A CO2 column concentration monitoring method based on multi-source data and hybrid geographic weighted regression

Abstract

This invention discloses a method for monitoring CO2 column concentration based on multi-source data and hybrid geographic weighted regression. The method includes: S1, acquiring spatiotemporally matched carbon satellite XCO2 datasets, model assimilation inversion reanalysis XCO2 datasets, meteorological datasets, and NDVI datasets to construct a spatiotemporally matched dataset for the study area; S2, constructing a hybrid geographic weighted regression model based on a geographic weighted regression model and a spatiotemporally geographic weighted regression model, inputting the spatiotemporally matched dataset into the hybrid geographic weighted regression model, and extracting auxiliary variables and carbon satellite XCO2 as dependent variables to construct a weighted regression equation; S3, obtaining model XCO2 estimates for all spatial locations in the study area based on the weighted regression equation, thereby obtaining a high spatiotemporally covered XCO2 dataset. This invention enables full-coverage monitoring of XCO2 data changes in the study area at both spatial and temporal scales, improving the ability to monitor spatiotemporal changes of XCO2 in the study area using satellite observation monitoring.

Description

Technical Field

[0001] This invention relates to the field of spatiotemporal full-coverage inversion monitoring of CO2 column concentration based on multi-source data, and particularly to a method for monitoring CO2 column concentration based on multi-source data and hybrid geographic weighted regression. Background Technology

[0002] Carbon dioxide (CO2) is the most important greenhouse gas in the atmosphere. Over the past decade, massive anthropogenic carbon emissions from burning fossil fuels and biomass have led to a rapid increase in atmospheric CO2 concentration. Annual analysis by global monitoring laboratories shows that the global average CO2 molar fraction reached a record high of 422.8 ppm in 2024. The continued rise in CO2 concentration is one of the main drivers of global warming, leading to a series of global problems such as sea-level rise, glacial melting, frequent extreme weather events, declining crop yields, and the spread of diseases. To mitigate the continued rise in atmospheric CO2 concentration, countries around the world have signed relevant climate conventions or agreements, pledging to control and reduce CO2 emissions to address global climate change. Monitoring and assessing the spatiotemporal changes in atmospheric CO2 concentration are crucial for carbon cycle research and emission reduction.

[0003] With increasing focus on greenhouse gas control and emission reduction, various research institutions have launched carbon satellite observation projects, such as the Scanning Imaging Absorption Spectrometer for Atmospheric Carbon Y (SCIMACHY) aboard the European Space Agency's Environmental Satellite (ENVISAT), Japan's Greenhouse Gases Observing Satellite (GOSAT) and GOSAT-2, NASA's Orbiting Carbon Observatory-2 (OCO-2) and OCO-3, and China's Global Carbon Dioxide Monitoring Scientific Experiment Satellite (TanSat). More carbon monitoring satellites will be launched in the future. Compared to ground-based monitoring, these carbon satellites provide unified monitoring, broader coverage, and more continuous spatial distribution for retrieving column-average atmospheric carbon dioxide (CO2) mole fraction (XCO2). Numerous applied studies have demonstrated that XCO2 data obtained through satellite remote sensing can effectively capture the spatial distribution, regional differences, source-sink variations, seasonal changes, and trends of atmospheric carbon dioxide (CO2) concentration at global and large-scale regional levels, and shows good consistency with observations from the Total Carbon Column Observing Network (TCCON) ground stations. However, the current number of carbon monitoring satellites is insufficient to achieve full geographical coverage. Furthermore, factors such as atmospheric aerosols, clouds, and surface albedo limit the application of satellite observations in capturing the spatiotemporal variations of XCO2 across the entire spatial coverage area. Therefore, achieving full-coverage CO2 column concentration monitoring by utilizing multi-source data combined with XCO2 data from carbon satellites with limited coverage is a current technical challenge, aiming to expand the monitoring area and improve monitoring accuracy. Summary of the Invention

[0004] The purpose of this invention is to provide a CO2 column concentration monitoring method based on multi-source data and hybrid geographic weighted regression. An auxiliary variable dataset is constructed using model assimilation, inversion, and reanalysis XCO2 datasets, meteorological datasets, and NDVI datasets. A spatiotemporal matching dataset is then constructed by spatiotemporally matching the auxiliary variable dataset with the carbon satellite XCO2 dataset. A weighted regression equation is constructed using a hybrid geographic weighted regression model, and equations for the surrounding neighboring regions are constructed based on effective distance thresholds. This achieves accurate inversion calculations of XCO2 data with high spatiotemporal coverage, improving the ability to monitor spatiotemporal changes of XCO2 in the study area using satellite observation.

[0005] The objective of this invention is achieved through the following technical solution:

[0006] A method for monitoring CO2 column concentration based on multi-source data and hybrid geographic weighted regression, the method comprising:

[0007] S1. Obtain the spatiotemporal matching carbon satellite XCO2 dataset, model assimilation inversion reanalysis XCO2 dataset, meteorological dataset, and NDVI dataset for the study area to construct a spatiotemporal matching dataset.

[0008] S2. Construct a hybrid geographic weighted regression model based on a geographic weighted regression model and a spatiotemporal geographic weighted regression model. Input the spatiotemporal matching dataset into the hybrid geographic weighted regression model. The hybrid geographic weighted regression model extracts auxiliary variables from the model assimilation inversion reanalysis XCO2 dataset, meteorological dataset, and NDVI dataset, and extracts carbon satellite XCO2 from the carbon satellite XCO2 dataset as the dependent variable to construct the following weighted regression equation:

[0009] ,in This is the estimated value of XCO2 from the model at spatial location i; Let i be the regression coefficient of spatial location i in the study area. The regression coefficient of the auxiliary variable k Obtained using the nearest neighbor allocation method; Data for spatial location i and auxiliary variable k;

[0010] S3. Using the weighted regression equation of method S2, obtain the model XCO2 estimates for all spatial locations in the study area, thus obtaining the XCO2 dataset with high spatiotemporal coverage.

[0011] To better implement the present invention, in method S1, carbon satellite XCO2 data in the carbon satellite XCO2 dataset are synthesized and gridded according to the monthly scale to obtain a gridded monthly carbon satellite XCO2 dataset. The monthly carbon satellite XCO2 dataset is used as the carbon satellite inversion XCO2 dataset and, together with the model assimilation inversion reanalysis XCO2 dataset, meteorological dataset, and NDVI dataset, undergoes preprocessing including uniform resolution processing and spatiotemporal matching processing to construct the spatiotemporal matching dataset.

[0012] Preferably, the data from the model assimilation, inversion, and reanalysis XCO2 dataset, meteorological dataset, and NDVI dataset are resampled and assigned to the corresponding spatial grid of the monthly carbon satellite XCO2 dataset according to the spatiotemporal matching method using the moving average method and / or bilinear interpolation method. The data from the spatiotemporal matching dataset only partially covers the grid of the study area.

[0013] Preferably, in method S2, the auxiliary variables extracted by the hybrid geographic weighted regression model include model reanalysis XCO2, 10 m horizontal and vertical wind speeds, atmospheric boundary layer height, surface air pressure, 2 m temperature, total precipitation, evaporation, net solar radiation at the surface, and NDVI data. The model reanalysis XCO2 data comes from the model assimilation and inversion reanalysis XCO2 dataset. The 10 m horizontal and vertical wind speeds, atmospheric boundary layer height, surface air pressure, 2 m temperature, total precipitation, evaporation, and net solar radiation at the surface are mainly derived from meteorological datasets. The NDVI data comes from the NDVI dataset, and the NDVI data is the normalized vegetation index.

[0014] Preferably, the hybrid geographic weighted regression model is constructed based on the relationship between the distance between the target sample location and the locations of its neighboring sample locations j and the effective distance threshold, using a combination of geographic weighted regression and spatiotemporal geographic weighted regression, as follows:

[0015] ,in, The lunar-scale carbon satellite XCO2 value at spatial location i of the target sample point; The distance between the target sample point location i and the surrounding neighboring sample point locations j; For the set effective distance threshold, , These are the spatial coordinates and spatiotemporal coordinates of position i, respectively. For the spatial location i and the auxiliary variable k, For residuals, , , respectively, are the local regression coefficients of spatial location i and auxiliary variable k.

[0016] Preferably, the regression coefficient estimation of spatial location i The method is as follows:

[0017] The distance between the spatial location i of the target sample point and the locations j of its neighboring sample points is calculated. With effective distance threshold The magnitude of the regression coefficients is determined, and then the regression coefficients are estimated using the following expression. :

[0018] Where X and Y are the vectors of the auxiliary variables and the dependent variable, respectively. Let X be the transpose of vector X. Let i be the weight matrix for spatial location i; if the distance between the location i of the target sample point and the locations j of its neighboring sample points is... Less than the effective distance threshold The spatial relationships and weight matrix are determined using a geographic weighted regression model, and local spatial regression is performed based on the observation information of the target sample point and its neighboring sample points; if the distance between the location i of the target sample point and the location j of its neighboring sample points is... Not less than the effective distance threshold The spatiotemporal relationship and weight matrix are determined using a spatiotemporal geographic weighted regression model, and spatiotemporal local regression is performed based on the observation information of the target sample point and its surrounding neighboring sample points; the regression coefficients of the spatial location i of the study area are... Regression coefficient estimation using spatial neighborhood or historical spatiotemporal neighborhood .

[0019] Preferably, the hybrid geographic weighted regression model is used for the hybrid geographic weighted regression value of the carbon satellite XCO2 at the spatial location i of the target sample point in the carbon satellite XCO2 dataset. The expression is:

[0020] ,in To estimate the regression coefficients for the spatial location i of the target sample point, This is the data for spatial location i.

[0021] Preferably, the hybrid geographic weighted regression model extracts auxiliary variables from the XCO2 dataset, meteorological dataset, and NDVI dataset to form an auxiliary variable dataset; the high spatiotemporal coverage XCO2 dataset is combined with the hybrid geographic weighted regression values ​​and the model XCO2 estimates of the weighted regression equation for data matching and full coverage.

[0022] Preferably, the weights between the spatial position i of the target sample point and its neighboring point j in the weight matrix are... The expression is as follows: ,in Let be the distance between the spatial location i of the target sample point and its neighboring point j, and b be the bandwidth parameter.

[0023] Preferably, the present invention further includes the following method:

[0024] S4. Following method S2, obtain the estimated value of model XCO2 for spatial location i arranged in time series, and obtain XCO2 time series variation data; use the XCO2 time series variation data to visualize the monthly scale variation, seasonal variation, interannual variation and trend variation of XCO2.

[0025] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0026] (1) This invention uses the XCO2 dataset, meteorological dataset and NDVI dataset to construct an auxiliary variable dataset, and uses the auxiliary variable dataset to construct a spatiotemporal matching dataset by spatiotemporal matching with the carbon satellite XCO2 dataset. It constructs a weighted regression equation by using a hybrid geographic weighted regression model and constructs equations for the surrounding neighboring regions according to the effective distance threshold relationship. This achieves accurate inversion calculation of XCO2 data with high spatiotemporal coverage and improves the ability to monitor the spatiotemporal changes of XCO2 in the satellite observation and monitoring area of ​​the study area.

[0027] (2) The hybrid geographic weighted regression model of this invention extracts auxiliary variables including meteorological factors and vegetation factors from the auxiliary variable dataset. Through the hybrid geographic weighted regression model inversion calculation, high-quality XCO2 data with full spatiotemporal coverage can be obtained, realizing full coverage XCO2 data change monitoring in the study area at both spatial and temporal scales.

[0028] (3) This invention solves the problem of large spatial gaps and spatiotemporal unevenness in carbon satellite observation data. By combining geographic weighted regression and spatiotemporal geographic weighted regression to construct a hybrid geographic weighted regression model, the accuracy of spatial prediction can be effectively improved. At the same time, it can provide an explanation of the spatial heterogeneity relationship between XCO2 and auxiliary variables. It can generate a high-quality spatiotemporal coverage XCO2 dataset, which can be used in the assimilation and inversion of atmospheric transport models to obtain carbon emission inventories, providing effective data and technical support for carbon emission monitoring and assessment in the study area. Attached Figure Description

[0029] Figure 1 This is a flowchart of the CO2 column concentration monitoring method of the present invention;

[0030] Figure 2 This is a schematic diagram illustrating the principle of the CO2 column concentration monitoring method in the embodiments;

[0031] Figure 3 As an example, a spatial distribution map of XCO2 estimated by a hybrid geographically weighted regression (HGWR) model for a certain month and year is extracted from a portion of the study area.

[0032] Figure 4As an example, this study area includes a section showing the variation trend of XCO2 in different seasons over multiple years.

[0033] Figure 5 As an example, a map showing the annual average XCO2 and its changing trend of a portion of the study area over multiple years is extracted.

[0034] Figure 6 The image shows the spatial distribution of average XCO2 in a portion of the study area during different seasons, as shown in the example. Detailed Implementation

[0035] The present invention will be further described in detail below with reference to embodiments:

[0036] Example

[0037] like Figure 1 As shown, a method for monitoring CO2 column concentration based on multi-source data and hybrid geographic weighted regression is described, the method comprising:

[0038] S1. Obtain the spatiotemporal matching carbon satellite XCO2 dataset, model assimilation inversion reanalysis XCO2 dataset, meteorological dataset, and NDVI dataset for the study area to construct a spatiotemporal matching dataset.

[0039] In some embodiments, carbon satellite XCO2 data in the carbon satellite XCO2 dataset are synthesized and gridded on a monthly scale to obtain a gridded monthly carbon satellite XCO2 dataset. The monthly carbon satellite XCO2 dataset is used as the carbon satellite inversion XCO2 dataset and, together with the model assimilation inversion reanalysis XCO2 dataset, meteorological dataset, and NDVI dataset, undergoes preprocessing including uniform resolution processing and spatiotemporal matching processing to construct a spatiotemporal matching dataset.

[0040] Data from the XCO2 dataset, meteorological dataset, and NDVI dataset, obtained through model assimilation, inversion, and reanalysis, are resampled and assigned to the corresponding spatial grids of the monthly carbon satellite XCO2 dataset using a moving average method and / or bilinear interpolation, according to spatiotemporal matching. (This invention constructs a geographic map of the study area, which is divided into spatial grids. The scale of the spatial grids is based on the data scale of the carbon satellite XCO2 dataset. Generally, the spatial resolution of the geographic map's preset spatial grid should be lower than the spatial resolution of the carbon satellite sampling. In this embodiment, a moving average method is used to synthesize monthly-scale carbon satellite XCO2 data from daily carbon satellite XCO2 data and assign it to the spatial grids of the geographic map.) Due to the limitations of carbon satellite XCO2 data (which only covers a portion of the geographic map grid), the spatiotemporal matching dataset only partially covers the study area's grid. The meteorological and NDVI datasets were resampled and assigned to the geographic map grid according to the spatiotemporal matching method using moving average and / or bilinear interpolation. The meteorological and NDVI datasets can fully cover the geographic map. However, due to the limitations of carbon satellite XCO2 data, the constructed spatiotemporal matching dataset only partially covers the study area's grid. The NDVI dataset (i.e., the Normalized Difference Vegetation Index dataset) can be generated from satellite imagery, and its preprocessing includes image stitching, projection transformation, resampling (using bilinear interpolation), and cropping.

[0041] In this embodiment, a geographical region in the Chinese mainland is designated as the study area. The study area's carbon satellite inversion XCO2 dataset (Orbiting Carbon Observatory 2 satellite inversion Level 2 daily XCO2 data), model reanalysis XCO2 dataset (European Centre for Medium-Range Weather Forecasts (ECMWF) Copernicus Atmospheric Monitoring Service (CAMS) global greenhouse gas reanalysis XCO2 data), model reanalysis meteorological data (ERA5 fifth-generation global climate and atmosphere reanalysis meteorological data), NDVI dataset (i.e., MOD13A3 NDVI data), and observation data from the Global Carbon Plumage Concentration Observation Network (TCCON) stations (as validation data for subsequent validation of high spatiotemporal coverage XCO2 data) are acquired and organized. Specifically, the carbon satellite inversion XCO2 dataset for the study area was derived from the OCO-2 Level 2 daily XCO2 product (https: / / www.jpl.nasa.gov / missions / orbiting-carbon-observatory-2-oco-2 / ), serving as the carbon satellite inversion XCO2 dataset. The model assimilation inversion reanalysis XCO2 dataset was derived from the CAMS monthly mean XCO2 product (https: / / ads.atmosphere.copernicus.eu / datasets), serving as the model inversion reanalysis XCO2 dataset. The ERA 5 monthly mean reanalysis meteorological data product for the study area (https: / / cds.climate.copernicus.eu / datasets), including eight meteorological elements—10 m horizontal and vertical wind speed, atmospheric boundary layer height, surface pressure, 2 m temperature, total precipitation, evaporation, and net solar radiation—was collected and compiled as the model reanalysis meteorological dataset. We collected and organized the monthly MODIS vegetation index data product (MOD13A3) (https: / / modis.gsfc.nasa.gov / data / ) for the study area as the NDVI dataset.

[0042] In this example where a specific geographical area is the study area, carbon satellite XCO2 data is synthesized and gridded. The gridding process uses a spatial resolution of 0.1°×0.1° to generate a grid, resulting in gridded monthly-scale carbon satellite XCO2 data. Model reanalysis XCO2 data, model reanalysis meteorological data, and NDVI data are preprocessed to obtain a preprocessed auxiliary variable dataset with uniform resolution. Spatiotemporal matching is then performed on the gridded carbon satellite XCO2 data and the preprocessed auxiliary variables to obtain a spatiotemporally matched dataset. Specifically, in this embodiment, R language and moving average are used to resynthesize the daily OCO-2 Level 2 data into monthly OCO-2 XCO2 data and assign it to the generated spatial grid, resulting in monthly OCO-2 XCO2 data with a resolution of 0.1°×0.1°. This gridded OCO-2 XCO2 data only retains grid points with more than 5 valid data days in each month, thus only partially covering the space. The CAMS XCO2 data and ERA5 meteorological data are resampled to a spatial resolution of 0.1°×0.1° using bilinear interpolation, resulting in CAMS XCO2 data and ERA5 meteorological data with a spatial resolution of 0.1°×0.1°. In ArcMAP 10.8, image stitching, projection conversion, resampling (using bilinear interpolation), and cropping are performed on the MOD13A3 NDVI data product to obtain NDVI data with a spatial resolution of 0.1°×0.1°. CAMS XCO2 data, ERA5 meteorological data, and NDVI data with uniform resolution (0.1°×0.1°) were used as preprocessed auxiliary variable datasets for subsequent estimation models. Then, gridded OCO-2 XCO2 data were used as a benchmark and spatiotemporally matched with the preprocessed auxiliary variable datasets to obtain a spatiotemporally matched dataset.

[0043] S2. Construct a hybrid geographic weighted regression model based on a geographic weighted regression model and a spatiotemporal geographic weighted regression model. Input the spatiotemporal matching dataset into the hybrid geographic weighted regression model. The hybrid geographic weighted regression model extracts auxiliary variables from the model assimilation inversion reanalysis XCO2 dataset, meteorological dataset, and NDVI dataset, and extracts carbon satellite XCO2 from the carbon satellite XCO2 dataset as the dependent variable to construct the following weighted regression equation:

[0044] ,in This is the estimated value of XCO2 from the model at spatial location i; The regression coefficients for spatial location i in the study area (selected from the regression coefficient estimates corresponding to the mixed geographic weighted regression model). ), The regression coefficients of auxiliary variable k are (the regression coefficients of auxiliary variable k are obtained after processing the mixed geographical weighted regression model). Obtained using the nearest neighbor allocation method; This refers to the data for spatial location i and auxiliary variable k. In this example where a certain geographical area is the study area, the total number of auxiliary variables k is 10 (i.e., model reanalysis XCO2, 10 m horizontal and vertical wind speeds, atmospheric boundary layer height, surface air pressure, 2 m temperature, total precipitation, evaporation, net solar radiation at the surface, and NDVI data).

[0045] In this invention, the auxiliary variables extracted by the hybrid geographic weighted regression model include model reanalysis XCO2, 10 m horizontal and vertical wind speeds, atmospheric boundary layer height, surface air pressure, 2 m temperature, total precipitation, evaporation, net solar radiation at the surface, and NDVI data. The model reanalysis XCO2 data comes from the model assimilation and inversion reanalysis XCO2 dataset. The 10 m horizontal and vertical wind speeds, atmospheric boundary layer height, surface air pressure, 2 m temperature, total precipitation, evaporation, and net solar radiation at the surface are mainly derived from meteorological datasets. The NDVI data comes from the NDVI dataset, and the NDVI data is the normalized vegetation index.

[0046] The hybrid geographic weighted regression model, based on the relationship between the distance between the target sample location and the locations of its neighboring sample locations j and the effective distance threshold, combines geographic weighted regression and spatiotemporal geographic weighted regression to construct the following equation:

[0047] ,in, The monthly-scale carbon satellite XCO2 value at spatial location i of the target sample point (the monthly-scale carbon satellite XCO2 value obtained by inversion based on a hybrid geographic weighted regression model). The distance between the target sample point location i and the surrounding neighboring sample point locations j; The effective distance threshold is set (which can be selected based on the distance distribution of the hybrid geographic weighted regression model). , These are the spatial coordinates and spatiotemporal coordinates of position i, respectively. For the spatial location i and the auxiliary variable k, For residuals, obey and . , and are the local regression coefficients of spatial location i and auxiliary variable k, respectively. In the formula... Indicates if Less than the effective distance threshold , in the formula Indicates if Greater than or equal to the effective distance threshold .

[0048] Local regression coefficients are obtained by applying weighted least squares estimation to the neighboring observations of the target point's spatial location i (using spatiotemporal local regression using observation information of the target point and spatially and temporally nearby points). The regression coefficients for spatial location i are estimated. The method is as follows:

[0049] The distance between the spatial location i of the target sample point and the locations j of its neighboring sample points is calculated. With effective distance threshold The magnitude of the regression coefficients is determined, and then the regression coefficients are estimated using the following expression. :

[0050] Where X and Y are the vectors of the auxiliary variables and the dependent variable, respectively. Let X be the transpose of vector X. Let i be the weight matrix for spatial location i; if the distance between the location i of the target sample point and the locations j of its neighboring sample points is... Less than the effective distance threshold Spatial relationships and weight matrices are determined using a geographic weighted regression model, and local spatial regression is performed based on the observation information of the target sample point and its neighboring sample points. The distance between the location i of the target sample point and the location j of its neighboring sample points is considered. Not less than the effective distance threshold The spatiotemporal relationship and weight matrix are determined using a spatiotemporal geographic weighted regression model, and spatiotemporal local regression is performed based on the observation information of the target sample point and its surrounding neighboring sample points; in this embodiment, the regression coefficient of spatial location i in the study area is... Regression coefficient estimation using spatial neighborhood or historical spatiotemporal neighborhood This invention uses observational information from the target point and nearby spatial and temporal points to perform spatiotemporal local regression; thus, when spatial observational information is relatively sparse, the accuracy of the estimation is improved by supplementing it with observational information from nearby times, since regional variables such as atmospheric CO2 concentration usually have obvious spatiotemporal correlations.

[0051] In some embodiments, the weights between the spatial location i of the target sample point and its neighboring point j in the weight matrix The expression is as follows: ,in Let be the distance between the spatial location i of the target sample point and its neighboring point j, and b be the bandwidth parameter (used to control the decay rate of the weights). In the formula... Indicates if Less than or equal to the bandwidth parameter b, in the formula Indicates if It is greater than or equal to the bandwidth parameter b.

[0052] The hybrid geographic weighted regression model performs hybrid geographic weighted inversion calculations on the spatial location i of the target sample point of the carbon satellite XCO2 dataset to obtain the hybrid geographic weighted regression value. The expression is:

[0053] ,in To estimate the regression coefficients for the spatial location i of the target sample point, This is the data for spatial location i. Specifically, , where X is a vector of auxiliary variables (the auxiliary variables extracted by the mixed geographic weighted regression model include model reanalysis XCO2, 10 m horizontal and vertical wind speeds, atmospheric boundary layer height, surface air pressure, 2 m temperature, total precipitation, evaporation, net solar radiation at the surface, and NDVI data), and y is the surrounding nearby observation information of location i (i.e., the carbon satellite XCO2 data corresponding to location i in its surrounding vicinity). Let X be the transpose of vector X. Let be the weight matrix for spatial location i.

[0054] In the example where a specific geographical region is selected as the study area in this embodiment, the hybrid geographically weighted regression model of this invention is created by fusing a geographically weighted regression model and a spatiotemporal geographically weighted regression model. If this invention uses the hybrid geographically weighted regression model, single geographically weighted regression, and single spatiotemporal geographically weighted regression respectively in the weighted regression equations in method S2 to perform regression inversion, the comparison results of the estimation results deviation are shown in Table 1 below:

[0055] Table 1. Comparison of estimation results between the hybrid geographic weighted regression model of this invention and single geographic weighted regression and single spatiotemporal geographic weighted regression.

[0056]

[0057] Table 1 shows the absolute bias (MAE) of CAMS data and estimation results from different methods relative to OCO XCO2 data for different distances (d) between the target grid point and the nearest neighbor observation. The results show that the bias of both the geographically weighted regression (GWR_MAE) and the spatiotemporal geographically weighted regression (CMAS_MAE) increases with the increase of the distance between the target grid point and the nearest neighbor observation, indicating poor performance in the case of observation comparison coefficients. However, the bias of the CMAS XCO2 data does not exhibit this characteristic. The spatiotemporal geographically weighted regression model performs better than the geographically weighted regression model when observation data is sparse due to the inclusion of nearby time data. When the distance between the target grid point and the nearest neighbor grid point is greater than 50 km, the geographically weighted regression method has a larger error; conversely, it performs better than the spatiotemporal geographically weighted regression method when the distance is smaller. Therefore, based on the statistical results, the effective distance threshold is determined in this example. For a distance of 50 km, when the distance between the target grid point and the nearest observation point is less than 50 km, a geographically weighted regression model is used to determine the spatial relationship; otherwise, a spatiotemporal geographically weighted regression model is used. This yields the mixed regression parameters for the locations of gridded carbon dioxide satellites with effective XCO2 information each month. Then, the nearest neighbor assignment method is used to obtain regression coefficient estimates for the spatial distribution information of the regression parameters at a spatial resolution of 0.1°×0.1°. .

[0058] S3. Following the weighted regression equation of method S2, obtain the model XCO2 estimates for all spatial locations in the study area, thus obtaining a high spatiotemporal coverage XCO2 dataset. For example... Figure 2 As shown, the hybrid geographic weighted regression model of this invention extracts auxiliary variables from the XCO2 dataset, meteorological dataset, and NDVI dataset to form an auxiliary variable dataset. The high spatiotemporal coverage XCO2 dataset is combined with the hybrid geographic weighted regression values ​​and the model XCO2 estimates of the weighted regression equation for data matching and full coverage.

[0059] In this example, which selects a specific geographical area as the study area, three statistical indicators—root mean square error (RMSE), coefficient of determination (R²), and mean bias—are used. Ten-fold cross-validation and site-specific validation are employed to validate the high spatiotemporal coverage XCO2 data. The validation results are shown in Table 2 below.

[0060] Table 2 Statistical Table of Verification Indicators for High Spatiotemporal Coverage XCO2 Data

[0061]

[0062] Table 2 shows the monthly CAMS XCO2 data for the study area and the validation results of the hybrid geographically weighted regression model. The results indicate that the accuracy of the estimation results from the hybrid geographically weighted regression model (HGWR) is significantly improved compared to the XCO2 obtained from CAMS model assimilation and inversion. Specifically, the 10-fold cross-validation results show higher RMSE, lower bias, and higher R-value. 2 The statistical index values ​​were 0.88 ppm, -0.01 ppm, and 0.97, respectively. The RMSE, Bias, and R were verified based on TCCON site data (i.e., site observation data obtained in this embodiment based on the global carbon column concentration observation network TCCON). 2 The statistical values ​​were 0.90 ppm, 0.13 ppm and 0.96, respectively, indicating that the hybrid geographic weighted regression model can better estimate XCO2 information and generate monthly XCO2 data for China's land area with full spatiotemporal coverage. Figure 3The study shows the spatial distribution of XCO2 estimated by a hybrid geographic weighted regression (HGWR) model for a specific year and month in a selected area of ​​the study region. It also compares the XCO2 data from the OCO-2 carbon satellite and the XCO2 data from the CAMS model reanalysis. The results show that the OCO-2 data is largely spatially blank and cannot reflect the spatial variation of XCO2. The CAMS XCO2 data and the HGWR-estimated XCO2 results for the study region in a specific year and month exhibit similar spatial patterns. However, compared to the CAMS data, the HGWR-estimated XCO2 has higher spatial resolution and reflects more details of XCO2 spatial variation.

[0063] S4. Following method S2, obtain the estimated value of model XCO2 for spatial location i arranged in time series, and obtain XCO2 time series variation data; use the XCO2 time series variation data to visualize the monthly scale variation, seasonal variation, interannual variation and trend variation of XCO2.

[0064] In some embodiments, a hybrid geographic weighted regression model is used to generate XCO2 data of the study area at a spatial resolution of 0.1°×0.1° for multiple years and months. The average XCO2 concentration data at different time scales of season, year and multi-year are calculated and their spatial distribution characteristics are analyzed. The monthly average XCO2 value of the entire study area is calculated and its time series changes are analyzed. The Mann-Kendall test is used to analyze the trend characteristics of XCO2. Figure 4 To illustrate the variation trends of XCO2 in different seasons over multiple years in a portion of the study area, Figure 4 The study shows the trend of XCO2 time series changes in different seasons and months in selected areas of the study area over multiple years. Overall, the CO2 concentration in the study area shows a continuous upward trend over multiple years, with obvious seasonal cyclical changes, with the highest concentration in spring and the lowest concentration in summer. Figure 5 As an example, this study area includes a portion of the region showing the annual average XCO2 levels and their spatial distribution trends over multiple years. Figure 5 This shows the spatial distribution of the annual average XCO2 and its trend analysis for multiple years in the study area. Figure 5 a represents the annual average distribution of XCO2 in the study area over multiple years. Figure 5 b represents the distribution of XCO2 variation trends over multiple years in the study area. Figure 6 This is a partial map showing the spatial distribution of average XCO2 in different seasons within the study area, as shown in the example. Figure 6 This shows the spatial distribution of average XCO2 in different seasons in the study area. Figure 6 a represents the spatial distribution of XCO2 in a certain region of the study area during spring. Figure 6 b represents the spatial distribution of XCO2 in a certain region of the study area during summer. Figure 6 c represents the spatial distribution of XCO2 in a certain region of the study area during autumn. Figure 6 (d represents the spatial distribution of XCO2 in a certain area of ​​the study area during winter). It can be seen that the spatial distribution pattern of XCO2 in different seasons of multiple years in the study area is different, especially between summer and other seasons. The spatial distribution trend of spring, autumn and winter is similar to the multi-year average.

[0065] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for monitoring CO2 column concentration based on multi-source data and hybrid geographic weighted regression, characterized in that: The methods include: S1. Obtain the spatiotemporal matching carbon satellite XCO2 dataset, model assimilation inversion reanalysis XCO2 dataset, meteorological dataset, and NDVI dataset for the study area to construct a spatiotemporal matching dataset. S2. Construct a hybrid geographic weighted regression model based on a geographic weighted regression model and a spatiotemporal geographic weighted regression model. Input the spatiotemporal matching dataset into the hybrid geographic weighted regression model. The hybrid geographic weighted regression model extracts auxiliary variables from the model assimilation inversion reanalysis XCO2 dataset, meteorological dataset, and NDVI dataset, and extracts carbon satellite XCO2 from the carbon satellite XCO2 dataset as the dependent variable to construct the following weighted regression equation: ,in This is the estimated value of XCO2 from the model at spatial location i; Let i be the regression coefficient of spatial location i in the study area. The regression coefficient of the auxiliary variable k Obtained using the nearest neighbor allocation method; The data are spatial location i and auxiliary variable k; the hybrid geographic weighted regression model is constructed based on the relationship between the distance and effective distance threshold between the target sample location and the surrounding neighboring sample locations j, using both geographic weighted regression and spatiotemporal geographic weighted regression, as follows: ,in, The lunar-scale carbon satellite XCO2 value at spatial location i of the target sample point; The distance between the target sample point location i and the surrounding neighboring sample point locations j; For the set effective distance threshold, , These are the spatial coordinates and spatiotemporal coordinates of position i, respectively. For the spatial location i and the auxiliary variable k, For residuals, , These are the local regression coefficients of spatial location i and auxiliary variable k, respectively; S3. Using the weighted regression equation of method S2, obtain the model XCO2 estimates for all spatial locations in the study area, thus obtaining the XCO2 dataset with high spatiotemporal coverage.

2. The CO2 column concentration monitoring method based on multi-source data and hybrid geographically weighted regression according to claim 1, characterized in that: In method S1, carbon satellite XCO2 data in the carbon satellite XCO2 dataset are synthesized and gridded according to the monthly scale to obtain a gridded monthly carbon satellite XCO2 dataset. The monthly carbon satellite XCO2 dataset is used as the carbon satellite inversion XCO2 dataset and, together with the model assimilation inversion reanalysis XCO2 dataset, meteorological dataset, and NDVI dataset, undergoes preprocessing including uniform resolution processing and spatiotemporal matching processing to construct the spatiotemporal matching dataset.

3. The CO2 column concentration monitoring method based on multi-source data and hybrid geographically weighted regression according to claim 2, characterized in that: Data from the XCO2 dataset, meteorological dataset, and NDVI dataset were resampled and assigned to the corresponding spatial grid of the monthly carbon satellite XCO2 dataset using the moving average method and / or bilinear interpolation method according to spatiotemporal matching. The spatiotemporal matching dataset only partially covers the grid of the study area.

4. The CO2 column concentration monitoring method based on multi-source data and hybrid geographically weighted regression according to claim 1, characterized in that: In Method S2, the auxiliary variables extracted by the mixed geographic weighted regression model include model reanalysis XCO2, 10 m horizontal and vertical wind speeds, atmospheric boundary layer height, surface air pressure, 2 m temperature, total precipitation, evaporation, net solar radiation at the surface, and NDVI data. Model reanalysis XCO2 is derived from the model assimilation and inversion reanalysis XCO2 dataset. 10 m horizontal and vertical wind speeds, atmospheric boundary layer height, surface air pressure, 2 m temperature, total precipitation, evaporation, and net solar radiation at the surface are derived from meteorological datasets. NDVI data is derived from the NDVI dataset, and NDVI data is the normalized vegetation index.

5. The CO2 column concentration monitoring method based on multi-source data and hybrid geographically weighted regression according to claim 1, characterized in that: Regression coefficient estimation of spatial location i The method is as follows: The distance between the spatial location i of the target sample point and the locations j of its neighboring sample points is calculated. With effective distance threshold The magnitude of the regression coefficients is determined, and then the regression coefficients are estimated using the following expression. : Where X and Y are the vectors of the auxiliary variables and the dependent variable, respectively. Let X be the transpose of vector X. Let i be the weight matrix for spatial location i; if the distance between the location i of the target sample point and the locations j of its neighboring sample points is... Less than the effective distance threshold The spatial relationships and weight matrix are determined using a geographic weighted regression model, and local spatial regression is performed based on the observation information of the target sample point and its neighboring sample points; if the distance between the location i of the target sample point and the location j of its neighboring sample points is... Not less than the effective distance threshold The spatiotemporal relationship and weight matrix are determined using a spatiotemporal geographic weighted regression model, and spatiotemporal local regression is performed based on the observation information of the target sample point and its surrounding neighboring sample points; the regression coefficients of the spatial location i of the study area are... Regression coefficient estimation using spatial neighborhood or historical spatiotemporal neighborhood .

6. The CO2 column concentration monitoring method based on multi-source data and hybrid geographic weighted regression according to claim 1 or 5, characterized in that: The hybrid geographic weighted regression model is used to calculate the spatial location i of the target sample point for the carbon satellite XCO2 in the XCO2 dataset. The expression is: ,in To estimate the regression coefficients for the spatial location i of the target sample point, This is the data for spatial location i.

7. The CO2 column concentration monitoring method based on multi-source data and hybrid geographically weighted regression according to claim 6, characterized in that: The hybrid geographic weighted regression model extracts auxiliary variables from the XCO2 dataset, meteorological dataset, and NDVI dataset to form an auxiliary variable dataset; the high spatiotemporal coverage XCO2 dataset is combined with the hybrid geographic weighted regression values ​​and the model XCO2 estimates of the weighted regression equation to achieve data matching and full coverage.

8. The CO2 column concentration monitoring method based on multi-source data and hybrid geographic weighted regression according to claim 5, characterized in that: The weights between the spatial location i of the target sample point and its neighboring point j in the weight matrix The expression is as follows: ,in Let be the distance between the spatial location i of the target sample point and its neighboring point j, and b be the bandwidth parameter.

9. The CO2 column concentration monitoring method based on multi-source data and hybrid geographic weighted regression according to claim 1, characterized in that: It also includes the following methods: S4. Following method S2, obtain the estimated value of model XCO2 for spatial location i arranged in time series, and obtain XCO2 time series variation data; use the XCO2 time series variation data to visualize the monthly scale variation, seasonal variation, interannual variation and trend variation of XCO2.

Images