 observations, introduces atmospheric NO2 concentration information to characterize the impact of anthropogenic CO2 emissions, and further improves the model accuracy of the atmospheric XCO2 remote sensing estimation model. The present invention makes up for the defect that there are a large number of monitoring blank areas in atmospheric XCO2 data, reflects the spatial continuous distribution pattern of atmospheric XCO2, and provides technical support for atmospheric XCO2 monitoring.

[0009] Technical solution:

[0010] A method for spatializing atmospheric XCO2 based on multi-source remote sensing data, the method comprising the following steps:

[0011] Step A, obtaining atmospheric XCO2 remote sensing data observed by multiple satellites, and preprocessing the obtained multi-source satellite observation data;

[0012] Step B, converting the atmospheric XCO2 remote sensing data of multi-source satellite observation data into a unified CO2 profile, interpolating or expanding the number of CO2 profile layers of each satellite to a unified number of CO2 profile layers based on the atmospheric vertical pressure layer, and then converting the XCO2 measurement value of the satellite transit time into the XCO2 measurement value of the unified reference time, and finally performing spatiotemporal normalization processing on the multi-source satellite data, unifying the data with different spatiotemporal resolutions to the same spatiotemporal scale, and obtaining a fused data set;

[0013] Step C, obtaining spatial independent variable data including atmospheric NO2 data, vegetation index, elevation data and temperature data, and resampling all spatial independent variable data to the same spatial resolution;

[0014] Step D, performing Gaussian filtering and smoothing processing on the resampled atmospheric NO2 data, calculating the correlation coefficients between the atmospheric NO2 data extracted with different smoothing effects and the atmospheric XCO2 remote sensing data, and selecting the smoothing result of the atmospheric NO2 data with the highest correlation coefficient as the input variable of the atmospheric XCO2 remote sensing estimation model;

[0015] Step E, extracting the spatial independent variable value corresponding to the atmospheric XCO2 remote sensing data by performing spatiotemporal matching through longitude and latitude and time; constructing and training a geographically weighted regression model with atmospheric XCO2 remote sensing data as the dependent variable, atmospheric NO2 data, vegetation index, elevation data and temperature data as independent variables, calculating the regression coefficient of the point according to the weighted least squares method, determining the optimal bandwidth and weight of the geographically weighted regression model by the adaptive method, solving the regression coefficient of each independent variable, and obtaining the atmospheric XCO2 remote sensing estimation model;

[0016] Step F, applying the atmospheric XCO2 remote sensing estimation model to the spatially continuous independent variable data, calculating the atmospheric XCO2 continuous data, and obtaining the atmospheric XCO2 spatially continuous distribution.

[0017] Furthermore, in step A, the atmospheric XCO2 remote sensing data from multiple greenhouse gas observation satellites is obtained, and the process of preprocessing the obtained multi-source satellite atmospheric XCO2 data includes the following sub-steps:

[0018] A1, respectively obtain the atmospheric XCO2 data of GOSAT satellite, the atmospheric XCO2 data of OCO-2 satellite and the atmospheric XCO2 data of OCO-3 satellite. The data version of the atmospheric XCO2 data of GOSAT satellite is ACOS_L2_Lite_FP.9r, the data version of the atmospheric XCO2 data of OCO-2 satellite is OCO2_L2_Lite_FP.11r, and the data version of the atmospheric XCO2 data of OCO-3 satellite is OCO3_L2_Lite_FP.10.4r;

[0019] A2, obtain high-quality atmospheric XCO2 remote sensing data through the quality control mark screening provided by the data product. The quality control mark is: xco 2_ quality_flag=0.

[0020] Furthermore, in step B, the process of obtaining the fused data set includes the following sub-steps:

[0021] B1, taking the CO2 profile of Carbon Tracker model data as the benchmark, the atmospheric XCO2 remote sensing data inverted by GOSAT ACOS satellite, OCO-2 satellite and OCO-3 satellite are converted into a unified CO2 prior profile using the following formula:

[0022]

[0023] Where XCO 2adj,t is the corrected XCO2 value of the satellite to be converted at observation time t, XCO 2raw,t is the XCO2 value of the original satellite to be converted at observation time t, h is the pressure weighting function, I is the unit matrix, A is the average kernel function calculated by the inversion algorithm, X CT,t is the CO2 profile distribution of the Carbon Tracker model corresponding to satellite observation time t, X a,t is the prior CO2 profile distribution of the satellite to be converted;

[0024] B2, based on the atmospheric vertical pressure layer, the CO2 profile layers of the GOSAT ACOS satellite, OCO-2 satellite and OCO-3 satellite are interpolated or expanded to the CO2 profile layers of the Carbon Tracker model;

[0025] B3, linearly interpolate the 8-period profile concentration distribution of the CT model data to the transit time of the satellite data, introduce a time shift correction coefficient based on the daily variation simulated by the Carbon Tracker model; convert the CO2 measurement value at the satellite transit time t Converted to CO2 measurement at reference time 13:30:

[0026]

[0027] Where rt is the reference time 13:30, XCO 2con,rt is the XCO2 value converted according to the satellite reference time rt, h t 、h rt are the CO2 profiles of the CT model corresponding to time t and rt, respectively;

[0028] B4, for each type of corrected satellite data, the data with a difference of more than 3 standard deviations from the mean are filtered out, and the atmospheric XCO2 remote sensing data within the synthesis period are integrated and averaged within a radius of 10 km, with 1 month or 1 year as the synthesis period, and the data with different spatiotemporal resolutions are unified to the same spatiotemporal scale to obtain a fused data set.

[0029] Furthermore, in step C, the atmospheric NO2 product of the Sentinel-5P satellite tropospheric observation instrument TROPOMI is used to extract the daily data of atmospheric NO2 concentration with a spatial resolution of 5 km; the 16-day synthetic data of the MODIS / MOD13A3 product is converted into a monthly synthetic normalized vegetation index NDVI with a resolution of 1 km using time weighting; SRTM / DEM is used to obtain elevation data, the data version of the elevation data is SRTMGL1_v003, and the spatial resolution is 30 m; the temperature data is obtained using the ERA5-Land monthly synthetic reanalysis data, and the spatial resolution of the temperature data is 10 km.

[0030] Furthermore, in step C, all spatial independent variable data are uniformly resampled to a spatial resolution of 10 km.

[0031] Furthermore, in step D, the resampled atmospheric NO2 data is subjected to Gaussian filtering and smoothing, and the process of calculating the correlation coefficients between the atmospheric NO2 data extracted with different smoothing effects and the atmospheric XCO2 remote sensing data includes the following sub-steps:

[0032] D1, Gaussian filtering and smoothing are performed on the acquired atmospheric NO2 data. The weights of each pixel in the window of the Gaussian smoothing filter are Gaussian distributed:

[0033]

[0034] In the formula, g(μ,v) is the weight corresponding to the observation point (μ,v); μ and v are the longitude and latitude coordinates of the observation point respectively; σ is the standard deviation. g(μ,v) is the weight corresponding to the position (μ,v); the center point in the window of the Gaussian smoothing filter is the origin, and the weights of other points are calculated according to their positions (μ,v) based on the Gaussian function, and the weighted average F(0,0) of the center point of the window is calculated:

[0035]

[0036] Where R is the window radius of the Gaussian smoothing filter, f(μ,v) is the original pixel value at position (μ,v), and g(μ,v) is the weight corresponding to position (μ,v);

[0037] D2, use Gaussian smoothing filters with different window sizes to process atmospheric NO2 data, use the longitude and latitude of the XCO2 observation point to extract the corresponding atmospheric NO2, calculate the correlation coefficient between atmospheric NO2 and XCO2 extracted with different smoothing effects, and select the smoothing result of atmospheric NO2 data with the highest correlation coefficient as the input variable of the atmospheric XCO2 remote sensing estimation model.

[0038] Furthermore, in step E, the process of obtaining the atmospheric XCO2 remote sensing estimation model includes the following sub-steps:

[0039] By matching latitude, longitude and time, the spatial independent variable values ​​corresponding to the atmospheric XCO2 observation data are extracted. With atmospheric XCO2 remote sensing data as the dependent variable, atmospheric NO2 data, vegetation index, temperature data, and elevation data as independent variables, a geographically weighted regression model is constructed:

[0040] y(μ,ν)=β0(μ,ν)+β1(μ,ν)x1+β2(μ,ν)x2+β3(μ,ν)x3+β4(μ,ν)x4+ε

[0041] Where y(μ,v) represents the atmospheric XCO2 observation value at the observation point (μ,v); μ and v are the longitude and latitude coordinates of the observation point, respectively; β0(μ,v) is the intercept of the equation; β1(μ,v), β2(μ,v), β3(μ,v) and β4(μ,v) are the regression coefficients of the corresponding independent variables; x1, x2, x3 and x4 are the smoothing results of the vegetation index, temperature data, elevation data and atmospheric NO2 data at the observation point, respectively; ε is the random error term;

[0042] According to the weighted least squares method, the regression coefficient of the point is obtained, and the estimated value of the regression coefficient parameter β^(μ,v) is expressed as:

[0043] β ^ (μ,v)=[X T W(μ,ν)X] -1 X T W(μ,ν)Y

[0044] Where W(μ,v) is the spatial weight diagonal matrix, X is the independent variable matrix, and Y is the dependent variable vector;

[0045] The cross-validation method was used to calculate the bandwidth, the Gaussian function was used to calculate the spatial weight, and the adaptive method was used to determine the optimal bandwidth and weight of the geographically weighted regression model. The regression coefficient of the independent variable was solved to obtain the atmospheric XCO2 remote sensing estimation model.

[0046] Beneficial effects:

[0047] First, the atmospheric XCO2 spatialization method based on multi-source remote sensing data of the present invention can generate spatiotemporally continuous XCO2 data based on discrete satellite remote sensing XCO2 data, providing important auxiliary decision-making information for evaluating the effectiveness of carbon emission reduction and promoting carbon peak and carbon neutrality goals.

[0048] Second, the atmospheric XCO2 spatialization method based on multi-source remote sensing data of the present invention aims to solve the problem that previous studies only used single satellite observation data, whose data volume was limited and unevenly distributed. The present invention adjusts the three satellite data based on the CO2 broad line and time of the Carbon Tracker model data, and fuses the adjusted three satellite data using the ensemble averaging method, thereby improving the coverage of the observation data.

[0049] Third, the atmospheric XCO2 spatialization method based on multi-source remote sensing data of the present invention uses a geographically weighted regression algorithm to construct a remote sensing estimation model between atmospheric XCO2 and related variables. Compared with spatial interpolation methods and machine learning methods whose parameters are constant in the global space, the parameters of geographically weighted regression are variable in different spatial positions, which can well describe the relationship between independent variables and dependent variables as the space changes, and effectively improve the characterization of the spatial heterogeneity of atmospheric XCO2.

[0050] Fourth, the present invention's atmospheric XCO2 spatialization method based on multi-source remote sensing data takes into account the relationship between atmospheric CO2 remote sensing data and atmospheric NO2 data, and explores the correlation between different smoothing effects of atmospheric NO2 data and atmospheric CO2 remote sensing data, and uses the smoothing result of the most correlated atmospheric NO2 data as one of the input variables of the atmospheric XCO2 remote sensing estimation model, thereby effectively improving the model accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a flow chart of the method for spatializing atmospheric XCO2 based on multi-source remote sensing data of the present invention;

[0052] Figure 2 This is the distribution map of GOSAT satellite observation data in June 2020;

[0053] Figure 3 This is the distribution map of OCO-2 satellite observation data in June 2020;

[0054] Figure 4 This is the distribution map of OCO-3 satellite observation data in June 2020;

[0055] Figure 5 This is the fusion data distribution map for June 2020;

[0056] Figure 6 It is a scatter plot of XCO2 concentration estimated by the method of the present invention and XCO2 concentration observed by satellite;

[0057] Figure 7 This is the spatial distribution map of atmospheric XCO2 in a certain area. DETAILED DESCRIPTION

[0058] The following examples will enable those skilled in the art to more fully understand the present invention, but are not intended to limit the present invention in any way.

[0059] Figure 1 This is a flow chart of the atmospheric XCO2 spatialization method based on multi-source remote sensing data of the present invention. Figure 1 , the atmospheric XCO2 spatialization method comprises the following steps:

[0060] Step A, acquiring atmospheric XCO2 remote sensing data from multiple greenhouse gas observation satellites, and preprocessing the acquired multi-source satellite atmospheric XCO2 data;

[0061] Step B: convert the atmospheric XCO2 remote sensing data of multi-source satellites into a unified CO2 profile, interpolate or expand the number of CO2 profile layers of each satellite to a unified number of CO2 profile layers based on the atmospheric vertical pressure layer, and then convert the CO2 measurement value at the satellite transit time into the XCO2 measurement value at a unified reference time. Finally, perform spatiotemporal normalization on the multi-source satellite data, unify the data with different spatiotemporal resolutions into the same spatiotemporal scale, and obtain a fused data set.

[0062] Step C, obtaining spatial independent variable data including atmospheric NO2 data, vegetation index, elevation data and temperature data, and resampling all spatial independent variable data to the same spatial resolution;

[0063] Step D, performing Gaussian filtering and smoothing processing on the resampled atmospheric NO2 data, calculating the correlation coefficients between the atmospheric NO2 data extracted with different smoothing effects and the atmospheric XCO2 remote sensing data, and selecting the smoothing result of the atmospheric NO2 data with the highest correlation coefficient as the input variable of the atmospheric XCO2 remote sensing estimation model;

[0064] Step E, extracting the spatial independent variable value corresponding to the atmospheric XCO2 remote sensing data by performing spatiotemporal matching through longitude and latitude and time; constructing and training a geographically weighted regression model with atmospheric XCO2 remote sensing data as the dependent variable, atmospheric NO2 data, vegetation index, elevation data and temperature data as independent variables, calculating the regression coefficient of the point according to the weighted least squares method, determining the optimal bandwidth and weight of the geographically weighted regression model by the adaptive method, solving the regression coefficient of each independent variable, and obtaining the atmospheric XCO2 remote sensing estimation model;

[0065] Step F, applying the atmospheric XCO2 remote sensing estimation model to the spatially continuous independent variable data, calculating the atmospheric XCO2 continuous data, and obtaining the atmospheric XCO2 spatially continuous distribution.

[0066] The purpose of the present invention is to propose a method for spatialization of atmospheric XCO2 based on multi-source fusion XCO2 remote sensing data and atmospheric NO2 data. The method first fuses GOSAT, OCO-2, and OCO-3 satellite data, and based on multi-source fusion XCO2 remote sensing data, comprehensively considers terrain factors, meteorological factors, vegetation factors related to atmospheric XCO2, and NO2 pollutant gas with strong correlation with anthropogenic CO2 emissions, and uses geographically weighted regression method to construct a remote sensing estimation model to generate spatially continuous atmospheric XCO2.

[0067] 1) Preprocessing of atmospheric XCO2 remote sensing data

[0068] The GOSAT satellite observation data version used in this invention is ACOS_L2_Lite_FP.9r, the OCO-2 satellite observation data version used is OCO2_L2_Lite_FP.11r, and the OCO-3 satellite observation data version used is OCO3_L2_Lite_FP.10.4r. In order to ensure the reliability of the data, high-quality XCO2 data is obtained by screening the quality control information provided by the data product. The quality control mark is:

[0069] xco 2_ quality_flag=0(1).

[0070] 2) Multi-source atmospheric XCO2 remote sensing data fusion

[0071] When fusing multi-source satellite data, it is necessary to consider the differences caused by the different prior CO2 profiles used by each inversion algorithm. This paper uses the CO2 profile of the Carbon Tracker (CT) model data as a benchmark to convert the XCO2 data inverted by GOSAT ACOS, OCO-2 and OCO-3 satellites into a unified CO2 prior profile:

[0072]

[0073] Where XCO 2adj,t is the corrected XCO2 value at satellite observation time t, XCO 2raw,t is the XCO2 value at the original satellite observation time t, h is the pressure weighting function, I is the unit matrix, A is the average kernel function calculated by the inversion algorithm, X CT,t is the CO2 profile distribution of the CT model corresponding to the satellite observation time t, X a,t is the satellite's prior CO2 profile distribution.

[0074] Before the correction process, in order to solve the difference between the number of CO2 profile layers of each satellite data and the CT model, the number of CO2 profile layers of each satellite is interpolated or expanded to the number of CO2 profile layers of the CT model based on the vertical atmospheric pressure layer. GOSAT ACOS, OCO-2 and OCO-3 data contain 20 layers of CO2 profile concentration, which are interpolated or expanded to the 35 layers of CO2 profile concentration of the CT model according to the horizontal distribution of atmospheric pressure.

[0075] Since the transit times of GOSAT and OCO-2 are different (13:00 and 13:36 respectively), OCO-3 is carried on the International Space Station for observation from morning to night. In order to reduce the impact of the daily variation of atmospheric CO2 concentration, a time shift correction coefficient is introduced according to the daily variation simulated by the CT model. The time resolution of CT data is 3 hours. Before the time shift correction, the 8-period profile concentration distribution of the CT model data is linearly interpolated to the transit time of the satellite data. The CO2 measurement value at the satellite transit time is converted to the CO2 measurement value at the reference time 13:30 (UTC):

[0076]

[0077] Where rt is the reference time 13:30 (UTC), XCO 2con,rt is the XCO2 value converted according to the satellite reference time rt, h t 、h rt and are the CO2 profiles of the CT model at time t and rt respectively.

[0078] Due to the different spectral resolutions and observation revisit periods of different satellite sensors, the spatiotemporal distribution of their inversion data is extremely uneven. In order to make the fused data well used for spatial continuous distribution modeling, the multi-source satellite data are normalized in time and space. First, the data with a difference of more than 3 standard deviations from the mean of each corrected satellite data is filtered out, and the XCO2 data within this synthesis period are averaged within a radius of 10km, with months or years as the time unit, and the data with different spatiotemporal resolutions are unified to the same spatiotemporal scale to obtain a fused data set.

[0079] 3) Spatial auxiliary data processing

[0080] Atmospheric NO2 data are inverted using the Tropospheric Monitoring Instrument (TROPOMI) on the Sentinel-5P satellite to obtain daily atmospheric NO2 concentration data with a resolution of 0.01°, and the monthly mean values ​​are synthesized to obtain the monthly synthetic data of Sentinel-5P atmospheric NO2 concentration. The vegetation index product uses MODIS / MOD13A3 products, which use time weighting to convert the 16-day synthetic data of MOD13A2 into a monthly synthetic normalized vegetation index (NDVI) with a resolution of 1km. The elevation data uses SRTM / DEM, the data version is SRTMGL1_v003, and the spatial resolution is 30m. The temperature data uses ERA5-Land monthly synthetic reanalysis data, with a spatial resolution of 0.01°. All spatial independent variable data are uniformly resampled to a spatial resolution of 10km.

[0081] 4) Smoothing of atmospheric NO2 data

[0082] Considering that the half-life of NO2 in the atmosphere is short, about 35 hours, while CO2 has a longer life in the atmosphere, the correlation between the two should be greater in a short time, so the acquired atmospheric NO2 data is smoothed by Gaussian filtering, and the weight of each pixel in the window of the Gaussian smoothing filter is Gaussian distributed:

[0083]

[0084] In the formula, (μ, v) is the point coordinate, and σ is the standard deviation. The center point of the window is the origin, and the other points are weighted according to their positions (μ, v) according to the Gaussian function, and finally the weighted average of the center point of the window is calculated:

[0085]

[0086] Where R is the smoothing window radius, f(μ,v) is the original pixel value, and g(μ,v) is the corresponding weight.

[0087] In order to determine the optimal atmospheric NO2 smoothing effect, the atmospheric NO2 data were processed using Gaussian smoothing filters with different window sizes. The corresponding atmospheric NO2 data were extracted using the longitude and latitude of the XCO2 observation point. The correlation coefficients between the atmospheric NO2 data extracted with different smoothing effects and the atmospheric XCO2 remote sensing data were calculated. The smoothed atmospheric NO2 data with the highest correlation coefficient were selected as the input variable of the remote sensing estimation model.

[0088] 5) Construction of remote sensing estimation model

[0089] By matching latitude, longitude and time in space and time, the spatial independent variable values ​​corresponding to the atmospheric XCO2 observation data (atmospheric XCO2 remote sensing data) are extracted. With atmospheric XCO2 observation data as the dependent variable, atmospheric NO2 data, vegetation index, temperature data, and terrain data (elevation data) as independent variables, a Geographic Weighted Regression model (GWR) is constructed:

[0090] y(μ,ν)=β0(μ,ν)+β1(μ,ν)x1+β2(μ,ν)x2+β3(μ,ν)x3+β4(μ,ν)x4+ε(6);

[0091] Where y(μ, v) represents the atmospheric XCO2 observation value at the observation point (μ, v); μ and v are the longitude and latitude coordinates of the observation point, respectively; β0(μ, v) is the intercept of the equation; β1(μ, v)~β4(μ, v) are the regression coefficients of the corresponding independent variables; x1~x4 are the vegetation index, temperature, topography, and atmospheric NO2 data at the observation point (μ, v), respectively; ε is the random error term.

[0092] According to the weighted least squares method, the regression coefficient of the point is obtained, and the estimated value of the regression coefficient parameter β^(μ,v) is expressed as:

[0093] β^(μ,v)=[X T W(μ,v)X] -1 X T W(μ,v)Y (7);

[0094] Where W(μ,v) is the spatial weight diagonal matrix, X is the independent variable matrix, and Y is the dependent variable vector.

[0095] When the model is calculating the regression parameter β^, the most important thing is to determine the spatial weight matrix W. The weight of the observation point to the regression point is a function of the distance between the two points. The spatial weight is determined by the spatial kernel function and the bandwidth. The present invention uses the cross-validation method to calculate the bandwidth and the Gaussian function to calculate the spatial weight, and uses the adaptive method to determine the optimal bandwidth and weight of the model.

[0096] 6) Calculation of atmospheric XCO2 continuous data

[0097] The present invention uses a geographically weighted regression algorithm, takes atmospheric XCO2 remote sensing data as dependent variables, takes the optimal atmospheric NO2 smoothing data, vegetation index, temperature data, and terrain data as independent variables to construct a remote sensing estimation model, uses a cross-validation method to calculate the bandwidth, and uses a Gaussian function to calculate the spatial weight. The optimal bandwidth and weight of the geographically weighted regression model are determined by an adaptive method, and the regression coefficient of the independent variable is finally solved to obtain an atmospheric XCO2 remote sensing estimation model. The constructed optimal atmospheric XCO2 remote sensing estimation model is applied to spatially continuous independent variable data, and the atmospheric XCO2 continuous data is calculated to obtain the atmospheric XCO2 spatial continuous distribution.

[0098] Examples

[0099] The present invention uses multi-source atmospheric XCO2 remote sensing data as well as atmospheric NO2, topography, temperature, and vegetation index to obtain spatially continuous atmospheric XCO2 distribution. The following are specific implementation steps for examples. Figure 1 .

[0100] Taking a certain place as the study area, atmospheric XCO2 remote sensing data are extracted from GOSAT, OCO-2, and OCO-3 data products, and high-quality XCO2 data are obtained through quality control (Formula 1), as follows: Figure 2 , Figure 3 and Figure 4 shown.

[0101] Firstly, the CO2 profile of Carbon Tracker (CT) model data is used as the benchmark, and the XCO2 data inverted by GOSATACOS, OCO-2 and OCO-3 satellites are converted to a unified CO2 prior profile by formula 2. Then, the CO2 measurements at different satellite transit times are converted to the CO2 measurements at the reference time 13:30 (UTC) by formula 3. Finally, the multi-source satellite data are normalized in time and space, and the data with a difference of more than 3 times the standard deviation from the mean are filtered out for each corrected satellite data. The XCO2 data within a radius of 10 km are integrated and averaged based on the time unit of month, and the data with different temporal and spatial resolutions are unified to the same temporal and spatial scale to obtain a fused data set. The fused data is shown in Figure 2. Figure 5 shown.

[0102] Secondly, we obtained daily atmospheric NO2 concentration data with a resolution of 0.01° from Sentinel-5P_TROPOMI, and synthesized the monthly mean to obtain monthly scale data; obtained monthly synthetic normalized difference vegetation index NDVI with a resolution of 1km from MODIS vegetation index product MOD13A3; obtained monthly synthetic temperature data with a resolution of 0.1° from ERA5-Land; and obtained elevation data with a spatial resolution of 30m from SRTM / DEM. All spatial independent variable data were uniformly resampled to a spatial resolution of 1km.

[0103] Secondly, the atmospheric NO2 concentration is smoothed by Gaussian filtering. In order to find the atmospheric NO2 smoothing effect with the best correlation with the atmospheric XCO2 remote sensing data, the Gaussian filter windows are set to 51×51, 101×101, 151×151 and 201×201 respectively, and the atmospheric NO2 is smoothed by Gaussian filtering with different window sizes. In order to determine the optimal correlation between the atmospheric NO2 data and the atmospheric XCO2 remote sensing data, the corresponding atmospheric NO2 data are extracted using the longitude and latitude of the XCO2 observation point, and the correlation coefficients between the atmospheric NO2 data with different smoothing effects and the atmospheric XCO2 remote sensing data are calculated. The atmospheric NO2 smoothing result with the highest correlation coefficient is selected as the input variable of the atmospheric XCO2 remote sensing estimation model. In this example, the optimal window is 151×151 pixels, and the atmospheric NO2 data smoothed with this window size is used as the final model independent variable.

[0104] Finally, the spatial independent variable values ​​corresponding to the atmospheric XCO2 observation data were extracted by temporal and spatial matching of longitude, latitude and time. Based on the geographically weighted regression algorithm, the atmospheric XCO2 remote sensing data was used as the dependent variable, and the atmospheric NO2 data, vegetation index, temperature data and terrain data of the corresponding pixels were used as independent variables to construct the atmospheric XCO2 remote sensing estimation model. The cross-validation method was used to calculate the bandwidth, the Gaussian function was used to calculate the spatial weight, and the adaptive method was used to determine the optimal bandwidth and weight of the model. Finally, the model independent variable coefficients were solved to obtain the optimal atmospheric XCO2 remote sensing estimation model.

[0105] In this example, the optimal atmospheric XCO2 remote sensing estimation model obtained through the above steps is applied to spatially continuous independent variables such as atmospheric NO2 data, vegetation index, temperature data, and terrain data, and the continuous atmospheric XCO2 data of the study area is calculated to obtain the spatial distribution of atmospheric XCO2. Figure 6 A cross-validation accuracy scatter plot is given when the method of the present invention is applied to the spatialization of atmospheric XCO2 in a certain region. The samples are mainly distributed near the 1:1 line, the R value is 0.812, the mean absolute error RMSE and the mean absolute error MAE are 0.950ppm and 0.679ppm respectively, indicating that the estimation accuracy is relatively high. Figure 7The spatial distribution of atmospheric XCO2 in a certain region in June 2020 obtained by applying the method of the present invention is given. Atmospheric XCO2 is mainly between 405 and 417 ppm, showing significant spatial differences. The atmospheric XCO2 in the southern region is higher, significantly higher than that in the northern region.

[0106] The estimation results of the present invention show that when constructing an atmospheric XCO2 estimation model, considering NO2 pollution gas, which has a strong correlation with anthropogenic CO2 emissions, can effectively improve the model accuracy. The model calculation results reflect the fine spatial distribution of atmospheric XCO2 and can provide technical support for remote sensing monitoring of atmospheric XCO2.

[0107] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.

Claims

1. A method for spatialization of atmospheric XCO2 based on multi-source remote sensing data, characterized in that: The atmospheric XCO2 spatialization method comprises the following steps: Step A, acquiring atmospheric XCO2 remote sensing data from multiple greenhouse gas observation satellites, and preprocessing the acquired multi-source satellite atmospheric XCO2 remote sensing data; Step B, converting the multi-source satellite atmospheric XCO2 remote sensing data into a unified CO2 profile, interpolating or expanding the number of satellite CO2 profile layers to a unified CO2 profile layer based on the atmospheric vertical pressure layer, and then converting the XCO2 measurement value of the satellite transit time into the XCO2 measurement value of the unified reference time, and finally performing spatiotemporal normalization processing on the multi-source satellite data, unifying the data with different spatiotemporal resolutions to the same spatiotemporal scale, and obtaining a fused data set; Step C, obtaining spatial independent variable data including atmospheric NO2 data, vegetation index, elevation data and temperature data, and resampling all spatial independent variable data to the same spatial resolution; Step D, performing Gaussian filtering and smoothing processing on the resampled atmospheric NO2 data, calculating the correlation coefficients between the atmospheric NO2 data extracted with different smoothing effects and the atmospheric XCO2 remote sensing data, and selecting the atmospheric NO2 data smoothing result with the highest correlation coefficient as the input variable of the atmospheric XCO2 remote sensing estimation model; Step E, extracting the spatial independent variable value corresponding to the atmospheric XCO2 remote sensing data by performing spatiotemporal matching through longitude and latitude and time; constructing and training a geographically weighted regression model with atmospheric XCO2 remote sensing data as the dependent variable, atmospheric NO2 data, vegetation index, elevation data and temperature data as independent variables, calculating the regression coefficient of the point according to the weighted least squares method, determining the optimal bandwidth and weight of the geographically weighted regression model by the adaptive method, solving the regression coefficient of each independent variable, and obtaining the atmospheric XCO2 remote sensing estimation model; Step F, applying the atmospheric XCO2 remote sensing estimation model to the spatially continuous independent variable data, calculating the atmospheric XCO2 continuous data, and obtaining the atmospheric XCO2 spatially continuous distribution.

2. The method for spatialization of atmospheric XCO2 based on multi-source remote sensing data according to claim 1, characterized in that: In step A, the atmospheric XCO2 remote sensing data from multiple greenhouse gas observation satellites is obtained, and the process of preprocessing the acquired multi-source satellite atmospheric XCO2 data includes the following sub-steps: A1, respectively obtain the atmospheric XCO2 data of GOSAT satellite, the atmospheric XCO2 data of OCO-2 satellite and the atmospheric XCO2 data of OCO-3 satellite. The data version of the atmospheric XCO2 data of GOSAT satellite is ACOS_L2_Lite_FP.9r, the data version of the atmospheric XCO2 data of OCO-2 satellite is OCO2_L2_Lite_FP.11r, and the data version of the atmospheric XCO2 data of OCO-3 satellite is OCO3_L2_Lite_FP.10.4r; A2, obtain high-quality atmospheric XCO2 remote sensing data through the quality control mark screening provided by the data product. The quality control mark is: xco 2_ quality_flag=0.

3. The method for spatialization of atmospheric XCO2 based on multi-source remote sensing data according to claim 1, characterized in that: In step B, the process of obtaining the fused data set includes the following sub-steps: B1, taking the CO2 profile of Carbon Tracker model data as the benchmark, the atmospheric XCO2 remote sensing data inverted by GOSAT ACOS satellite, OCO-2 satellite and OCO-3 satellite are converted into a unified CO2 prior profile using the following formula: Where XCO 2adj,t is the corrected XCO2 value at satellite observation time t, XCO 2raw,t is the XCO2 value at the original satellite observation time t, h is the pressure weighting function, I is the unit matrix, A is the average kernel function calculated by the inversion algorithm, X CT,t is the CO2 profile distribution of the Carbon Tracker model corresponding to satellite observation time t, X a,t is the satellite’s prior CO2 profile distribution; B2, based on the atmospheric vertical pressure layer, the CO2 profile layers of the GOSAT ACOS satellite, OCO-2 satellite and OCO-3 satellite are interpolated or expanded to the CO2 profile layers of the Carbon Tracker model; B3, linearly interpolate the 8-period profile concentration distribution of the CT model data to the transit time of the satellite data, introduce a time shift correction coefficient based on the daily variation simulated by the Carbon Tracker model; convert the CO2 measurement value at the satellite transit time t Converted to CO2 measurement at reference time 13:30: Where rt is the reference time 13:30, XCO 2con,rt is the XCO2 value converted according to the satellite reference time rt, h t 、h rt are the CO2 profiles of the CT model corresponding to time t and rt, respectively; B4, for each type of corrected satellite data, the data with a difference of more than 3 standard deviations from the mean are filtered out, and the atmospheric XCO2 remote sensing data within the synthesis period are integrated and averaged within a radius of 10 km, with 1 month or 1 year as the synthesis period, and the data with different spatiotemporal resolutions are unified to the same spatiotemporal scale to obtain a fused data set.

4. The method for spatialization of atmospheric XCO2 based on multi-source remote sensing data according to claim 1, characterized in that: In step C, the daily data of atmospheric NO2 concentration with a resolution of 5 km are inverted using the Tropospheric Observation Instrument TROPOMI carried by the Sentinel-5P satellite; the 16-day composite data of the MODIS / MOD13A3 product is converted into a monthly composite normalized difference vegetation index NDVI with a resolution of 1 km using time weighting; SRTM / DEM is used to obtain elevation data, the data version of the elevation data is SRTMGL1_v003, and the spatial resolution is 30 m; the temperature data is obtained using the ERA5-Land monthly composite reanalysis data, and the spatial resolution of the temperature data is 10 km.

5. The method for spatialization of atmospheric XCO2 based on multi-source remote sensing data according to claim 1, characterized in that: In step C, all spatial independent variable data are uniformly resampled to a spatial resolution of 10 km.

6. The method for spatialization of atmospheric XCO2 based on multi-source remote sensing data according to claim 1, characterized in that: In step D, the resampled atmospheric NO2 data is subjected to Gaussian filtering and smoothing, and the process of calculating the correlation coefficients between the atmospheric NO2 remote sensing data extracted with different smoothing effects and the atmospheric XCO2 remote sensing data includes the following sub-steps: D1, Gaussian filtering and smoothing are performed on the acquired atmospheric NO2 data. The weights of each pixel in the window of the Gaussian smoothing filter are Gaussian distributed: In the formula, g(μ,v) is the weight corresponding to the observation point (μ,v); μ and v are the longitude and latitude coordinates of the observation point respectively; σ is the standard deviation. The center point in the window of the Gaussian smoothing filter is the origin, and the weights of other points are calculated according to their positions (μ,v) based on the Gaussian function, and the weighted average F(0,0) of the center point of the window is calculated: Where F(0,0) is the Gaussian smoothing value of the window center; R is the window radius of the Gaussian smoothing filter; f(μ,v) is the original pixel value at position (μ,v); g(μ,v) is the weight corresponding to position (μ,v); D2, use Gaussian smoothing filters with different window sizes to process atmospheric NO2 data, use the longitude and latitude of the XCO2 observation point to extract the corresponding atmospheric NO2, calculate the correlation coefficient between atmospheric NO2 and XCO2 extracted with different smoothing effects, and select the smoothing result of atmospheric NO2 data with the highest correlation coefficient as the input variable of the atmospheric XCO2 remote sensing estimation model.

7. The method for spatialization of atmospheric XCO2 based on multi-source remote sensing data according to claim 1, characterized in that: In step E, the process of obtaining the atmospheric XCO2 remote sensing estimation model includes the following sub-steps: By performing spatiotemporal matching of longitude, latitude and time, the spatial independent variable values ​​corresponding to the atmospheric XCO2 observation data are extracted; the atmospheric XCO2 remote sensing data is used as the dependent variable, and the atmospheric NO2 data, vegetation index, temperature data and elevation data are used as independent variables to construct a geographically weighted regression model: y(μ,ν)=β0(μ,ν)+β1(μ,ν)x1+β2(μ,ν)x2+β3(μ,ν)x3+β4(μ,ν)x4+ε Where y(μ,v) represents the atmospheric XCO2 observation value at the observation point (μ,v); μ and v are the longitude and latitude coordinates of the observation point, respectively; β0(μ,v) is the intercept of the equation; β1(μ,v), β2(μ,v), β3(μ,v) and β4(μ,v) are the regression coefficients of the corresponding independent variables; x1, x2, x3 and x4 are the smoothing results of the vegetation index, temperature data, elevation data and atmospheric NO2 data at the observation point, respectively; ε is the random error term; According to the weighted least squares method, the regression coefficient of the point is obtained, and the estimated value of the regression coefficient parameter β^(μ,v) is expressed as: b ^ (μ,v)=[X T W(μ,ν)X] -1 X T W(μ,v)Y Where W(μ,v) is the spatial weight diagonal matrix; X is the independent variable matrix; Y is the dependent variable vector; The cross-validation method was used to calculate the bandwidth, the Gaussian function was used to calculate the spatial weight, and the adaptive method was used to determine the optimal bandwidth and weight of the geographically weighted regression model. The regression coefficient of the independent variable was solved to obtain the atmospheric XCO2 remote sensing estimation model.

Images