Datasets:

---

LLMDH

/

OpenScience

like 0

Follow

LLM - Digital Humanities 16

ERROR: type should be string, got "https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. Haowei Zhang1★, Boming Liu2★, Xin Ma2, Ge Han3, Qinglin Yang3, Yichi Zhang3, Tianqi Shi2, Jianye Yuan1, Wanqi Zhong2, \nYanran Peng1, Jingjing Xu1, Wei Gong1 1School of Electronic Information, Wuhan University, Wuhan 473072, China \n5 \n2State Key Laboratory of Information Engineering in Surveying, Mapping, and Remote Sensing, Wuhan University, Wuhan \n430079, China \n3School of Remote Sensing and Information Engineering, Wuhan University, Wuhan 430079, China \n★These authors contributed to the work equally and should be regarded as co-first authors. Correspondence to: XinMa (maxinwhu@whu.edu.cn) \n10 Correspondence to: XinMa (maxinwhu@whu.edu.cn) \n10 Abstract. For carbon dioxide concentration (XCO2) distribution, the improvement of spatial and temporal resolution is very \nimportant in some scientific studies (e.g., studies of the carbon cycle and assessment of carbon emissions based on top-down \ntheory). However, carbon sniffing satellites based on passive theory (e.g., Gosat-2, OCO-2, and OCO-3) are susceptible to \ncloud and aerosol interference when the data are captured. Therefore, the data collected by carbon sniffing satellites have Abstract. For carbon dioxide concentration (XCO2) distribution, the improvement of spatial and temporal resolution is very \nimportant in some scientific studies (e.g., studies of the carbon cycle and assessment of carbon emissions based on top-down \ntheory). However, carbon sniffing satellites based on passive theory (e.g., Gosat-2, OCO-2, and OCO-3) are susceptible to \ncloud and aerosol interference when the data are captured. Therefore, the data collected by carbon sniffing satellites have relatively low utilization, especially in some regions where data gaps exist. Here, we present the Carbon Dioxide Coverage \n15 \n(CDC) dataset, an innovative theory to obtain high spatial and temporal resolution maps of XCO2 distribution by combining \nspatial attributes and extracted temporal attributes from the GOSAT satellite series data. This theory is divided into the \nfollowing three parts. Firstly, several background values in the raw GOSAT data were removed through data pre-processing, \nand for spatial attributes, GOSAT satellite data gap areas were filled by combining adjacent GOSAT data and empirical 15 Bayesian kriging (EBK) theory in the study area. Secondly, for the temporal attributes, we constructed a time profile parameter \n20 \nlibrary, based on the GOSAT data of the time series to extract the temporal parameters from a specific formula at each point \nof the study area. Carbon dioxide cover: carbon dioxide column concentration \nseamlessly distributed globally during 2009–2020 Haowei Zhang1★, Boming Liu2★, Xin Ma2, Ge Han3, Qinglin Yang3, Yichi Zhang3, Tianqi Shi2, Jianye Yuan1, Wanqi Zhong2, \nYanran Peng1, Jingjing Xu1, Wei Gong1 Haowei Zhang1★, Boming Liu2★, Xin Ma2, Ge Han3, Qinglin Yang3, Yichi Zhang3, Tianqi Shi2, Jianye Yuan1, Wanqi Zhong2, \nYanran Peng1, Jingjing Xu1, Wei Gong1 Finally, for the integration of temporal and spatial information, based on the GOSAT satellite data and the \npopulated data based on spatial attributes, we assign the temporal parameter information from the time parameter library to \neach pixel location in the study area, combining the transfer component analysis (TCA) theory, and then combine the assigned Bayesian kriging (EBK) theory in the study area. Secondly, for the temporal attributes, we constructed a time profile parameter \n20 \nlibrary, based on the GOSAT data of the time series to extract the temporal parameters from a specific formula at each point \nof the study area. Finally, for the integration of temporal and spatial information, based on the GOSAT satellite data and the \npopulated data based on spatial attributes, we assign the temporal parameter information from the time parameter library to \neach pixel location in the study area, combining the transfer component analysis (TCA) theory, and then combine the assigned parameters with specific formulas to complete the prediction of XCO2 distribution. For temporal resolution, both the \n25 \nGOSAT_FTS_L3_V2.95 and CDC datasets are monthly-averaged resolution datasets from 2010 to 2020. And for spatial \nresolution, the CDC dataset is 0.25° resolution with a significant improvement compared to GOSAT_FTS_L3_V2.95 which \nis 2.5° resolution. And the dataset contained 136 files. Besides, for the data validation part, we used OCO-2 satellite data from \n2009 to 2020 and TCCON data at mid and low latitudes, respectively. This CDC dataset and the original data from the TCCON sites were compared on a monthly-averaged scale. And the results showed that R2 was 0.9686, and RMSE was 1.3811 ppm. 30 \nWe also derived statistical monthly averaged XCO2 from OCO-2 data and compared it with the data set from our theory. And \nour evaluation index R was greater than 0.7, by comparison with OCO-2 during 2014-2020. Finally, to assess the accuracy of sites were compared on a monthly-averaged scale. And the results showed that R2 was 0.9686, and RMSE was 1.3811 ppm. 30 \nWe also derived statistical monthly averaged XCO2 from OCO-2 data and compared it with the data set from our theory. And \nour evaluation index R was greater than 0.7, by comparison with OCO-2 during 2014-2020. Finally, to assess the accuracy of 1 1 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. the algorithm, we compared the predicted results with the input data for the period of 2009-2020. Haowei Zhang1★, Boming Liu2★, Xin Ma2, Ge Han3, Qinglin Yang3, Yichi Zhang3, Tianqi Shi2, Jianye Yuan1, Wanqi Zhong2, \nYanran Peng1, Jingjing Xu1, Wei Gong1 And the comparison results \nshow that the mean value of R2 is 0.93 and the mean value of RMSE is 0.53 ppm during 2010-2020. Data gaps produced by \nsniffer satellites are disturbed by factors such as clouds and aerosols and can be filled by this mapping technique is mentioned \nin this paper. This technique improves the utilization of XCO2 and the accuracy and resolution of the CDC dataset is sufficient \nfor scientific applications. And the CDC dataset is publicly available at https://doi.org/10.6084/m9.figshare.17826404.v4 \n(Zhang et al., 2022), which is of significance for a multitude of scientific carbon research. the algorithm, we compared the predicted results with the input data for the period of 2009-2020. And the comparison results \nshow that the mean value of R2 is 0.93 and the mean value of RMSE is 0.53 ppm during 2010-2020. Data gaps produced by \nsniffer satellites are disturbed by factors such as clouds and aerosols and can be filled by this mapping technique is mentioned \n5 \nin this paper. This technique improves the utilization of XCO2 and the accuracy and resolution of the CDC dataset is sufficient \nfor scientific applications. And the CDC dataset is publicly available at https://doi.org/10.6084/m9.figshare.17826404.v4 \n(Zhang et al., 2022), which is of significance for a multitude of scientific carbon research. 35 1 Introduction Global climate change exerts increased risks and impacts on natural and human life, such as rising sea levels, heat waves, \n40 \nfloods and droughts, erosion of food security, and slowing economic growth (Field et al., 2014; Diaz et al., 2017). The \nskyrocketing level of greenhouse gas in recent decades is the main cause of global climate change (Black et al., 2011). Therefore, monitoring the changes in spatiotemporal carbon dioxide concentration (XCO2) in the global atmosphere is crucial. The sniffer satellites in orbit at present are carrying passive detectors (Basilio et al., 2014; Nakajima et al., 2017), and the quality of the data collected is limited by several factors, such as cloud cover, lack of observations in high-latitude areas and \n45 \nat night, and sensitivity to aerosols. Therefore, the acquisition of spatio-temporal maps of XCO2 distribution with high accuracy \nand resolution is essential to facilitate the study of the carbon cycle, carbon sources, carbon sinks, carbon neutrality, and carbon \nemissions assessed through top-down theory. Scientists have conducted downscaling studies on carbon detection series satellite data. Tomasada et al. (2009;2008) and Liu et al. (2012) generated monthly-averaged CO2 distribution maps by using ordinary kriging interpolation of GOSAT Level 2 \n50 \n(L2) products. Hammerling et al. (2012) obtained CO2 maps mainly by processing simulated satellite observations by using a \nmoving kriging window. Mueller et al. (2008) reconstructed global monthly-averaged CO2 fluxes from ground observations \nby using the geostatistical inverse modeling theory. Moreover, Katzfuss et al. (2011;2012) completed spatiotemporal \nsmoothing of global XCO2 data, the theory of which focused on a fully Bayesian hierarchical approach. Zeng et al. (2013) proposed a spatiotemporal kriging theory, applied it to model GOSAT data in China, and obtained the monthly-averaged \n55 \ndistribution of XCO2. The interpolation method commonly used for satellite XCO2 observations is the conventional geostatistical spatial prediction \nmethod, which considers spatial autocorrelation only (Tomosada et al. 2009; Tomosada et al. 2008; Liu et al. 2012). This \nmethod requires a long time series of data so as to ensure sufficient data for stable variometric estimations, but it ignores the time structure in the data. In addition, on the basis of spatial interpolation, several scholars further integrated time information \n60 \ninto the interpolation method and obtained good results (Zeng et al. 2013; Yang et al. 2020; Gribov et al. 2012; Ma et al. 2021). 1 Introduction Although these methods produce good results from a mathematical point of view, in studies that utilized these methods, the \nprior time profile information of XCO2 was rarely considered, resulting in insufficient adjustment of the temporal information, \nas reflected by the large differences between the monthly-averaged XCO2 and the true value. Therefore, in this study, we 2 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. integrate the prior information of the original data into a new spatiotemporal interpolation theory that considers the time \n65 \nvariation of concentration distribution to effectively improve data accuracy. In other words, we propose a new method to \nimprove the utilization of XCO2 data. First, several background values in the raw GOSAT data were removed through data \npre-processing, and for spatial attributes, GOSAT satellite data gap areas were filled by combining adjacent GOSAT data and \nempirical Bayesian kriging (EBK) theory in the study area. Secondly, for the temporal attributes, we constructed a time profile integrate the prior information of the original data into a new spatiotemporal interpolation theory that considers the time \n65 \nvariation of concentration distribution to effectively improve data accuracy. In other words, we propose a new method to \nimprove the utilization of XCO2 data. First, several background values in the raw GOSAT data were removed through data \npre-processing, and for spatial attributes, GOSAT satellite data gap areas were filled by combining adjacent GOSAT data and \nempirical Bayesian kriging (EBK) theory in the study area. Secondly, for the temporal attributes, we constructed a time profile empirical Bayesian kriging (EBK) theory in the study area. Secondly, for the temporal attributes, we constructed a time profile \nparameter library, based on the GOSAT data of the time series to extract the temporal parameters from a specific formula at \n70 \neach point of the study area. Finally, for the integration of temporal and spatial information, based on the GOSAT satellite \ndata and the populated data based on spatial attributes, we assign the temporal parameter information from the time parameter \nlibrary to each pixel location in the study area, combining the transfer component analysis (TCA) theory, and then combine \nthe assigned parameters with specific formulas to complete the prediction of XCO2 distribution. parameter library, based on the GOSAT data of the time series to extract the temporal parameters from a specific formula at \n70 \neach point of the study area. 1 Introduction Finally, for the integration of temporal and spatial information, based on the GOSAT satellite \ndata and the populated data based on spatial attributes, we assign the temporal parameter information from the time parameter \nlibrary to each pixel location in the study area, combining the transfer component analysis (TCA) theory, and then combine \nthe assigned parameters with specific formulas to complete the prediction of XCO2 distribution. The focus of this work is to provide a global dataset of the monthly-averaged XCO2 at 0.25° based on the theory presented in \n75 \nthe paper and the discrete XCO2 measured by the GOSAT satellite. The CDC dataset extends from 2009 to 2020 and from 50° \nS to 50° N. The validation of the CDC dataset will be performed by comparing it with those from OCO-2, TCCON, and the \ninput GOSAT dataset (which was not involved in the generation of the CDC dataset). Namely, the accuracy validation of the \nCDC dataset is divided into the following parts in this paper. First, based on the theory proposed in this work and the The focus of this work is to provide a global dataset of the monthly-averaged XCO2 at 0.25° based on the theory presented in \n75 \nthe paper and the discrete XCO2 measured by the GOSAT satellite. The CDC dataset extends from 2009 to 2020 and from 50° \nS to 50° N. The validation of the CDC dataset will be performed by comparing it with those from OCO-2, TCCON, and the \ninput GOSAT dataset (which was not involved in the generation of the CDC dataset). Namely, the accuracy validation of the \nCDC dataset is divided into the following parts in this paper. First, based on the theory proposed in this work and the GOSAT_L3 data, we compare the spatio-temporal prediction data generated in each TCCON site with the data from the \n80 \ncorresponding TCCON site. Second, we derived statistical monthly-averaged XCO2 from OCO-2 data and compared it with \nthe data set from our theory. Finally, to assess the accuracy of the algorithm, we compared the results of the model predictions \nwith the input data for the period 2009-2020. 1 Introduction The advantages of the global CDC dataset are (1) its large spatial coverage (From approximately 55° S to 55° N with a The advantages of the global CDC dataset are (1) its large spatial coverage (From approximately 55 S to 55 N with a \nresolution of 0.25°) and (2) 12-year time series (Monthly-averaged XCO2 from 2009 to 2020). Thus, the CDC dataset can be \n85 \nused to study the global XCO2 at timescales ranging from seasons to decades and from cities to countries. Besides, the XCO2 \ndata calculated by the model presented in this paper can be input into the atmospheric chemical transport model and can also \ncontribute to the study of the carbon cycle. And the satellite data of global observations (such as OCO-2, OCO-3, GOSAT, \nGOSAT-2 and Tansat) have been widely used for the calculation of global carbon sources and sinks. Therefore, this technique resolution of 0.25°) and (2) 12-year time series (Monthly-averaged XCO2 from 2009 to 2020). Thus, the CDC dataset can be \n85 \nused to study the global XCO2 at timescales ranging from seasons to decades and from cities to countries. Besides, the XCO2 \ndata calculated by the model presented in this paper can be input into the atmospheric chemical transport model and can also \ncontribute to the study of the carbon cycle. And the satellite data of global observations (such as OCO-2, OCO-3, GOSAT, \nGOSAT-2 and Tansat) have been widely used for the calculation of global carbon sources and sinks. Therefore, this technique improves the utilization of XCO2 and the accuracy and resolution of the CDC dataset is sufficient for scientific applications. 90 \nConsidering that GOSAT can help to obtain XCO2 at the globle scale, whose data is used as the primary dataset in this work. It enables the development of strategies to reduce XCO2 at the global scale. The dataset and related codes are publicly available \nat https://doi.org/10.6084/m9.figshare.17826404.v4 (Zhang et al., 2022), which are of significance for a multitude of scientific \nresearch and applications. improves the utilization of XCO2 and the accuracy and resolution of the CDC dataset is sufficient for scientific applications. 90 \nConsidering that GOSAT can help to obtain XCO2 at the globle scale, whose data is used as the primary dataset in this work. It enables the development of strategies to reduce XCO2 at the global scale. 1 Introduction The dataset and related codes are publicly available \nat https://doi.org/10.6084/m9.figshare.17826404.v4 (Zhang et al., 2022), which are of significance for a multitude of scientific \nresearch and applications. 3 3 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. 2.1 Data description The time span of GOSAT satellite data (2009–2020) is longer than that of OCO-2, OCO-3, and Tansat. Thus, we selected the \nbias-corrected data of GOSAT_FTS_L3_V2.95. And the accuracy of the comparison between the GOSAT data product and \nthe TCCON site was 0.56 ppm (Noël et al. 2021; Watanabe et al. 2015;). And the GOSAT orbits at an altitude of approximately pp\n(\n;\n;)\npp\ny\n666 km, with 10.5 km of spatial resolution and three-day temporal resolution. The time resolution of GOSAT-2 satellite is 6 \n100 \ndays, IFOV is 9.7km. The GOSAT and GOSAT-2 satellites have been operational since 2009 and 2018, respectively, and the data collected by the \nGOSAT-1/2 satellites have the potential to reveal new information on the carbon cycle. Studies of the carbon cycle have been \ncarried out based on atmospheric chemistry models. Such models usually require input of measured XCO2 data to constrain 666 km, with 10.5 km of spatial resolution and three-day temporal resolution. The time resolution of GOSAT-2 satellite is 6 \n100 \ndays, IFOV is 9.7km. The GOSAT and GOSAT-2 satellites have been operational since 2009 and 2018, respectively, and the data collected by the \nGOSAT-1/2 satellites have the potential to reveal new information on the carbon cycle. Studies of the carbon cycle have been \ncarried out based on atmospheric chemistry models. Such models usually require input of measured XCO2 data to constrain the atmospheric chemistry model. However, the OCO-2_L2_Lite_FP9r provides data locations that are gradually shifted over \n105 \ntime by satellite observations. And the GOSAT_L3 product only provides a long time series of cumulative observations for a \nfixed location, thus large vacant data areas exist in the global for the GOSAT_L3 product. Our proposed monthly-averaged \nXCO2 map can complete the carbon cycle input on a large scale spatially and over a long time series. Therefore, our monthly-\naveraged XCO2 map is helpful for carbon cycle studies. Because the six data channels of the sensor carried by the GOSAT the atmospheric chemistry model. However, the OCO-2_L2_Lite_FP9r provides data locations that are gradually shifted over \n105 \ntime by satellite observations. And the GOSAT_L3 product only provides a long time series of cumulative observations for a \nfixed location, thus large vacant data areas exist in the global for the GOSAT_L3 product. Our proposed monthly-averaged \nXCO2 map can complete the carbon cycle input on a large scale spatially and over a long time series. 2.2 Validation data \n115 2.2 Validation data \n115 \nTo evaluate the accuracy of the monthly-averaged XCO2 data from our algorithm, we used global data of the Total Carbon \nColumn Observing Network (TCCON) during 2009-2020. TCCON ( Iraci et al. 2017; Dubey et al. 2017; Wennberg et al. 2017; Dubey et al. 2017; Blumenstock et al. 2017; Feist et al. 2017; Warneke et al. 2017; Sussmann et al. 2017; Sussmann et \nal. 2017; Petri et al. 2017; Maziere et al. 2017; Morino et al. 2017; Goo et al. 2017; Shiomi et al. 2017; Morino et al. 2017; To evaluate the accuracy of the monthly-averaged XCO2 data from our algorithm, we used global data of the Total Carbon \nColumn Observing Network (TCCON) during 2009-2020. TCCON ( Iraci et al. 2017; Dubey et al. 2017; Wennberg et al. 2017; Dubey et al. 2017; Blumenstock et al. 2017; Feist et al. 2017; Warneke et al. 2017; Sussmann et al. 2017; Sussmann et \nal. 2017; Petri et al. 2017; Maziere et al. 2017; Morino et al. 2017; Goo et al. 2017; Shiomi et al. 2017; Morino et al. 2017; To evaluate the accuracy of the monthly-averaged XCO2 data from our algorithm, we used global data of the Total Carbon \nColumn Observing Network (TCCON) during 2009-2020. TCCON ( Iraci et al. 2017; Dubey et al. 2017; Wennberg et al. 2017; Dubey et al. 2017; Blumenstock et al. 2017; Feist et al. 2017; Warneke et al. 2017; Sussmann et al. 2017; Sussmann et \nal. 2017; Petri et al. 2017; Maziere et al. 2017; Morino et al. 2017; Goo et al. 2017; Shiomi et al. 2017; Morino et al. 2017; Morino et al. 2017; Griffith et al. 2017; Pollard et al. 2017; Sherlock et al. 2017; Liu et al. 2017; Wennberg et al. 2017; \n120 \nWennberg et al. 2017;) is composed of ground-based Fourier transform spectrometers that record direct solar spectra in the \nnear-infrared spectral region. And we show the global distribution of TCCON sites in Figure 1. The spectrometer used in \nTCCON can provide accurate and precise column-averaged abundances of CO2. And the results showed that R2 was 0.9686, \nand RMSE was 1.3811. We also collected the original XCO2 from OCO-2 for comparison and to obtain abundant observations Morino et al. 2017; Griffith et al. 2017; Pollard et al. 2017; Sherlock et al. 2017; Liu et al. 2017; Wennberg et al. 2017; \n120 \nWennberg et al. 2.1 Data description Therefore, our monthly-\naveraged XCO2 map is helpful for carbon cycle studies. Because the six data channels of the sensor carried by the GOSAT satellite operate in the near-infrared part of the solar spectrum, the GOSAT satellite cannot collect data when the Earth reflects \n110 \nlittle sunlight, such as in polar regions during winter. For additional instrument’s information, readers may refer to \nhttp://www.gosat.nies.go.jp/en/about_2_observe.html. Furthermore, the GOSAT-1/2 satellite provides column-averaged \nXCO2 by measuring the spectrum reflected by sunlight in the infrared region over a global scale. However, the interference of \nclouds and aerosols offen results in a sparse spatiotemporal coverage for XCO2 products of GOSAT. satellite operate in the near-infrared part of the solar spectrum, the GOSAT satellite cannot collect data when the Earth reflects \n110 \nlittle sunlight, such as in polar regions during winter. For additional instrument’s information, readers may refer to \nhttp://www.gosat.nies.go.jp/en/about_2_observe.html. Furthermore, the GOSAT-1/2 satellite provides column-averaged \nXCO2 by measuring the spectrum reflected by sunlight in the infrared region over a global scale. However, the interference of \nclouds and aerosols offen results in a sparse spatiotemporal coverage for XCO2 products of GOSAT. satellite operate in the near-infrared part of the solar spectrum, the GOSAT satellite cannot collect data when the Earth reflects \n110 \nlittle sunlight, such as in polar regions during winter. For additional instrument’s information, readers may refer to \nhttp://www.gosat.nies.go.jp/en/about_2_observe.html. Furthermore, the GOSAT-1/2 satellite provides column-averaged \nXCO2 by measuring the spectrum reflected by sunlight in the infrared region over a global scale. However, the interference of \nclouds and aerosols offen results in a sparse spatiotemporal coverage for XCO2 products of GOSAT. 2.3.1 Spatial Prediction Through EBK Theory \n135 Therefore, we choose EBK \ninterpolation method as the data processing method. interpolated region, and use single variogram to predict the value of the unknown location. But the EBK method estimates the \n145 \nerror of the semi-variograms. And the EBK theory will be more accurate compared to other kriging theories because it takes \ninto account the uncertainty involved in the estimation of the semi-variogram (Pilz et al.2007). Therefore, we choose EBK \ninterpolation method as the data processing method. 2.3.1 Spatial Prediction Through EBK Theory \n135 To obtain the distributions of monthly-averaged XCO2 in the study area, we performed monthly-averaged calculations on the \nraw GOSAT_L3 data on the basis of a 0.25° grid for each month separately. Then, we used the existing EBK method to fill \nthe areas not covered effectively and reasonably by the GOSAT data in each month. EBK can automatically perform the most \ndifficult steps in the process of building an effective kriging model (Gribov et al. 2012; Krivoruchko et al.). The EBK can To obtain the distributions of monthly-averaged XCO2 in the study area, we performed monthly-averaged calculations on the \nraw GOSAT_L3 data on the basis of a 0.25° grid for each month separately. Then, we used the existing EBK method to fill \nthe areas not covered effectively and reasonably by the GOSAT data in each month. EBK can automatically perform the most \ndifficult steps in the process of building an effective kriging model (Gribov et al. 2012; Krivoruchko et al.). The EBK can automatically calculate parameters through the process of constructing subsets and simulations, while the same type of Kriging \n140 \ninterpolation requires manual adjustment of parameters to receive accurate results (Krivoruchko et al. 2012; Krivoruchko et \nal.2019). And weighted least squares is used to estimate the semi-variogram in the same type kriging interpolation method, \nbut the parameters in EBK are estimated by the limited maximum likelihood method. And the EBK method differs from other \nkriging methods in that other kriging methods assume that the estimated semi-variograms is the true semi-variograms of the automatically calculate parameters through the process of constructing subsets and simulations, while the same type of Kriging \n140 \ninterpolation requires manual adjustment of parameters to receive accurate results (Krivoruchko et al. 2012; Krivoruchko et \nal.2019). And weighted least squares is used to estimate the semi-variogram in the same type kriging interpolation method, \nbut the parameters in EBK are estimated by the limited maximum likelihood method. And the EBK method differs from other \nkriging methods in that other kriging methods assume that the estimated semi-variograms is the true semi-variograms of the interpolated region, and use single variogram to predict the value of the unknown location. But the EBK method estimates the \n145 \nerror of the semi-variograms. And the EBK theory will be more accurate compared to other kriging theories because it takes \ninto account the uncertainty involved in the estimation of the semi-variogram (Pilz et al.2007). 2.2 Validation data \n115 2017;) is composed of ground-based Fourier transform spectrometers that record direct solar spectra in the \nnear-infrared spectral region. And we show the global distribution of TCCON sites in Figure 1. The spectrometer used in \nTCCON can provide accurate and precise column-averaged abundances of CO2. And the results showed that R2 was 0.9686, \nand RMSE was 1.3811. We also collected the original XCO2 from OCO-2 for comparison and to obtain abundant observations of XCO2 from OCO-2 in a large range. And our evaluation index R was greater than 0.7, by comparison with OCO-2 during \n125 \n2014-2020. of XCO2 from OCO-2 in a large range. And our evaluation index R was greater than 0.7, by comparison with OCO-2 during \n125 \n2014-2020. 4 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. 2.3 Theoretical framework The framework in Figure 2 depicts the general methodology. The method is divided into three parts: Spatial Prediction Through \nEBK Theory, Prior Time Curve Parameter Library, and Integration of Temporal Attributes through TCA Theory, respectively. In the Spatial Prediction Through EBK Theory section, the XCO2 gaps are filled in the spatial attributes through EBK Theory. 30 \nIn the Prior Time Curve Parameter Library section a time profile parameter library is constructed to express the temporal The framework in Figure 2 depicts the general methodology. The method is divided into three parts: Spatial Prediction Through \nEBK Theory, Prior Time Curve Parameter Library, and Integration of Temporal Attributes through TCA Theory, respectively. The framework in Figure 2 depicts the general methodology. The method is divided into three parts: Spatial Prediction Through \nEBK Theory, Prior Time Curve Parameter Library, and Integration of Temporal Attributes through TCA Theory, respectively. In the Spatial Prediction Through EBK Theory section, the XCO2 gaps are filled in the spatial attributes through EBK Theory. 130 \nIn the Prior Time Curve Parameter Library section, a time profile parameter library is constructed to express the temporal \nattributes. In the Integration of Temporal Attributes through TCA Theory section, temporal and spatial information is \nintegrated based on TCA theory, and then combine the assigned parameters with specific formulas to complete the prediction \nof XCO2 distribution. In the Spatial Prediction Through EBK Theory section, the XCO2 gaps are filled in the spatial attributes through EBK Theory. 130 \nIn the Prior Time Curve Parameter Library section, a time profile parameter library is constructed to express the temporal \nattributes. In the Integration of Temporal Attributes through TCA Theory section, temporal and spatial information is \nintegrated based on TCA theory, and then combine the assigned parameters with specific formulas to complete the prediction \nof XCO2 distribution. 2.3.2 Prior Time Curve Parameter Library 𝐹(𝑡) = 𝑎+ 𝑏∗𝑡+ 𝑐∗cos /\n!\"#\n$ 0 + 𝑑∗sin /\n!\"#\n$ 0 + 𝑒∗cos /\n!\"#\n$ 0 + 𝑔∗sin /\n!\"#\n$ 0 , \n \n \n (1) 𝐹(𝑡) = 𝑎+ 𝑏∗𝑡+ 𝑐∗cos /\n!\"#\n$ 0 + 𝑑∗sin /\n!\"#\n$ 0 + 𝑒∗cos /\n!\"#\n$ 0 + 𝑔∗sin /\n!\"#\n$ 0 , 160 (1) (1) where 𝑎 refers to the yearly averaged XCO2; 𝑐 , 𝑑, 𝑒, and 𝑔 are the coefficients of the seasonal component; 𝑏 is the coefficient \nof the interannual component; 𝑓 is the sampling frequency (𝑓 = 12 for a year); and 𝑡 is the sampling interval. 2.3.3 Integration of Temporal Attributes through TCA Theory To fill the XCO2 gap region, we used the EBK theory based on spatial attributes in Section 2.3.1, and constructed a time profile \nparameter library based on temporal attributes in Section 2.3.2. But the problem is: how to merge the parameters representing \n165 \ntemporal attributes in the time profile parameter library with the filled XCO2 gaps based on spatial attributes (namely, the EBK \ntheory)? The TCA theory solves the allocation problem of parameters 𝑏 and 𝑐, which represent the XCO2 time profile of the \nwhole research area in the time curve parameter library. First, TCA assumes the same conditional distribution of source and \ntarget domains. second, maps the data into high-dimensional reproducing kernel Hilbert space, and then uses maximum mean discrepancy to find a mapping matrix that minimizes the marginal distribution between different domains to increase the source \n170 \ndomain the similarity with the target domain. finally, use the data of the source and target domains and the mapping matrix to \ntrain the classifier and complete the labeling of the target domain. The core of TCA is to find a mapping matrix that satisfies \nthe conditions (Dong et al. 2021; Pan et al. 2010; Dong et al. 2020; Dong et al. 2017). In this study, for the spatial point locations corresponding to the temporal profile parameter library, the fitted data are set as the source domain based on Eq (1). And, the spatial interpolation data are set as the target domain based on EBK theory in \n175 \nthe study area. Thus, each temporal profile was distributed from the source domain to the corresponding target domain based \non TCA theory transfer learning. Each pixel was again fitted based on Equation 1, combined with the time-adjusted parameters \n𝑐 , 𝑑, 𝑒, and 𝑔, assigned by TCA theory in the target domain from the temporal profile parameter library, in order to obtain the \nremaining parameters 𝑎 and 𝑏. And the final fitted data represent the spatio-temporal interpolation data. the source domain based on Eq (1). And, the spatial interpolation data are set as the target domain based on EBK theory in \n175 \nthe study area. Thus, each temporal profile was distributed from the source domain to the corresponding target domain based \non TCA theory transfer learning. 2.3.2 Prior Time Curve Parameter Library To fill the region of data gaps in space, we use EBK theory in Section 2.3.1. However, this EBK theory only considers the \n150 \nadjacent XCO2 data in the current month at the data gap location. Then, using only a theory based on spatial attributes to fill \nthe data gap locations would have a problem: the relationship between XCO2 data at adjacent times is cut off from a continuous \ntime scale. Thus, based on EBK theory, data gap filling may result in the current month being anomalous relative to XCO2 at \nadjacent times. For this reason, we constructed a time profile parameter library, based on the GOSAT data of the time series to extract the \n155 \ntemporal parameters from a specific formula at each point of the study area. We used GOSAT_L3 data as the input to build \nthe time curve library because the data of GOSAT_L3 stably provide the monthly-averaged XCO2 data of successive months For this reason, we constructed a time profile parameter library, based on the GOSAT data of the time series to extract the \n155 \ntemporal parameters from a specific formula at each point of the study area. We used GOSAT_L3 data as the input to build \nthe time curve library because the data of GOSAT_L3 stably provide the monthly-averaged XCO2 data of successive months For this reason, we constructed a time profile parameter library, based on the GOSAT data of the time series to extract the \n155 \ntemporal parameters from a specific formula at each point of the study area. We used GOSAT_L3 data as the input to build \nthe time curve library because the data of GOSAT_L3 stably provide the monthly-averaged XCO2 data of successive months 5 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. at the global scale. Furthermore, we used Eq. (1) to express the time change of XCO2 and to fit the GOSAT_L3 data for \nobtaining the parameters a and b. at the global scale. Furthermore, we used Eq. (1) to express the time change of XCO2 and to fit the GOSAT_L3 data for \nobtaining the parameters a and b. 2.3.3 Integration of Temporal Attributes through TCA Theory Each pixel was again fitted based on Equation 1, combined with the time-adjusted parameters \n𝑐 , 𝑑, 𝑒, and 𝑔, assigned by TCA theory in the target domain from the temporal profile parameter library, in order to obtain the \nremaining parameters 𝑎 and 𝑏. And the final fitted data represent the spatio-temporal interpolation data. 2.4 Accuracy Assessment \n180 \nThe accuracy verification process was divided into three main parts. First, this CDC dataset and the original data from the \nTCCON sites were compared on a monthly-averaged scale. Second, we derived statistical monthly-averaged XCO2 from OCO-\n2 data and compared it with the data set from our theory. Finally, to assess the accuracy of the algorithm, we compared the \nresults of the model predictions with the input data for the period 2009-2020. 3.1 Evaluation using TCCON observations Figure \n3 shows the predicted XCO2 and XCO2 observations at 23 TCCON sites. Compared with other similar works, the overall \nevaluation metric RMSE for our product data was 1.38 and improved by 22.9 % (Li et al. 2022; Zeng et al. 2014), and the spatial resolution (0.25°) became more refined compared with the mainstream spatial resolution of 1°. The time span of the \n205 \ndata set is 12 years from 2009 to 2020. Therefore, our data set fully satisfies the calculations of carbon sources, sinks, and \nemissions in a long time series. spatial resolution (0.25°) became more refined compared with the mainstream spatial resolution of 1°. The time span of the \n205 \ndata set is 12 years from 2009 to 2020. Therefore, our data set fully satisfies the calculations of carbon sources, sinks, and \nemissions in a long time series. 3 Results 3 Results 2.4 Accuracy Assessment \n180 = 1 −\n)\n(+!,-!)\"\n#\n!$%\n)\n(+!,/)\"\n#\n!$%\n, \n \n \n \n \n (4) (3) 190 (4) where 𝑁 is the number of prediction locations, 𝑃' is the predicted value, and 𝑅' is the observed value. e 𝑁 is the number of prediction locations, 𝑃' is the predicted value, and 𝑅' is the observed value. where 𝑁 is the number of prediction locations, 𝑃' is the predicted value, and 𝑅' is the observed va 3.1 Evaluation using TCCON observations The monthly-average XCO2 distribution of the CDC dataset products from 2010 to 2020 is presented from Figure 7 to Figure The monthly-average XCO2 distribution of the CDC dataset products from 2010 to 2020 is pre 195 17. And the data dictionary corresponding to the products we show in Table 3. Considering the range of XCO2 predicted by \n195 \nthe algorithm, we matched the predicted monthly-averaged XCO2 from our algorithm with the measured monthly-averaged \nXCO2 from TCCON sites at low- and mid-latitudes at the global scale from 2009 to 2020. Then, we used two mathematical \nindicators (R2 and RMSE) to quantitatively evaluate our algorithm. Table 1 lists the statistics for the predicted XCO2 and \nTCCON-observed XCO2. Our algorithm’s R2 is above 0.95, and its RMSE is below 1.5 in most of the individual TCCON sites. We also comprehensively analyzed the predicted data in 24 TCCON sites. The results showed that R2 was 0.9686, and RMSE \n200 \nwas 1.3811. Pearson’s correlation coefficient was adopted to evaluate the relationship between the predicted XCO2 and the \nXCO2 from TCCON sites. We annotated P<0.01 in Table 1 to indicate that the data have a strong statistical correlation. Figure \n3 shows the predicted XCO2 and XCO2 observations at 23 TCCON sites. Compared with other similar works, the overall \nevaluation metric RMSE for our product data was 1.38 and improved by 22.9 % (Li et al. 2022; Zeng et al. 2014), and the We also comprehensively analyzed the predicted data in 24 TCCON sites. The results showed that R2 was 0.9686, and RMSE \n200 \nwas 1.3811. Pearson’s correlation coefficient was adopted to evaluate the relationship between the predicted XCO2 and the \nXCO2 from TCCON sites. We annotated P<0.01 in Table 1 to indicate that the data have a strong statistical correlation. Figure \n3 shows the predicted XCO2 and XCO2 observations at 23 TCCON sites. Compared with other similar works, the overall \nevaluation metric RMSE for our product data was 1.38 and improved by 22.9 % (Li et al. 2022; Zeng et al. 2014), and the We also comprehensively analyzed the predicted data in 24 TCCON sites. The results showed that R2 was 0.9686, and RMSE \n200 \nwas 1.3811. Pearson’s correlation coefficient was adopted to evaluate the relationship between the predicted XCO2 and the \nXCO2 from TCCON sites. We annotated P<0.01 in Table 1 to indicate that the data have a strong statistical correlation. 2.4 Accuracy Assessment \n180 The accuracy verification process was divided into three main parts. First, this CDC dataset and the original data from the \nTCCON sites were compared on a monthly-averaged scale. Second, we derived statistical monthly-averaged XCO2 from OCO-\n2 data and compared it with the data set from our theory. Finally, to assess the accuracy of the algorithm, we compared the \nresults of the model predictions with the input data for the period 2009-2020. To quantify the rationality of the proposed theory in this paper, the coefficient of determination (R2) and the root mean square \n185 \nerror (RMSE) are chosen in this manuscript. The R2 can be used to evaluate the linear correlation between the results and the \nactual values. The RMSE is used to evaluate the bias of the prediction. The RMSE and R2 can be defined as follows： To quantify the rationality of the proposed theory in this paper, the coefficient of determination (R2) and the root mean square \n185 \nerror (RMSE) are chosen in this manuscript. The R2 can be used to evaluate the linear correlation between the results and the \nactual values. The RMSE is used to evaluate the bias of the prediction. The RMSE and R2 can be defined as follows： To quantify the rationality of the proposed theory in this paper, the coefficient of determination (R2) and the root mean square \n185 \nerror (RMSE) are chosen in this manuscript. The R2 can be used to evaluate the linear correlation between the results and the \nactual values. The RMSE is used to evaluate the bias of the prediction. The RMSE and R2 can be defined as follows： 6 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. RMSE = =\n%\n& ∑\n|𝑃' −𝑅'|! &\n'(%\n, \n \n \n \n \n (2) \n𝑦=\n%\n& ∑\n𝑃'\n&\n'(%\n, \n \n \n \n \n \n \n (3) \n𝑅! = 1 −\n)\n(+!,-!)\"\n#\n!$%\n)\n(+!,/)\"\n#\n!$%\n, \n \n \n \n \n (4) RMSE = =\n%\n& ∑\n|𝑃' −𝑅'|! &\n'(%\n, \n \n \n \n \n (2) \n𝑦=\n%\n& ∑\n𝑃'\n&\n'(%\n, \n \n \n \n \n \n \n (3) \n𝑅! = 1 −\n)\n(+!,-!)\"\n#\n!$%\n)\n(+!,/)\"\n#\n!$%\n, \n \n \n \n \n (4) RMSE = =\n%\n& ∑\n|𝑃' −𝑅'|! &\n'(%\n, \n \n \n \n \n (2) \n𝑦=\n%\n& ∑\n𝑃'\n&\n'(%\n, \n \n \n \n \n \n \n (3) \n𝑅! 3.2 Evaluation using OCO-2 observations To evaluate the accuracy of the algorithm’s predicted data at the global scale, we considered a To evaluate the accuracy of the algorithm’s predicted data at the global scale, we considered another greenhouse gas satellite, \nOCO-2, from the United States. OCO-2 and GOSAT satellites are XCO2 monitoring satellites that use the passive inversion \n210 \nmode. Although the sensors onboard the two satellites are different, the data from both are a measure of XCO2 columns. Several scholars have compared XCO2 data from OCO-2 and GOSAT-2 and concluded that the observed data values of the \ntwo satellites are consistent and smooth (Liang et al. 2017). For these reasons, we selected measured XCO2 data from OCO-2 \nas a comparison for verification. By doing so, we can verify our products in a wider range and with more data than fixed OCO-2, from the United States. OCO-2 and GOSAT satellites are XCO2 monitoring satellites that use the passive inversion \n210 \nmode. Although the sensors onboard the two satellites are different, the data from both are a measure of XCO2 columns. Several scholars have compared XCO2 data from OCO-2 and GOSAT-2 and concluded that the observed data values of the \ntwo satellites are consistent and smooth (Liang et al. 2017). For these reasons, we selected measured XCO2 data from OCO-2 \nas a comparison for verification. By doing so, we can verify our products in a wider range and with more data than fixed TCCON sites. We removed bad data in accordance with the data quality label provided by OCO-2 and obtained the monthly- \n215 7 7 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. averaged XCO2 data through statistics. The statistical results showed that all R values were greater than 0.7, and a significant \ncorrelation was observed at the 0.01 level (Table 2). OCO-2 data services were opened and closed in 2014 and 2020, \nrespectively, and we could only obtain partial OCO-2 data. Therefore, our evaluation index (R) is relatively low in 2014 and \n2020 due to the insufficient data volume. Accordingly, a density scatter diagram of each year is drawn in Figure 4, and most 220 of the data are distributed on the 1:1 line. The color change from blue to red in Figure 4 indicates a gradual increase in data \n220 \noverlap. Furthermore, the comparison results distributed near the 1:1 line are high-density data in Figure 4. 3.2 Evaluation using OCO-2 observations The comparison \nof OCO-2 data during 2014–2020 revealed that our data have high accuracy and stability. We found multiple parts per million \ndeviations present between the OCO-2 and algorithm products in Figure 4, which is due to the difference in revisit period. Compared to the revisit period of 16 days for OCO-2, the repeat period of GOSAT-2 satellite is 6 days. Therefore, GOSAT will can sample more data than OCO-2 in a month's time. Besides, the official algorithms of OCO-2 and GOSAT-2 products \n225 \nare different, so the model results generated based on GOSAT-2 data will produce multiple parts per million deviations \ncompared to the OCO-2 product during 2015-2019 period. will can sample more data than OCO-2 in a month's time. Besides, the official algorithms of OCO-2 and GOSAT-2 products \n225 \nare different, so the model results generated based on GOSAT-2 data will produce multiple parts per million deviations \ncompared to the OCO-2 product during 2015-2019 period. 3.3 Evaluation using GOSAT_L3 observations Because satellite observations are missing in some years (e.g., 2009, 2014, and 2015), we need to \n240 \ncombine adjacent months to complete a continuous time period of data input. Therefore, the temporal structure of this data \ninput may have an impact on the accuracy of the model. January to December. Because satellite observations are missing in some years (e.g., 2009, 2014, and 2015), we need to \n240 \ncombine adjacent months to complete a continuous time period of data input. Therefore, the temporal structure of this data \ninput may have an impact on the accuracy of the model. January to December. Because satellite observations are missing in some years (e.g., 2009, 2014, and 2015), we need to \n240 \ncombine adjacent months to complete a continuous time period of data input. Therefore, the temporal structure of this data \ninput may have an impact on the accuracy of the model. 3.3 Evaluation using GOSAT_L3 observations To assess the accuracy of the algorithm, we compared the results of the model predictions To assess the accuracy of the algorithm, we compared the results of the model predictions with the input data for the period \n2009-2020. To validate evenly globally, we removed one column of GOSAT_L3 data for each 20° longitude interval. The \n230 \nremoved data will be used as the validation set for validation. And this R2 and RMSE are used as evaluation metrics to evaluate \nthe validation set and the predicted data. Besides, we show the validation results of the CDC dataset according to the year \ninterval in Figure 5 from 2009 to 2020. And the comparison results show that the mean value of R2 is 0.93 and the mean value \nof RMSE is 0.53 ppm during 2010-2020. This indicates that the accuracy of our data products is recognized from the GOSAT_L3 input data. In Figure 6, we show the errors for each year of predicted data. And, the fluctuations of the error bands \n235 \nshaded in Figure 6 are small, which indicates that the errors of the data set are in a stable state from 2010 to 2020. Because of \nthe data from June to December in 2009, the low precision metrics indicate that the model is not suitable for incomplete years. In general, products from our models can fill the vacant areas of XCO2 globally. As described in the theory section, our method \nrequired the input of 12 consecutive months of XCO2 data to make predictions, with the ideal data input period being from GOSAT_L3 input data. In Figure 6, we show the errors for each year of predicted data. And, the fluctuations of the error bands \n235 \nshaded in Figure 6 are small, which indicates that the errors of the data set are in a stable state from 2010 to 2020. Because of \nthe data from June to December in 2009, the low precision metrics indicate that the model is not suitable for incomplete years. In general, products from our models can fill the vacant areas of XCO2 globally. As described in the theory section, our method \nrequired the input of 12 consecutive months of XCO2 data to make predictions, with the ideal data input period being from January to December. 4 Data and code availability Version 3 of the CDC Database is available in h5 format at https://doi.org/10.6084/m Version 3 of the CDC Database is available in h5 format at https://doi.org/10.6084/m9.figshare.17826404.v4 (Zhang et al., \n2022). For the data extraction approach and the data dictionary of the CDC dataset, we provide the ReadMe.pdf file in the \n255 \nCDC dataset repository. For the introduction of data extraction in the ReadMe.pdf file, it contains Panoly and HDFView \nsoftware as well as read examples through python platform. TCCON dataset can be accessed https://tccondata.org/. Because \nthe CDC dataset range is covered at mid to low latitudes, we match the latitude range of the CDC dataset (namely, \napproximately from [55N, 55S]) with the corresponding TCOON site data on the TCCON website. Besides, the OCO-2 dataset \ncan be accessed https://disc.gsfc.nasa.gov/datasets?page=1&keywords=OCO-2. And, the OCO-2 data version number used in \n260 \nthe validation set is OCO2 L2 Lite FP 9r. The compressed code has been uploaded to the repository and the file name is p\ng\ng\n(\ng\n2022). For the data extraction approach and the data dictionary of the CDC dataset, we provide the ReadMe.pdf file in the \n255 \nCDC dataset repository. For the introduction of data extraction in the ReadMe.pdf file, it contains Panoly and HDFView \nsoftware as well as read examples through python platform. TCCON dataset can be accessed https://tccondata.org/. Because \nthe CDC dataset range is covered at mid to low latitudes, we match the latitude range of the CDC dataset (namely, \napproximately from [55N, 55S]) with the corresponding TCOON site data on the TCCON website. Besides, the OCO-2 dataset can be accessed https://disc.gsfc.nasa.gov/datasets?page=1&keywords=OCO-2. And, the OCO-2 data version number used in \n260 \nthe validation set is OCO2_L2_Lite_FP 9r. The compressed code has been uploaded to the repository and the file name is \nCode.zip in the RawDataAndCode folder of the data repository. can be accessed https://disc.gsfc.nasa.gov/datasets?page=1&keywords=OCO-2. And, the OCO-2 data version number used in \n260 \nthe validation set is OCO2_L2_Lite_FP 9r. The compressed code has been uploaded to the repository and the file name is \nCode.zip in the RawDataAndCode folder of the data repository. 3.4 Evaluating Dataset Uncertainty We divided the data uncertainty into three categories. This label '1' indicates that there are We divided the data uncertainty into three categories. This label '1' indicates that there are no satellite observations at the \nlocation where the label is located, and that the spatial and temporal properties of this location are adjusted. In other words, \n245 \nthe error of the data product at the position indicated by label '1' may be the largest in the three types of labels. This label '2' \nrepresents the presence of GOSAT-2_L3 observations at the location where the label is located, but with the temporal attribute 8 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. adjusted. That is, the error of the data product at the location indicated by label '2' may be small, and this kind of data has \nmedium error in the three types of labels. This label ‘3’represents the presence of GOSAT-2_L3 data from satellite observations \nat the location where the label is located, and the data from this location are used to build the data for the time profile library. That is, the data product error at the position indicated by label '3' is minimal. Finally, we added a data layer 'uncertainty' to \nshow the uncertainty in the latest dataset. 250 5 Conclusions In general, we obtained XCO2 based on GOSAT-2 data that can accurately fill the XCO2 gap region in the global through the \nmodel presented in this paper from 2009 to 2020. And for the data coverage, our data area mainly covers the middle and low \n280 \nlatitudes in the global. Besides, the XCO2 data calculated by the model presented in this paper can be input into the atmospheric \nchemical transport model and can also contribute to the study of the carbon cycle. And the satellite data of global observations \n(such as OCO-2, OCO-3, GOSAT, GOSAT-2 and Tansat) have been widely used for the calculation of global carbon sources \nand sinks. This mapping technique with high accuracy and resolution can fill the spatiotemporal gaps in satellite measurement, \nwhich can meet the needs of scientific applications. And the GOSAT is the primary dataset being used in this work. It enables \n285 \nthe development of strategies to reduce XCO2 at the global scale. of the algorithm, we compared the results of the model predictions with the input data for the period 2009-2020. And the \ncomparison results show that the mean value of R2 is 0.93 and the mean value of RMSE is 0.53 ppm during 2010-2020. In general, we obtained XCO2 based on GOSAT-2 data that can accurately fill the XCO2 gap region in the global through the \nmodel presented in this paper from 2009 to 2020. And for the data coverage, our data area mainly covers the middle and low \n280 \nlatitudes in the global. Besides, the XCO2 data calculated by the model presented in this paper can be input into the atmospheric \nchemical transport model and can also contribute to the study of the carbon cycle. And the satellite data of global observations \n(such as OCO-2, OCO-3, GOSAT, GOSAT-2 and Tansat) have been widely used for the calculation of global carbon sources \nand sinks. This mapping technique with high accuracy and resolution can fill the spatiotemporal gaps in satellite measurement, \nwhich can meet the needs of scientific applications. And the GOSAT is the primary dataset being used in this work. It enables \n285 \nthe development of strategies to reduce XCO2 at the global scale. 280 7 Author contributions HZ, XM and GH designed the research and developed the whole methodological framework; BL supervised the CDC dataset; \nQY, YZ and JY collects data for validation; HZ wrote the original draft of the manuscript; WZ, YP, JX and WG revised the \nmanuscript. 8 Competing interests The authors declare no competing interests. 9 Acknowledgements This work was supported by the National Natural Science Foundation of China (Grant No. 42171464, 41971283, 41801261, \n41827801 and 41801282), the National Key Research and Development Program of China (2017YFC0212600), The Key \n295 \nResearch and Development Project of Hubei Province (2021BCA216). The numerical calculations in this paper have been \ndone on the supercomputing system in the Supercomputing Center of Wuhan University. This work was supported by the National Natural Science Foundation of China (Grant No. 42 Basilio, R. R., Pollock, H., and Hunyadi-Lay, S. L.: OCO2 (Orbiting Carbon Observatory-2) mission operations planning and \ninitial operations experiences, Proc. SPIE. Sensors, Systems, and Next-Generation Satellites XVIII, 9241, 924105, (2014) \nBlack, R., Bennett, S. R. G., Tomas, S. M. & Beddington, J. R. Migration as adaptation. Nature 478, 447–449 (2011) 5 Conclusions In this paper, we propose a new method to improve the utilization of XCO2 data (as shown in Figure 7). First, several \nbackground values in the raw GOSAT data were removed through data pre-processing, and for spatial attributes, GOSAT \n265 \nsatellite data gap areas were filled by combining adjacent GOSAT data and empirical Bayesian kriging (EBK) theory in the \nstudy area. Secondly, for the temporal attributes, we constructed a time profile parameter library, based on the GOSAT data \nof the time series to extract the temporal parameters from a specific formula at each point of the study area. Finally, for the \nintegration of temporal and spatial information, based on the GOSAT satellite data and the populated data based on spatial attributes, we assign the temporal parameter information from the time parameter library to each pixel location in the study \n270 \narea, combining the transfer component analysis (TCA) theory, and then combine the assigned parameters with specific \nformulas to complete the prediction of XCO2 distribution. Besides, we evaluated the accuracy of the algorithm through three parts. First, this CDC dataset and the original data from the \nTCCON sites were compared on a monthly-averaged scale. And the results showed that R2 was 0.9686, and RMSE was 1.3811; Second, we derived statistical monthly-averaged XCO2 from OCO-2 data and compared it with the data set from our theory. 275 \nAnd our evaluation index R was greater than 0.7, by comparison with OCO-2 during 2014-2020; Finally, to assess the accuracy Second, we derived statistical monthly-averaged XCO2 from OCO-2 data and compared it with the data set from our theory. 275 \nAnd our evaluation index R was greater than 0.7, by comparison with OCO-2 during 2014-2020; Finally, to assess the accuracy 9 9 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. of the algorithm, we compared the results of the model predictions with the input data for the period 2009-2020. And the \ncomparison results show that the mean value of R2 is 0.93 and the mean value of RMSE is 0.53 ppm during 2010-2020. of the algorithm, we compared the results of the model predictions with the input data for the period 2009-2020. And the \ncomparison results show that the mean value of R2 is 0.93 and the mean value of RMSE is 0.53 ppm during 2010-2020. References Basilio, R. R., Pollock, H., and Hunyadi-Lay, S. L.: OCO2 (Orbiting Carbon Observatory-2) mission operations planning and \ninitial operations experiences, Proc. SPIE. Sensors, Systems, and Next-Generation Satellites XVIII, 9241, 924105, (2014) \nBlack, R., Bennett, S. R. G., Tomas, S. M. & Beddington, J. R. Migration as adaptation. Nature 478, 447–449 (2011) Basilio, R. R., Pollock, H., and Hunyadi-Lay, S. L.: OCO2 (Orbiting Carbon Observatory-2) mission operations planning and \ninitial operations experiences, Proc. SPIE. Sensors, Systems, and Next-Generation Satellites XVIII, 9241, 924105, (2014) \nBlack, R., Bennett, S. R. G., Tomas, S. M. & Beddington, J. R. Migration as adaptation. Nature 478, 447–449 (2011) 10 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. Blumenstock, T., F. Hase, M. Schneider, O.E. García, E. Sepúlveda. TCCON data from Izana, Tenerife, Spain, Release \nGGG2014R1. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) De Maziere, M., M. K. Sha, F. Desmet, C. Hermans, F. Scolas, N. Kumps, J.-M. Metzger, V. Duflot, J.-P. Cammas. TCCON \n305 \ndata from Reunion Island (La Reunion), France, Release GGG2014R0. TCCON data archive, hosted by CaltechDATA, \nCalifornia Institute of Technology, Pasadena, CA, U.S.A. (2017) \nDiaz, D. & Moore, F. Quantifying the economic risks of climate change. Nat. Clim. Change 7, 774–782 (2017) \nDong. Y, Shi. W, Du. B, Hu. X and Zhang. L, “Asymmetric Weighted Logistic Metric Learning for Hyperspectral Target De Maziere, M., M. K. Sha, F. Desmet, C. Hermans, F. Scolas, N. Kumps, J.-M. Metzger, V. Duflot, J.-P. Cammas. TCCON \n305 \ndata from Reunion Island (La Reunion), France, Release GGG2014R0. TCCON data archive, hosted by CaltechDATA, \nCalifornia Institute of Technology, Pasadena, CA, U.S.A. (2017) \nDi\nD & M\nF Q\ntif i\nth\ni\ni k\nf li\nt\nh\nN t Cli\nCh\n7 774 782 (2017) Diaz, D. & Moore, F. Quantifying the economic risks of climate change. Nat. Clim. Change 7, 774–782 (2017) Dong. Y, Shi. W, Du. B, Hu. X and Zhang. L, “Asymmetric Weighted Logistic Metric Learning for Hyperspectral Target \nDetection,” IEEE Transactions on Cybernetics, (2021) \n10 g\ng\ny\ng\ng\ng\nyp\np\ng\nDetection,” IEEE Transactions on Cybernetics, (2021) \n310 Detection,” IEEE Transactions on Cybernetics, (2021) \n310 \nDong.Y, Du. B, Zhang. L, and Zhang. L, “Dimensionality reduction and classification of hyperspectral images using ensemble \ndiscriminative local metric learning,” IEEE Trans. Geosci. Remote Sens., 55, no. 5, pp. 2509–2524, (2017) \nDong. References Y, Liang. T, Zhang. Y, and Du. B, “Spectral-spatial weighted kernel manifold embedded distribution alignment for \nremote sensing image classification,” IEEE Trans. Cybern., early access (2020) ,\ny\n, (\n)\nDong.Y, Du. B, Zhang. L, and Zhang. L, “Dimensionality reduction and classification of hyperspectral images using ensemble \ndiscriminative local metric learning,” IEEE Trans. Geosci. Remote Sens., 55, no. 5, pp. 2509–2524, (2017) \nDong. Y, Liang. T, Zhang. Y, and Du. B, “Spectral-spatial weighted kernel manifold embedded distribution alignment for Dong.Y, Du. B, Zhang. L, and Zhang. L, “Dimensionality reduction and classification of hyperspectral images using ensemble \ndiscriminative local metric learning,” IEEE Trans. Geosci. Remote Sens., 55, no. 5, pp. 2509–2524, (2017) Dong. Y, Liang. T, Zhang. Y, and Du. B, “Spectral-spatial weighted kernel manifold embedded distribution alignment for \nremote sensing image classification,” IEEE Trans. Cybern., early access (2020) ier, B. Henderson, D. Green, N. Allen, C. Roehl, J.-F. Blavier, Z. Butterfield, S. Love, J. Hamelman Dubey, M., R. Lindenmaier, B. Henderson, D. Green, N. Allen, C. Roehl, J.-F. Blavier, Z. Butterfield, S. Love, J. Hamelmann, \n315 \nD. Wunch. TCCON data from Four Corners, NM, USA, Release GGG2014R0. TCCON data archive, hosted by \nCaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) \nDubey, M., B. Henderson, D. Green, Z. Butterfield, G. Keppel-Aleks, N. Allen, J.-F. Blavier, C. Roehl, D. Wunch, R. Lindenmaier. TCCON data from Manaus, Brazil, Release GGG2014R0. TCCON data archive, hosted by CaltechDATA, Dubey, M., R. Lindenmaier, B. Henderson, D. Green, N. Allen, C. Roehl, J.-F. Blavier, Z. Butterfield, S. Love, J. Hamelmann, \n315 \nD. Wunch. TCCON data from Four Corners, NM, USA, Release GGG2014R0. TCCON data archive, hosted by \nCaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) D. Wunch. TCCON data from Four Corners, NM, USA, Release GGG2014R0. TCCON data archive, hosted by \nCaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) Dubey, M., B. Henderson, D. Green, Z. Butterfield, G. Keppel-Aleks, N. Allen, J.-F. Blavier, C. Roehl, D. Wunch, R. Lindenmaier. TCCON data from Manaus, Brazil, Release GGG2014R0. TCCON data archive, hosted by CaltechDATA, Dubey, M., B. Henderson, D. Green, Z. Butterfield, G. Keppel-Aleks, N. Allen, J.-F. Blavier, C. Roehl, D. Wunch, R. Lindenmaier. TCCON data from Manaus, Brazil, Release GGG2014R0. TCCON data archive, hosted by CaltechDATA, \nCalifornia Institute of Technology, Pasadena, CA, U.S.A. (2017) \n320 Lindenmaier. TCCON data from Manaus, Brazil, Release GGG2014R0. References TCCON data archive, hosted by CaltechDATA, \nCalifornia Institute of Technology, Pasadena, CA, U.S.A. (2017) \n320 California Institute of Technology, Pasadena, CA, U.S.A. (2017) \n320 \nFeist, D. G., S. G. Arnold, N. John, M. C. Geibel. TCCON data from Ascension Island, Saint Helena, Ascension and Tristan \nda Cunha, Release GGG2014R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, \nPasadena, CA, U.S.A. (2017) \nGoo, T.-Y., Y.-S. Oh, V. A. Velazco. TCCON data from Anmeyondo, South Korea, Release GGG2014R0. TCCON data Feist, D. G., S. G. Arnold, N. John, M. C. Geibel. TCCON data from Ascension Island, Saint Helena, Ascension and Tristan \nda Cunha, Release GGG2014R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, \nPasadena, CA, U.S.A. (2017) Pasadena, CA, U.S.A. (2017) \nGoo, T.-Y., Y.-S. Oh, V. A. Velazco. TCCON data from Anmeyondo, South Korea, Release GGG2014R0. TCCON data Goo, T.-Y., Y.-S. Oh, V. A. Velazco. TCCON data from Anmeyondo, South Korea, Release GGG2014R0. TCCON data \narchive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) ,\n,\n,\ny\n,\n,\narchive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) \n325 325 archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) \n325 \nGriffith, D. W. T., V. A. Velazco, N. Deutscher, C. Murphy, N. Jones, S. Wilson, R. Macatangay, G. Kettlewell, R. R. Buchholz, \nM. Riggenbach. TCCON data from Wollongong, Australia, Release GGG2014R0. TCCON data archive, hosted by \nCaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) \nGribov. A, and Krivoruchko. K, “New Flexible Non-Parametric Data Transformation for Trans-Gaussian Kriging,” Griffith, D. W. T., V. A. Velazco, N. Deutscher, C. Murphy, N. Jones, S. Wilson, R. Macatangay, G. Kettlewell, R. R. Buchholz, \nM. Riggenbach. TCCON data from Wollongong, Australia, Release GGG2014R0. TCCON data archive, hosted by \nCaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) Gribov. A, and Krivoruchko. K, “New Flexible Non-Parametric Data Transformation for Trans-Gaussian Kriging,” \nGeostatistics Oslo. 17, 51- 65 (2012) \n330 Geostatistics Oslo. 17, 51- 65 (2012) \n0 \nHammerling. D. M, Michalak. A. M, and Kawa. S. R, “Mapping of CO2 at high spatiotemporal resolution using satellite\nobservations: Global distributions from OCO-2,” J. Geophys. Res. 117, no. D6, D06306-1–D06306-10 (2012) \nIraci, L., J. Podolske, P. Hillyard, C. Roehl, P. O. Wennberg, J.-F. Blavier, J. Landeros, N. Allen, D. Wunch, J. Zavaleta, E. References 4, 430–446 (2011) \n340 \nK. Krivoruchko and A. Gribov, “Pragmatic Bayesian kriging for nonstationary and moderately non-Gaussian data,” Math. Planet Earth \nKrivoruchko, Konstantin. \"Empirical bayesian kriging.\" ArcUser Fall 6, no. 10 (2012) \nKrivoruchko K, Gribov A. Evaluation of empirical Bayesian kriging. Spatial Statistics., (2019) Katzfuss. M and Cressie. N, “Bayesian hierarchical spatio-temporal smoothing for very large datasets,” Environmentrics. 23, \nno. 1, 94–107 (2012) Katzfuss. M and Cressie. N, “Bayesian hierarchical spatio-temporal smoothing for very large datasets,” Environmentrics. 23, \nno. 1, 94–107 (2012) \nKatzfuss. M and Cressie. N, “Spatio-temporal smoothing and EM estimation for massive remote-sensing data sets,” J. Time \nSer. Anal. 32, no. 4, 430–446 (2011) \n340 Katzfuss. M and Cressie. N, “Spatio-temporal smoothing and EM estimation for massive remote-sensing data sets,” J. Time \nSer. Anal. 32, no. 4, 430–446 (2011) 340 Ser. Anal. 32, no. 4, 430–446 (2011) \n340 \nK. Krivoruchko and A. Gribov, “Pragmatic Bayesian kriging for nonstationary and moderately non-Gaussian data,” Math. Planet Earth \nK i\nhk\nK\ni\n\"E\ni i\nl b\ni\nk i i\n\" A U\nF ll 6\n10 (2012) K. Krivoruchko and A. Gribov, “Pragmatic Bayesian kriging for nonstationary and moderately non-Gaussian data,” Math. Planet Earth Krivoruchko, Konstantin. \"Empirical bayesian kriging.\" ArcUser Fall 6, no. 10 (2012) Krivoruchko, Konstantin. \"Empirical bayesian kriging.\" ArcUser Fall 6, no. 10 (2012) Krivoruchko K, Gribov A. Evaluation of empirical Bayesian kriging. Spatial Statistics., (2019) Krivoruchko K, Gribov A. Evaluation of empirical Bayesian kriging. Spatial Statistics., (2019) Field, C. B. Climate change 2014–Impacts, adaptation and vulnerability: Regional aspects. (Cambridge University Press, 2014) \n345 \nLi. J, Jia. K, Wei. X, Xia. M, and et al., “High-spatiotemporal resolution mapping of spatiotemporally continuous atmospheric \nCO2 concentrations over the global continent”. International Journal of Applied Earth Observation and Geoinformation, \n108, 102743, (2022) \nLiang. A, Gong. W, Han. G, and Xiang. C, “Comparison of satellite-observed XCO2 from GOSAT, OCO-2, and ground-based eld, C. B. Climate change 2014–Impacts, adaptation and vulnerability: Regional aspects. (Cambridg Field, C. B. Climate change 2014 Impacts, adaptation and vulnerability: Regional aspects. (Cambridge University Press, 2014) \n345 \nLi. J, Jia. K, Wei. X, Xia. M, and et al., “High-spatiotemporal resolution mapping of spatiotemporally continuous atmospheric \nCO2 concentrations over the global continent”. International Journal of Applied Earth Observation and Geoinformation, \n108, 102743, (2022) g\npp\n,\n108, 102743, (2022) \nLiang. A, Gong. W, Han. G, and Xiang. C, “Comparison of satellite-observed XCO2 from GOSAT, OCO-2, and ground-based Liang. References Q i l\nG O t\nR Alb t\nK D\nd\nH B\nd\nTCCON d t f\nA\nt\nFli ht R\nh C\nt 330 Geostatistics Oslo. 17, 51- 65 (2012) \n0 \nHammerling. D. M, Michalak. A. M, and Kawa. S. R, “Mapping of CO2 at high spatiotemporal resolution using satellite \nobservations: Global distributions from OCO-2,” J. Geophys. Res. 117, no. D6, D06306-1–D06306-10 (2012) Hammerling. D. M, Michalak. A. M, and Kawa. S. R, “Mapping of CO2 at high spatiotemporal resolution using satellite \nobservations: Global distributions from OCO-2,” J. Geophys. Res. 117, no. D6, D06306-1–D06306-10 (2012) Iraci, L., J. Podolske, P. Hillyard, C. Roehl, P. O. Wennberg, J.-F. Blavier, J. Landeros, N. Allen, D. Wunch, J. Zavaleta, E. Quigley, G. Osterman, R. Albertson, K. Dunwoody, H. Boyden. TCCON data from Armstrong Flight Research Center, 11 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. Edwards, CA, USA, Release GGG2014R1. TCCON data archive, hosted by CaltechDATA, California Institute of \n335 \nTechnology, Pasadena, CA, U.S.A. (2017) \nKatzfuss. M and Cressie. N, “Bayesian hierarchical spatio-temporal smoothing for very large datasets,” Environmentrics. 23, \nno. 1, 94–107 (2012) \nKatzfuss. M and Cressie. N, “Spatio-temporal smoothing and EM estimation for massive remote-sensing data sets,” J. Time \nSer. Anal. 32, no. 4, 430–446 (2011) \n340 Edwards, CA, USA, Release GGG2014R1. TCCON data archive, hosted by CaltechDATA, California Institute of \nTechnology, Pasadena, CA, U.S.A. (2017) Edwards, CA, USA, Release GGG2014R1. TCCON data archive, hosted by CaltechDATA, California Institute \n335 \nTechnology, Pasadena, CA, U.S.A. (2017) \nKatzfuss. M and Cressie. N, “Bayesian hierarchical spatio-temporal smoothing for very large datasets,” Environmentrics. 2\nno. 1, 94–107 (2012) \nKatzfuss. M and Cressie. N, “Spatio-temporal smoothing and EM estimation for massive remote-sensing data sets,” J. Tim\nSer. Anal. 32, no. 4, 430–446 (2011) \n340 \nK. Krivoruchko and A. Gribov, “Pragmatic Bayesian kriging for nonstationary and moderately non-Gaussian data,” Mat\nPlanet Earth \nKrivoruchko, Konstantin. \"Empirical bayesian kriging.\" ArcUser Fall 6, no. 10 (2012) \nKrivoruchko K, Gribov A. Evaluation of empirical Bayesian kriging. Spatial Statistics., (2019) 335 Edwards, CA, USA, Release GGG2014R1. TCCON data archive, hosted by CaltechDATA, California Institute of \n335 \nTechnology, Pasadena, CA, U.S.A. (2017) \nKatzfuss. M and Cressie. N, “Bayesian hierarchical spatio-temporal smoothing for very large datasets,” Environmentrics. 23, \nno. 1, 94–107 (2012) \nKatzfuss. M and Cressie. N, “Spatio-temporal smoothing and EM estimation for massive remote-sensing data sets,” J. Time \nSer. Anal. 32, no. References A, Gong. W, Han. G, and Xiang. C, “Comparison of satellite-observed XCO2 from GOSAT, OCO-2, and ground-based Liang. A, Gong. W, Han. G, and Xiang. C, “Comparison of satellite-observed XCO2 from GOSAT, OCO-2, and ground-based \nTCCON”. Remote Sensing, 9, 1033, (2017) \n350 Liang. A, Gong. W, Han. G, and Xiang. C, “Comparison of satellite-observed XCO2 from GOS TCCON”. Remote Sensing, 9, 1033, (2017) \n350 \nLiu C., Wang, W., Sun, Y. TCCON data from Hefei, China, Release GGG2014R0. TCCON data archive, hosted by \nCaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2010) \nLiu. Y, Wang. X, Guo. M, and Tani. H, “Mapping the TS SWIR L2 product of XCO2 and XCH4 data from the GOSAT by the \nkriging method—A case study in East Asia,” Int. J. Remote Sens. 33, no. 10, 3004–3025 (2012) Liu C., Wang, W., Sun, Y. TCCON data from Hefei, China, Release GGG2014R0. TCCON data archive, hosted by \nCaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2010) Liu. Y, Wang. X, Guo. M, and Tani. H, “Mapping the TS SWIR L2 product of XCO2 and XCH4 data from the GOSAT by the \nkriging method—A case study in East Asia,” Int. J. Remote Sens. 33, no. 10, 3004–3025 (2012) Ma, X., Zhang, H W., Han, G., Mao, F Y., et al., A Regional Spatiotemporal Downscaling Method for CO₂ Columns. IEEE \n355 \nTrans. Geosci. Remote. Sens. (2021) \nMueller. K. L, Gourdji. S. M, and Michalak. A. M, “Global monthly averaged CO2 fluxes recovered using a geostatistical Ma, X., Zhang, H W., Han, G., Mao, F Y., et al., A Regional Spatiotemporal Downscaling Method for CO₂ Columns. IEEE \n355 \nTrans. Geosci. Remote. Sens. (2021) \nMueller. K. L, Gourdji. S. M, and Michalak. A. M, “Global monthly averaged CO2 fluxes recovered using a geostatistical \ninverse modeling approach: 1. Results using atmospheric measurements,” J. Geophys. Res. 113, no. D21, D21114-1–\nD21114-15 (2008) Mueller. K. L, Gourdji. S. M, and Michalak. A. M, “Global monthly averaged CO2 fluxes recovered using a geostatistical \ninverse modeling approach: 1. Results using atmospheric measurements,” J. Geophys. Res. 113, no. D21, D21114-1–\nD21114-15 (2008) D21114-15 (2008) TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2020) Petri, C., C. Rousogenous, T. Warneke, M. Vrekoussis, S. Sciare, J. Notholt. TCCON data from Nicosia, Cyprus, Release \n370 \nGGG2014R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2020) \nPilz. J and Spöck. G, “Why do we need and how should we implement Bayesian kriging methods,” Stoch Environ Res Risk \nAssess.,22, pp. 621–632, (2007) Pilz. J and Spöck. G, “Why do we need and how should we implement Bayesian kriging methods,” Stoch Environ Res Risk \nAssess.,22, pp. 621–632, (2007) Pollard, D., J. Robinson, H. Shiona. 2019. TCCON data from Lauder, New Zealand, 125HR, Release GGG2014R0. TCCON \n375 \ndata archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) \nS. Jialin Pan and Q. Yang, “A survey on transfer learning,” IEEE Trans. Knowl. Data Eng., 22, no. 10, pp. 1345–1359, (2010) Pollard, D., J. Robinson, H. Shiona. 2019. TCCON data from Lauder, New Zealand, 125HR, Release GGG2014R0. TCCON \n375 \ndata archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) \nS. Jialin Pan and Q. Yang, “A survey on transfer learning,” IEEE Trans. Knowl. Data Eng., 22, no. 10, pp. 1345–1359, (2010) \nSussmann, R., M. Rettinger. TCCON data from Zugspitze, Germany, Release GGG2014R1. TCCON data archive, hosted by \nCaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) Pollard, D., J. Robinson, H. Shiona. 2019. TCCON data from Lauder, New Zealand, 125HR, Release GGG2014R0\n375 \ndata archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) \nS. Jialin Pan and Q. Yang, “A survey on transfer learning,” IEEE Trans. Knowl. Data Eng., 22, no. 10, pp. 1345–1359, (2010) \nSussmann, R., M. Rettinger. TCCON data from Zugspitze, Germany, Release GGG2014R1. TCCON data archive, hosted by \nCaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) S. Jialin Pan and Q. Yang, “A survey on transfer learning,” IEEE Trans. Knowl. Data Eng., 22, no. 10, pp. 1345–1359, (2010) Sussmann, R., M. Rettinger. TCCON data from Garmisch, Germany, Release GGG2014R2. TCCON data archive, hosted by \n380 \nCaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) Sussmann, R., M. Rettinger. TCCON data from Garmisch, Germany, Release GGG2014R2. D21114-15 (2008) Morino, I., V. A. Velazco, A. Hori, O. Uchino, D. W. T. Griffith. TCCON data from Burgos, Philippines, Release GGG2014R0. 360 Morino, I., V. A. Velazco, A. Hori, O. Uchino, D. W. T. Griffith. TCCON data from Burgos, Philippines, Release GGG2014R0. 360 \nTCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) \nMorino I T Matsuzaki A Shishime TCCON data from Tsukuba Ibaraki Japan 125HR Release GGG2014R2 TCCON Morino, I., V. A. Velazco, A. Hori, O. Uchino, D. W. T. Griffith. TCCON data from Burgos, Philippines, Release GGG2014R0. 360 \nTCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) \nMorino, I., T. Matsuzaki, A. Shishime. TCCON data from Tsukuba, Ibaraki, Japan, 125HR, Release GGG2014R2. TCCON \ndata archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) \nMorino, I., N. Yokozeki, T. Matzuzaki, A. Shishime. TCCON data from Rikubetsu, Hokkaido, Japan, Release GGG2014R2. ive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) Morino, I., T. Matsuzaki, A. Shishime. TCCON data from Tsukuba, Ibaraki, Japan, 125HR, Release GGG2014R2. TCCON \ndata archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) Morino, I., N. Yokozeki, T. Matzuzaki, A. Shishime. TCCON data from Rikubetsu, Hokkaido, Japan, Release GGG2014R2. d\nhi\nh\nd b\nl\nh\nli\ni\ni\nh\nl\nd\n(\n) Morino, I., N. Yokozeki, T. Matzuzaki, A. Shishime. TCCON data from Rikubetsu, Hokkaido Morino, I., N. Yokozeki, T. Matzuzaki, A. Shishime. TCCON data from Rikubetsu, Hokkaido, Japan, Release GGG2014R2. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) \n365 TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) \n365 \nNakajima. M, Hiroshi. S, Yotsumoto. K, Shiomi. K, Hirabayashi. T,\"Fourier transform spectrometer on GOSAT and GOSAT-\n2.\" International Conference on Space Optics—ICSO 2014. International Society for Optics and Photonics, 10563, (2017) 12 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. Noël, S., Reuter, M., Buchwitz, M., Borchardt, J., et al., XCO2 retrieval for GOSAT and GOSAT-2 based on the FOCAL \nalgorithm, Atmos. Meas. Tech., 14, 3837–3869, (2021) Petri, C., C. Rousogenous, T. Warneke, M. Vrekoussis, S. Sciare, J. Notholt. TCCON data from Nicosia, Cyprus, Release \n370 \nGGG2014R0. D21114-15 (2008) TCCON data archive, hosted by \n380 \nCaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2017) \nWarneke, T., J. Messerschmidt, J. Notholt, C. Weinzierl, N. Deutscher, C. Petri, P. Grupe, C. Vuillemin, F. Truong, M. Schmidt, \nM. Ramonet, E. Parmentier. TCCON data from Orleans, France, Release GGG2014R1. TCCON data archive, hosted by \nCaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. (2014) Shiomi, K., Kawakami, S., H. Ohyama, K. Arai, H. Okumura, C. Taura, T. Fukamachi, M. Sakashita. TCCON data from Saga, \n385 \nJapan, Release GGG2014R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, \nCA, U.S.A. (2017) \nSherlock, V., B. Connor, J. Robinson, H. Shiona, D. Smale, D. Pollard. 2017. TCCON data from Lauder, New Zealand, 125HR, \nRelease GGG2014R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, Shiomi, K., Kawakami, S., H. Ohyama, K. Arai, H. Okumura, C. Taura, T. Fukamachi, M. Sakashita. TCCON data from Saga, \n385 \nJapan, Release GGG2014R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, \nCA, U.S.A. (2017) Sherlock, V., B. Connor, J. Robinson, H. Shiona, D. Smale, D. Pollard. 2017. TCCON data from Lauder, New Zealand, 125HR, \nRelease GGG2014R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, \nU.S.A. (2017) \n390 U.S.A. (2017) \n390 \nTomosada. M, Kanefuji. K, Matsumoto. Y, and Tsubaki. H, “A Prediction method of the global distribution map of CO2 \ncolumn abundance retrieved from GOSAT observation derived from ordinary kriging,” in Proc. ICROS-SICE Int. Joint \nConf.4869–4873 (2009) \nTomosada. M, Kanefuji. K, Matsumoto. Y, and Tsubaki. H, “Application of the spatial statistics to the retrieved CO2 column (\n)\nTomosada. M, Kanefuji. K, Matsumoto. Y, and Tsubaki. H, “A Prediction method of the global distribution map of CO2 \ncolumn abundance retrieved from GOSAT observation derived from ordinary kriging,” in Proc. ICROS-SICE Int. Joint \nConf.4869–4873 (2009) column abundance retrieved from GOSAT observation derived from ordinary kriging, in Proc. ICROS-SICE Int. Joint \nConf.4869–4873 (2009) \nTomosada. M, Kanefuji. K, Matsumoto. Y, and Tsubaki. H, “Application of the spatial statistics to the retrieved CO2 column Tomosada. M, Kanefuji. K, Matsumoto. Y, and Tsubaki. H, “Application of the spatial statistics to the retrieved CO2 column \nabundances derived from GOSAT data,” in Proc. 4th WSEAS Int. Conf. Remote Sens., Venice, Italy, 67–73 (2008)\n395 abundances derived from GOSAT data,” in Proc. 4th WSEAS Int. Conf. D21114-15 (2008) Remote Sens., Venice, Italy, 67–73 (2008) \n395 \nWatanabe, H., Hayashi, K., Saeki, T., Maksyutov, S., et al., Global mapping of greenhouse gases retrieved from GOSAT Level \n2 products by using a kriging method. International Journal of Remote Sensing, 36(6), pp.1509-1528, (2015) \nWennberg, P. O., D. Wunch, C. Roehl, J.-F. Blavier, G. C. Toon, N. Allen, P. Dowell, K. Teske, C. Martin, J. Martin. TCCON \ndata from Lamont, Oklahoma, USA, Release GGG2014R1. TCCON data archive, hosted by CaltechDATA, California ,\n,\n,\ny,\n(\n)\nWatanabe, H., Hayashi, K., Saeki, T., Maksyutov, S., et al., Global mapping of greenhouse gases retrieved from GOSAT Level \n2 products by using a kriging method. International Journal of Remote Sensing, 36(6), pp.1509-1528, (2015) Institute of Technology, Pasadena, CA, U.S.A. (2017) \n400 400 13 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. Wennberg, P. O., D. Wunch, Y. Yavin, G. C. Toon, J.-F. Blavier, N. Allen, G. Keppel-Aleks. TCCON data from Jet Propulsion \nLaboratory, Pasadena, California, USA, Release GGG2014R0. TCCON data archive, hosted by CaltechDATA, California \nInstitute of Technology, Pasadena, CA, U.S.A. (2017) Wennberg, P. O., D. Wunch, Y. Yavin, G. C. Toon, J.-F. Blavier, N. Allen, G. Keppel-Aleks. TCCON data from Jet Propulsion \nLaboratory, Pasadena, California, USA, Release GGG2014R0. TCCON data archive, hosted by CaltechDATA, California \nInstitute of Technology, Pasadena, CA, U.S.A. (2017) \nWennberg, P. O., D. Wunch, C. Roehl, J.-F. Blavier, G. C. Toon, N. Allen. TCCON data from California Institute of \nh\nl\nd\nlif\ni\nl\nd\nhi\nh\nd b\nl\nh\nlif\ni Wennberg, P. O., D. Wunch, Y. Yavin, G. C. Toon, J.-F. Blavier, N. Allen, G. Keppel-Aleks. TCCON data from Jet Propulsion \nLaboratory, Pasadena, California, USA, Release GGG2014R0. TCCON data archive, hosted by CaltechDATA, California \nInstitute of Technology, Pasadena, CA, U.S.A. (2017) \nWennberg, P. O., D. Wunch, C. Roehl, J.-F. Blavier, G. C. Toon, N. Allen. TCCON data from California Institute of \nTechnology, Pasadena, California, USA, Release GGG2014R1. TCCON data archive, hosted by CaltechDATA, California \n05 \nInstitute of Technology, Pasadena, CA, U.S.A. (2017) \nWennberg, P. O., C. Roehl, D. Wunch, G. C. Toon, J.-F. Blavier, R. Washenfelder, G. Keppel-Aleks, N. Allen, J. Ayers. TCCON data from Park Falls, Wisconsin, USA, Release GGG2014R1. TCCON data archive, hosted by CaltechDATA, \nCalifornia Institute of Technology, Pasadena, CA, U.S.A. (2017) gy\n(\n)\nWennberg, P. O., D. Wunch, C. Roehl, J.-F. Blavier, G. C. Toon, N. Allen. D21114-15 (2008) TCCON data from California Institute of Wennberg, P. O., D. Wunch, C. Roehl, J.-F. Blavier, G. C. Toon, N. Allen. TCCON data f Wennberg, P. O., D. Wunch, C. Roehl, J.-F. Blavier, G. C. Toon, N. Allen. TCCON data from California Institute of \nTechnology, Pasadena, California, USA, Release GGG2014R1. TCCON data archive, hosted by CaltechDATA, California \nInstitute of Technology, Pasadena, CA, U.S.A. (2017) 405 Technology, Pasadena, California, USA, Release GGG2014R1. TCCON data archive, hosted by CaltechDATA, California \n05 \nInstitute of Technology, Pasadena, CA, U.S.A. (2017) Wennberg, P. O., C. Roehl, D. Wunch, G. C. Toon, J.-F. Blavier, R. Washenfelder, G. Keppel-Aleks, N. Allen, J. Ayers. TCCON data from Park Falls, Wisconsin, USA, Release GGG2014R1. TCCON data archive, hosted by CaltechDATA, \nCalifornia Institute of Technology, Pasadena, CA, U.S.A. (2017) Yang. Y, Deng. J, Huang. L, Zheng. Q, Wang. K, Tong. C, and Hong. Y,” Modeling and Prediction of NPP-VIIRS Nighttime \n410 \nLight Imagery Based on Spatiotemporal Statistical Method,” IEEE Trans. Geosci. Remote. Sens., 1-13 (2020) Yang. Y, Deng. J, Huang. L, Zheng. Q, Wang. K, Tong. C, and Hong. Y,” Modeling and Prediction of NPP-VIIRS Nighttime \n410 \nLight Imagery Based on Spatiotemporal Statistical Method,” IEEE Trans. Geosci. Remote. Sens., 1-13 (2020) \nZeng. Z, Lei. L, Hou. S, Ru. F, Guan. X, and Zhang. B,” A Regional Gap-Filling Method Based on Spatiotemporal Variogram \nModel of CO2 Columns,” IEEE Trans. Geosci. Remote. Sens.52, 3594-3603 (2013) \nZeng. Z, Lei. L, Hou. S, Ru. F, Guan. X, and Zhang. B “Incorporating temporal variability to improve geostatistical analysis g\ng y\np\np\n,\n,\n(\n)\nZeng. Z, Lei. L, Hou. S, Ru. F, Guan. X, and Zhang. B,” A Regional Gap-Filling Method Based on Spatiotemporal Variogram \nModel of CO2 Columns,” IEEE Trans. Geosci. Remote. Sens.52, 3594-3603 (2013) Zeng. Z, Lei. L, Hou. S, Ru. F, Guan. X, and Zhang. B “Incorporating temporal variability to improve geostatistical analysis \nof satellite-observed CO2 in China,” Chin. Sci. Bull.58, 1948-1954 (2013) \n415 Zeng. Z, Lei. L, Hou. S, Ru. F, Guan. X, and Zhang. B “Incorporating temporal variability to im of satellite-observed CO2 in China,” Chin. Sci. Bull.58, 1948-1954 (2013) \n415 \nZeng. Z, Lei. L, Hou. S, Ru. F, Guan. X, and Zhang. B “A regional gap-filling method based on spatiotemporal variogram \nmodel of CO2 columns,” IEEE Trans. Geosci. Remote Sens., vol. 52, no. 6, pp. 3594–3603, (2014) Zeng. Z, Lei. L, Hou. S, Ru. F, Guan. D21114-15 (2008) X, and Zhang. B “A regional gap-filling method based on spatiotemporal variogram \nmodel of CO2 columns,” IEEE Trans. Geosci. Remote Sens., vol. 52, no. 6, pp. 3594–3603, (2014) 420 425 425 430 14 14 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. 435 Table 1. Geographic locations of TCCON sites used for validation and the statistics used to compare predicted XCO2 and \nTCCON XCO2 observations. Table 1. Geographic locations of TCCON sites used for validation and the statistics used to compare predicted XCO2 and\nTCCON XCO2 observations. Tccon sites (Site abbreviations) \nLongitude \nLatitude \n𝑅! RMSE \nJet Propulsion Laboratory (JC) \n-118.18 \n34.20 \n0.98** \n1.07 \nCaltech (CI) \n-118.13 \n34.14 \n0.97** \n0.95 \nEdwards (DF) \n-117.88 \n34.96 \n0.98** \n0.82 \nFour Corners (FC) \n-108.48 \n36.80 \n0.96** \n0.31 \nLamont (OC) \n-97.49 \n36.60 \n0.98** \n1.04 \nPark Falls (PA) \n-90.27 \n45.94 \n0.98** \n1.24 \nManaus (MA) \n-60.60 \n-3.21 \n0.88** \n0.64 \nIzana (IZ) \n-16.48 \n28.30 \n0.98** \n1.18 \nAscension Island (AE) \n-14.33 \n-7.92 \n0.94** \n0.93 \nOrléans (OR) \n2.11 \n47.97 \n0.99** \n0.95 \nZugspitze (ZS) \n10.98 \n47.42 \n0.92** \n1.52 \nGarmisch (GM) \n11.06 \n47.48 \n0.98** \n1.05 \nNicosia (NI) \n33.38 \n35.14 \n0.93** \n0.73 \nRéunion Island (RA) \n55.49 \n-20.90 \n0.96** \n1.23 \nHefei (HF) \n117.17 \n31.90 \n0.87** \n1.51 \nBurgos (BU) \n120.65 \n18.53 \n0.89** \n1.01 \nAnmeyondo (AN) \n120.65 \n36.54 \n0.90** \n1.20 \nSaga (JS) \n130.29 \n33.24 \n0.97** \n1.26 \nEdwards (DB) \n130.89 \n-12.43 \n0.99** \n0.75 \nTsukuba (TK) \n140.12 \n36.05 \n0.91** \n1.89 \nRikubetsu (RJ) \n143.77 \n43.46 \n0.95** \n1.17 \nWollongong (WG) \n150.88 \n-34.41 \n0.99** \n0.82 \nLauder01&02&03 (LL) \n169.68 \n-45.04 \n0.97** \n1.44 \nAll sites \n- \n- \n0.97** \n1.38 \n** At the 0.01 level (two-tailed), the correlation is significant. 15 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. 445 Table 2. Statistics for predicted monthly-averaged XCO2 and OCO-2 monthly-averaged XCO2 observations. Year \n𝑅 \nNums \n2014 \n0.37** \n129089 \n2015 \n0.74** \n586906 \n2016 \n0.75** \n789007 \n2017 \n0.75** \n641161 \n2018 \n0.70** \n768083 \n2019 \n0.70** \n768083 \n2020 \n0.72** \n28564 \n** At the 0.01 level (two-tailed), the correlation is significant. 450 450 455 16 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. Table 3. The data dictionary for the CDC dataset Table 3. The data dictionary for the CDC dataset \nNumber \nfield names \nData type \nUnit \nFurther description \n1 \nLatitude \nMatrix \n(Degrees, \nminute) \nPoint of latitude \n2 \nLongitude \nMatrix \n(Degrees, \nminute) \nPoint of longitude \n3 \nSpatial XCO2 \nMatrix \nppm \nThe result of spatial interpolation \n4 \nSpatiotemporal XCO2 \nMatrix \nppm \nThe result of spatio-temporal \ninterpolation \n5 \nParment a \nMatrix \n- \nModel parameter a \n6 \nParment b \nMatrix \n- \nModel parameter b \n7 \nParment c \nMatrix \n- \nModel parameter c \n8 \nParment d \nMatrix \n- \nModel parameter d \n9 \nParment e \nMatrix \n- \nModel parameter c \n10 \nParment g \nMatrix \n- \nModel parameter g \n11 \nRMSE \nMatrix \n- \nModel evaluation index \n12 \nR Square \nMatrix \n- \nModel evaluation index \n13 \nCode Version \nFloat \n- \nCode version \n14 \nSpatial Resolution \nFloat \n- \nSpatial resolution \n15 \nNumbers of valid months \nInt \n- \nNumber of valid months in a year \n16 \nLabels TCA \nInt \n- \nLabel \n17 \nUncertain \nInt \n- \nUncertain label \n18 \nTime Curve Parameter Library \nMatrix \n- \nSpatial position coordinates in the \ntime curve parameter library \n0 470 475 17 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. 480 480 Figure 1. Map showing the location of TCCON sites in global. 485 Figure 1 Map showing the location of TCCON sites in global\n485 Figure 1. Map showing the location of TCCON sites in global. 485 490 495 18 18 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. 500 Figure 2. Framework of the proposed methodology. To remove \ncurrent grid data\nGet GOSAT data from the official website\nIs the current grid a valid month? No\nYes\nSpatial prediction results\nSpatial prediction of \neach grid by EBK \nTheory\nTemporal adjustment of each spatial \nprediction grid by applying TCA theory\nGlobal-scale valid monthly \nShp data sets\nFitted in each grid according to formula 1 \nThe parameters are extracted into \nthe time curve library\nSpatiotemporal Prediction Results\nCarbon dioxide columns concentration is seamlessly distributed \nglobally during 2009-2020 Carbon dioxide columns concentration is seamlessly distributed \nglobally during 2009-2020 Carbon dioxide columns concentration is seamlessly distributed \nglobally during 2009-2020 Global-scale valid monthly \nShp data sets Temporal adjustment of each spatial \nprediction grid by applying TCA theory Spatiotemporal Prediction Results Figure 2. Framework of the proposed methodology. 505 510 510 19 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. Figure 3. Scatter plots of predicted XCO2 and XCO2 observations at 23 TCCON sites. P XCO2 is the predicted XCO2. T XCO2 \nis the TCCON XCO2. 515 Figure 3. Scatter plots of predicted XCO2 and XCO2 observations at 23 TCCON sites. P XCO2 is the predicted XCO2. T XCO2 \ni th TCCON XCO Figure 3. Scatter plots of predicted XCO2 and XCO2 observations at 23 TCCON sites. P XCO2 is the predicted XCO2. T XCO2 \nis the TCCON XCO2. 520 20 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. 525 Figure 4. Density scatter plots of predicted XCO2 and observed one from OCO-2. Figure 4. Density scatter plots of predicted XCO2 and observed one from OCO-2. 530 535 21 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. 540 Figure 5. Scatter plots of predicted XCO2 and XCO2 observed from GOSAT_L3. P XCO2 is the predicted XCO2. T XCO2 is \nthe GOSAT_L3 XCO2. The blue dots in the graph represent the raw data in different years. The yellow line represents the line \nwhere the original data was fitted. 5 Figure 5. Scatter plots of predicted XCO2 and XCO2 observed from GOSAT_L3. P XCO2 is the predicted XCO2. T XCO2 is \nthe GOSAT_L3 XCO2. The blue dots in the graph represent the raw data in different years. The yellow line represents the line \nwhere the original data was fitted. 45 Figure 5. Scatter plots of predicted XCO2 and XCO2 observed from GOSAT_L3. P XCO2 is the predicted XCO2. T XCO2 is \nthe GOSAT_L3 XCO2. The blue dots in the graph represent the raw data in different years. The yellow line represents the line \nwhere the original data was fitted. 5 545 550 22 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. 555 Figure 6. Error with graph of predicted XCO2 from 2009 to 2020. PXCO2 is the predicted XCO2. The blue shading represents \nthe standard deviation of the CDC data set for the screened locations in the corresponding year in the figure. The yellow dots \nrepresent the mean of the CDC data set for the screened locations in the corresponding year in the figure. 60 Figure 6. Error with graph of predicted XCO2 from 2009 to 2020. PXCO2 is the predicted XCO2. The blue shading represents \nthe standard deviation of the CDC data set for the screened locations in the corresponding year in the figure. The yellow dots \nrepresent the mean of the CDC data set for the screened locations in the corresponding year in the figure. 60 Figure 6. Error with graph of predicted XCO2 from 2009 to 2020. PXCO2 is the predicted XCO2. The blue shading represents \nthe standard deviation of the CDC data set for the screened locations in the corresponding year in the figure. The yellow dots \nrepresent the mean of the CDC data set for the screened locations in the corresponding year in the figure. 60 560 565 23 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. 570 Figure 7. Product data from Our Algorithm in 2010. This CDC dataset covers approximately from 55°N to 55°S, with a spatial \nresolution of 0.25°. Figure 7. Product data from Our Algorithm in 2010. This CDC dataset covers approximately from 55°N to 55°S, with a spatial \nresolution of 0.25°. resolution of 0.25°. 24 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. 580 585 \nFigure 8. Product data from Our Algorithm in 2011. This CDC dataset covers approximately from 55°N to 55°S, with a spatial \nresolution of 0.25°. Jan\nFeb\nMar\nApr\nJul\nJun\nMay\nAug\nSep\nOct\nNov\nDec Mar Aug Jul Aug Jul Oct 585 Figure 8. Product data from Our Algorithm in 2011. This CDC dataset covers approximately from 55°N to 55°S, with a spatial \nresolution of 0.25°. 590 25 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. 595 \nFigure 9. Product data from Our Algorithm in 2012. This CDC dataset covers approximately from 55°N to 55°S, with a spatial \nresolution of 0.25°. Jan\nFeb\nMar\nApr\nJul\nJun\nMay\nAug\nSep\nOct\nNov\nDec Jan Sep Aug Aug Jul Sep Jul 595 Figure 9. Product data from Our Algorithm in 2012. This CDC dataset covers approximately from 55°N to 55°S, with a spatial \nresolution of 0.25°. 600 26 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. 605 Figure 10. Product data from Our Algorithm in 2013. This CDC dataset covers approximately from 55°N to 55°S, with a \n610 \nspatial resolution of 0.25°. Jan\nFeb\nMar\nApr\nJul\nJun\nMay\nAug\nSep\nOct\nNov\nDec Feb Mar Aug Jul Aug Jul Oct Figure 10. Product data from Our Algorithm in 2013. This CDC dataset covers approximately from 55°N to 55°S, with a \n610 \nspatial resolution of 0.25°. 615 27 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. 620 Figure 11. Product data from Our Algorithm in 2014. This CDC dataset covers approximately from 55°N to 55°S, with a \nspatial resolution of 0.25°. Jan\nFeb\nMar\nApr\nJul\nJun\nMay\nAug\nSep\nOct\nNov\nDec Mar Aug Jul Aug Jul Oct Figure 11. Product data from Our Algorithm in 2014. This CDC dataset covers approximately from 55°N to 55°S, with a \nspatial resolution of 0.25°. 625 28 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. 630 Figure 12. Product data from Our Algorithm in 2015. This CDC dataset covers approximately from 55°N to 55°S, with a \nspatial resolution of 0.25°. Jan\nFeb\nMar\nApr\nJul\nJun\nMay\nAug\nSep\nOct\nNov\nDec Mar Apr Aug Jul Aug Jul Oct Figure 12. Product data from Our Algorithm in 2015. This CDC dataset covers approximately from 55°N to 55°S, with a \nspatial resolution of 0.25°. spatial resolution of 0.25°. 635 635 29 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. 640 Figure 13. Product data from Our Algorithm in 2016. This CDC dataset covers approximately from 55°N to 55°S, with a \n645 \nspatial resolution of 0.25°. Jan\nFeb\nMar\nApr\nJul\nJun\nMay\nAug\nSep\nOct\nNov\nDec Mar Feb Jan Sep Aug Jul Aug Jul Oct Figure 13. Product data from Our Algorithm in 2016. This CDC dataset covers approximately from 55°N to 55°S, with a \n645 \nspatial resolution of 0.25°. 650 650 30 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. 655 Figure 14. Product data from Our Algorithm in 2017. This CDC dataset covers approximately from 55°N to 55°S, with a \nspatial resolution of 0.25°. Jan\nFeb\nMar\nApr\nJul\nJun\nMay\nAug\nSep\nOct\nNov\nDec\nDec Mar Sep Aug Jul Aug Jul Oct Figure 14. Product data from Our Algorithm in 2017. This CDC dataset covers approximately from 55°N to 55°S, with a \nspatial resolution of 0.25°. 660 31 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. 665 Figure 15. Product data from Our Algorithm in 2018. This CDC dataset covers approximately from 55°N to 55°S, with a \nspatial resolution of 0.25°. 0 \nJan\nFeb\nMar\nApr\nJul\nJun\nMay\nAug\nSep\nOct\nNov\nDec Mar Jan Jul Aug Jul Oct\nNov Oct Figure 15. Product data from Our Algorithm in 2018. This CDC dataset covers approximately from 55°N to 55°S, with a \nspatial resolution of 0.25°. spatial resolution of 0.25°. 670 670 675 32 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. 680 \nFigure 16. Product data from Our Algorithm in 2019. This CDC dataset covers approximately from 55°N to 55°S, with a \nspatial resolution of 0.25°. Jan\nFeb\nMar\nApr\nJul\nJun\nMay\nAug\nSep\nOct\nNov\nDec Mar Jul Jul 680 Figure 16. Product data from Our Algorithm in 2019. This CDC dataset covers approximately from 55°N to 55°S, with a \nspatial resolution of 0.25°. 685 33 https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. https://doi.org/10.5194/essd-2022-215\nPreprint. Discussion started: 19 July 2022\nc⃝Author(s) 2022. CC BY 4.0 License. 690 Figure 17. Product data from Our Algorithm in 2020. This CDC dataset covers approximately from 55°N to 55°S, with a \nspatial resolution of 0.25°. Jan\nFeb\nMar\nApr\nJul\nJun\nMay\nAug\nSep\nOct\nNov\nDec Jul Figure 17. Product data from Our Algorithm in 2020. This CDC dataset covers approximately from 55°N to 55°S, with a \nspatial resolution of 0.25°. spatial resolution of 0.25°. 695 34"