Comparative Evaluation of Top-Down GOSAT XCO 2 vs. Bottom-Up National Reports in the European Countries

: Submitting national inventory reports (NIRs) on emissions of greenhouse gases (GHGs) is obligatory for parties of the United Nations Framework Convention on Climate Change (UNFCCC). The NIR forms the basis for monitoring individual countries’ progress on mitigating climate change. Countries prepare NIRs using the default bottom–up methodology of the Intergovernmental Panel on Climate Change (IPCC), as approved by the Kyoto protocol. We provide tangible evidence of the discrepancy between ofﬁcial bottom–up NIR reporting (unit: tons) versus top–down XCO 2 reporting (unit: ppm) within the European continent, as measured by the Greenhouse Gases Observing Satellite (GOSAT). Bottom–up NIR (annual growth rate of CO 2 emission from 2010 to 2016: − 1.55%) does not show meaningful correlation (geographically weighted regression coefﬁcient = − 0.001, R 2 = 0.024) to top–down GOSAT XCO 2 (annual growth rate: 0.59%) in the European countries. The top ﬁve countries within the European continent on carbon emissions in NIR do not match the top ﬁve countries on GOSAT XCO 2 concentrations. NIR exhibits anthropogenic carbon-generating activity within country boundaries, whereas satellite signals reveal the trans-boundary movement of natural and anthropogenic carbon. Although bottom–up NIR reporting has already gained worldwide recognition as a method to track national follow-up for treaty obligations, the single approach based on bottom–up did not present background atmospheric CO 2 density derived from the air mass movement between the countries. In conclusion, we suggest an integrated measuring, reporting, and veriﬁcation (MRV) approach using top–down observation in combination with bottom–up NIR that can provide sufﬁcient countrywide objective evidence for national follow-up activities.


Introduction
National inventory reports (NIRs) are required for tracking national adherence to treaty obligations and recommendations arising from Nationally Determined Contributions (NDCs) mechanisms under the Paris Agreement and United Nations Framework Convention on Climate Change (UNFCCC) [1]. An NIR consists of CO 2 emissions and removal from five categories (energy, waste, industrial process and production use, agriculture and land use, land-use change and forestry [LULUCF]). The calculation of CO 2 emissions in an NIR is based on the carbon footprint concept, which considers all relevant carbon sources, sinks, and storage within the country boundary [2]. An NIR presents the quantity of greenhouse gases (GHGs) estimated with the emission factors and units of emission-generating activity emitted into the atmosphere by an individual, organization, process, product, or event from within a specified country boundary [3]. Once CO 2 is emitted into the atmosphere, it increases the atmosphere's net resident CO 2 [4]. In this

Materials and Methods
The NIR and GOSAT XCO 2 are represented by two different measurement units: bottom-up carbon footprint (ton) and top-down atmospheric CO 2 data (unit: ppm). To compare the two datasets under the same unit of measurement, the annual growth rate of the two datasets was calculated as a percentage (%). The integrated variation was calculated by subtracting the annual growth rate of the previous year from the annual growth rate of the following year. For instance, if a country had 1% variation in 2010-2011 and 1.5% variation in 2011-2012, the integrated variation from 2010 to 2012 was estimated as 0.5%. That is why the integrated variation is smaller than the individual variation.

Study Area
According to the Kyoto protocol, Annex I countries (which developed country parties in 1992) must submit an updated NIR each year. Uncertainties in NIR data are relatively low for Annex 1 countries due to well-developed statistical systems and their capacity to use higher-tier methods [14]. Europe is the second smallest continent in the world after Australia and Oceania, but as 44 countries are concentrated in one area, it is an ideal region for studying carbon emissions among countries. Europe is also ideal for investigating carbon inventories in various climates, due to diverse land cover and forests, such as the Mediterranean forest, temperate forest, tundra, coniferous forest, and steppes [15].
The Total Carbon Column Observing Network (TCCON) is a global network of instruments that measure the amount of carbon dioxide and other trace gases in the Earth's atmosphere [16]. TCCON provides the primary validation dataset for the GOSAT, and it has been used to validate other space-based measurements of CO 2 [7,17,18]. Out of a Sustainability 2021, 13, 6700 3 of 15 total of 23 worldwide TCCON sites, eight are in Europe; therefore, there is a relatively large distribution of verification points of satellite measurements in comparison to other continents. In this regard, GOSAT XCO 2 data acquired in Europe have been produced with more validation procedures than those from other continents. Therefore, Europe is an ideal continent with an adequate land mosaic for comparing bottom-up CO 2 emissions and top-down column abundance XCO 2 , because various types of countries with diverse human and natural environments are concentrated in a small area, and there are a large number of GOSAT XCO 2 validation points. We selected the 33 European countries belonging to Annex 1 countries. We did not include Iceland because it is too far from the European continent.

GOSAT XCO 2 Observations
The GOSAT orbits at an altitude of approximately 666 km, with 10.5 km of spatial resolution and three-day temporal resolution [19]. The observation instrument onboard GOSAT is the Thermal and Near Infrared Sensor for Carbon Observation (TANSO) composed of two subunits: the Fourier-Transform Spectrometer (FTS) and the Cloud and Aerosol Imager (CAI). The TANSO-FTS operates with spectral resolution in three narrow bands in the short-wavelength infrared (SWIR) region (0.76, 1.6, and 2.0 µm) and a wide thermal infrared band (5.5-14.3 µm). TANSO-CAI operates with spectral resolution in four narrow bands in the near-ultraviolet to a near-infrared region (0.38, 0.674, 0.87, and 1.6 µm). TANSO can detect optically thick clouds inside the TANSO-FTS instantaneous field of view (IFOV) and correct aerosols' effect in the TANSO-FTS spectrum data [20,21]. In this study, we used the GOSAT SWIR Level 2 product (column-averaged mixing ratios of CO 2 , V02.75) for 2010-2016 (total 20,929 data points) retrieved with the National Institute for Environmental Studies (NIES) CO 2 retrieval algorithm [22], which is corrected with the regression coefficients (0.89) between GOSAT XCO 2 and TCCON data [23].

Bottom-Up NIR
NIR contains detailed descriptive and numerical information and common reporting format (CRF) tables for all Kyoto protocol GHG [24,25]. On the contrary, GOSAT only provides the column abundance of CO 2 (the number of the gas molecule in a vertical unit column). To perform a comparative evaluation of the NIR data and GOSAT XCO 2 data, we used direct CO 2 emissions (CRF Table 10s2

Visualization Utilizing GOSAT Level 3 Products
GOSAT Level 3 data were generated to visualize the spatial distribution of XCO 2 on a regional scale by using the empirical Bayesian kriging (EBK) surface interpolation approach (Figure 1). EBK is a geostatistical interpolation method that automatically calculates a K-Bessel semi-variogram model through a process of sub-setting simulations [27,28]. GOSAT XCO 2 detects the column-averaged abundance of CO 2 in the atmosphere at the time of data acquisition (in real-time). Therefore, GOSAT XCO 2 strongly reflects regional CO 2 concentration, such as the presence of CO 2 sources and the difference in CO 2 flux due to the photosynthesis and respiration from the terrestrial biosphere [29]. CO 2 emitted from a source moves continuously with the flow of the atmosphere. Emissions strongly affect atmospheric CO 2 concentration; however, the farther away from the source, the lower the impact on the CO 2 concentration [30]. By dividing the input GOSAT point data into small subsets, EBK derived the different semi-variograms reflecting distance away from the emission source [27]. Figure 1b shows the results of the EBK, reflecting regional CO 2 distribution characteristics in accordance with the non-stationary flow of the atmosphere. Cross-validation is commonly used to validate how good the kriging model interpolates by comparing the estimated values with the observed values [31]. There are various criteria for the cross-validation of prediction errors for EBK models, such as root mean square error (RMSE), average standard error (ASE), mean standardized error (MSE), and root mean square standardized error (RMSSE) [32]. RMSE provides a measure of interpolation precision, where a lower RMSE indicates a more precise estimation. For this research, ASE was distributed in the range of 1.833 to 2.204 (ppm), whereas RMSE was in the range of 1.765 to 2.121 (ppm; Table 2). As EBK was applied for yearly GOSAT XCO 2 data, the difference of CO 2 concentration by season is the cause of ASE and RMSE deviations in this range [33]. ASE values close to RMSEs imply that the variability in prediction is correctly assessed. The results of this study revealed that ASE was close to RMSE, suggesting a reasonably acceptable level of error, considering that the interpolation was conducted for the fluctuation of carbon concentration occurring on the entire European continent. RMSSEs were calculated by dividing the root average of RMSE by ASE. An RMSSE close to 1 indicates fewer prediction standard errors [27]. MSE was calculated by dividing the sum of the difference between the measured and the predicted values by the kriging variance. An MSE close to 0 indicates that the bias of the EBK model is low [27]. In this study, RMSSE was close to 1, ranging from 0.969 to 0.993, and MSE was also close to 0, from −0.011 to 0.013, indicating that the model's accuracy is reasonably acceptable ( Table 2).

Geographically Weighted Regression (GWR)
Atmospheric carbon is a natural phenomenon that flows across national boundaries, interacting with the Earth and the atmosphere [34]. It is normal for top-down satellite

Geographically Weighted Regression (GWR)
Atmospheric carbon is a natural phenomenon that flows across national boundaries, interacting with the Earth and the atmosphere [34]. It is normal for top-down satellite data reflecting natural phenomena to retain information on spatial dependence or spatial heterogeneity among neighboring countries sharing a similar space. Therefore, CO 2 fluxes and concentrations are influenced by local parameters, such as climatic conditions and Sustainability 2021, 13, 6700 6 of 15 stationary CO 2 sources (e.g., power plants, oil refineries, etc.) within national boundaries. GWR searches for spatial heterogeneity by applying different regression models to data in different locations in space. Therefore, local regression coefficients were derived for individual countries when using a GWR regression model between NIR and satellite signals, whereas traditional linear regression analysis estimates a single global regression coefficient on the whole area of analysis. With GWR, it was possible to determine different local coefficients for each country and to identify spatially heterogeneous patterns that are difficult to grasp with traditional ordinary least squares (OLS) models.  [35][36][37]. The NIR 4 and 5 sectors rebound in increasing trends was triggered by the higher heat demand by households and services due to colder winter conditions and higher road transport demands from 2014 to 2016 [37][38][39].

A Comparative Evaluation of Top-Down vs. Bottom-Up
Sustainability 2021, 13, x FOR PEER REVIEW 6 of 16 data reflecting natural phenomena to retain information on spatial dependence or spatial heterogeneity among neighboring countries sharing a similar space. Therefore, CO2 fluxes and concentrations are influenced by local parameters, such as climatic conditions and stationary CO2 sources (e.g., power plants, oil refineries, etc.) within national boundaries. GWR searches for spatial heterogeneity by applying different regression models to data in different locations in space. Therefore, local regression coefficients were derived for individual countries when using a GWR regression model between NIR and satellite signals, whereas traditional linear regression analysis estimates a single global regression coefficient on the whole area of analysis. With GWR, it was possible to determine different local coefficients for each country and to identify spatially heterogeneous patterns that are difficult to grasp with traditional ordinary least squares (OLS) models.

A Comparative Evaluation of Top-Down vs. Bottom-Up
Annual changing trends of top-down GOSAT XCO2 (ppm) show strong contradictory trends between bottom-up NIR 4 and 5 sectors (Figure 2a) [35][36][37]. The NIR 4 and 5 sectors rebound in increasing trends was triggered by the higher heat demand by households and services due to colder winter conditions and higher road transport demands from 2014 to 2016 [37][38][39].  Although emissions from the NIR 4 and 5 sectors appeared to increase or decrease relative to the previous year, GOSAT XCO 2 continues to increase. The average GOSAT XCO 2 growth rate from 2014 to 2016 (0.65%) was higher than from 2010 to 2014 (0.56%; Figure 2b). The bottom-up NIR 4 and 5 sectors represent the residual terrestrial CO 2 sinks (terrestrial biosphere) and anthropogenic CO 2 sources. Decreased CO 2 uptakes from terrestrial CO 2 sinks and anthropogenic sources (NIR 4 and 5 sectors) influence atmospheric carbon residuals. Therefore, annual changing trends of top-down GOSAT XCO 2 (ppm) may be positively correlated with NIR CO 2 emissions from terrestrial carbon sinks and anthropogenic sources (NIR 4 and 5 sectors) [40]. GOSAT XCO 2 steeply increased in 2011-2012, as well as 2015-2016. It is assumed to be the natural phenomenon that occurred by the reduction in biospheric CO 2 uptake due to the heavy snowfall (2011-2012) and the El Niño event (2015-2016) [41].

A Comparative Evaluation among Individual Countries
The top five countries in NIR 4 and 5 sectors are not in good agreement with the top five countries with high GOSAT XCO 2 concentrations (Table 3). There are only five countries (Estonia, Hungary, Slovakia, Slovenia, and the United Kingdom) where GOSAT XCO 2 is decreasing among the 27 countries in which both NIR 4 and 5 sectors are decreasing. Similarly, the bottom five countries in NIR 4 and 5 sectors are not at all consistent with the bottom five countries for GOSAT XCO 2 concentrations ( Figure 3; Table 3). In addition, when GWR analysis was performed on top-down and bottom-up changing trends at the national level, there was no significant correlation (R 2 for NIR 4 sectors: 0.070; R 2 for NIR 5 sectors: 0.024) between independent variables and dependent variables (Table 4).

Discussion
The preconditions and tools used in the top-down versus bottom-up methods differ, as shown in Table 5. NIR reveals anthropogenic carbon-generating activity within country boundaries, whereas satellite signal exhibits the trans-boundary movement of atmospheric carbon. It is inevitable that different methods for calculating CO2 (i.e., calculating

Discussion
The preconditions and tools used in the top-down versus bottom-up methods differ, as shown in Table 5. NIR reveals anthropogenic carbon-generating activity within country boundaries, whereas satellite signal exhibits the trans-boundary movement of atmospheric carbon. It is inevitable that different methods for calculating CO 2 (i.e., calculating concentration units versus volume units) yield different results. Bottom-up IPCC guidelines calculate the total volume of CO 2 emissions by quantifying CO 2 per unit of a particular activity (an emission factor) and multiplying the numbers of emissions-generating activity occurring within the country boundaries [42,43]. Consequently, the emission factors and numbers of emissions activity within boundaries are crucial in determining the range of CO 2 emissions and variations. The units for emission-generating activities and emission factors greatly influence variations in net CO 2 emissions of NIR 4 and 5 sectors. In contrast, top-down GOSAT XCO 2 measures direct column-averaged dry-air mole fractions of CO 2 as ppm using radiance spectra from the sensitive wavelength band [20] (Table 5).
There are inconsistencies in CO 2 emission estimations based on initial assumptions about carbon footprint as CO 2 sources in bottom-up NIR [44], as emission coefficients fluctuate depending on a variety of variables, such as collection and reporting system of energy statistics, data definition, data processing, level of detail and local specific conditions [45]. IPCC guidelines present three bottom-up methods: (1) Tier 1 (global default model), (2) Tier 2 (nation-specific model), and (3) Tier 3 (local-specific model). Bottom-up estimates fluctuate based on how tier stages are set [8]. Unlike the NIR, GOSAT XCO 2 presents atmospheric density accumulated by different types of sources or sinks (background atmospheric CO 2 ) accumulated during the complete carbon life stages of more than 100 years [46], including other natural and anthropogenic sources or natural sinks that are not present in IPCC guidelines, such as volcano eruptions [47]. In addition, GOSAT XCO 2 detects the direct and indirect CO 2 emissions from terrestrial biotic activity derived from natural disturbances. GOSAT XCO 2 is the integrated CO 2 in a column from the surface to the top of the atmosphere. Therefore, GOSAT XCO 2 has very limited sensitivity to the bottom layer of the atmosphere where human activities are interacting with it. It has been shown that the XCO 2 variability is driven by both surface emissions and atmospheric transport contaminated with clouds and clear sky ratio, according to geolocation [48]. The developed countries listed in Annex I of the UNFCCC must submit national GHG inventories of anthropogenic emissions by sources and removals by sinks accounted with IPCC guidelines [52]. The energy sector is the most important CO 2 source affecting the carbon footprint among the five categories in IPCC guidelines as it accounts for approximately 77-80% of total net CO 2 emissions from EU-33 countries' NIR. In particular, solid fuel combustion from public electricity and heat production is a major CO 2 source in the energy sector, accounting for 60.6-67.9% of total CO 2 emission in the energy sector [38]. IPCC guidelines do not count the CO 2 emissions avoided from the use of blast furnaces by reusing by-products of industrial processes or by replacing conventional solid fossil fuels [53] such as anthracite and lignite. Much higher emission factors are being observed in Belgium (238.23 tCO 2 /TJ) and Sweden (206.37 tCO 2 /TJ) than other EU countries (89.95 tCO 2 /TJ to 122.96 tCO 2 /TJ) from solid fuels utilized in public electricity and heat production ( Figure 4). The blast furnace gas used in Belgium and Sweden emits much more CO 2 than the anthracite and lignite used in the rest of the EU countries. Blast furnace gas is the secondary fuel created as a free by-product derived from blast furnace coke burned in the iron and steel production process [51], while anthracite and lignite are the primary fuels used in conventional coal power plants [54]. The reduced combustible components from free by-products cause much more CO 2 emission than anthracite or lignite because the energy efficiency of blast furnace gas is considerably low ( Table 6). In Belgium and Sweden, most electricity (around 80%) is produced using biomass and by-product derived from waste in Sweden [55] and nuclear, natural gas, and renewable energy in Belgium [56], while very few places use coal (around 10%).
The CO 2 emissions accumulated over the complete life stages of a carbon-generating activity (LCA: life cycle assessment) are not estimated by IPCC guidelines because there is a risk of double-counting of emissions transferred from the industrial processes and product use (IPPU) sector [51] to the energy sector. As blast furnace gas is a by-product of industrial processes, there is no energy consumption to exploit, unlike fossil fuels or industrial waste processing. The current IPCC guidelines do not accommodate the positive carbon footprint effect caused by renewable energy and enhanced technologies such as CHP plants (combined heat and power) using blast furnace gas (Table 6). Table 6. Comparison of emission factors on solid fuels utilized in public electricity and heat production among EU-33 countries.

Category Belgium and Sweden Other EU-33 Countries
Fuel types in public electricity and heat production Blast furnace gas Anthracite, lignite, etc.

Productions
Secondary fuels (by-product during the production of metals or steels) Primary fuel (raw materials)  NIR is always confined within the national boundary of individual countries, unlike GOSAT XCO2, where there is no such barrier. The NIR is calculated based on the anthropogenic carbon dioxide according to energy consumption statistics in the different sectors and reflects the socio-economic condition of individual countries in terms of GHG emission. It is possible to use the NIR for monitoring the progress of individual countries to NIR is always confined within the national boundary of individual countries, unlike GOSAT XCO 2 , where there is no such barrier. The NIR is calculated based on the anthropogenic carbon dioxide according to energy consumption statistics in the different sectors and reflects the socio-economic condition of individual countries in terms of GHG emission. It is possible to use the NIR for monitoring the progress of individual countries to mitigate emissions from the earth surface, owing to statistical validation using the Monte-Carlo analysis and the time-period of measurement within confined national boundaries [61]. GOSAT XCO 2 is largely driven by the air mass movement between the countries. The air column monitored by GOSAT is not the real representation of emission from the country under investigation. The prevailing lapse rate guided by the direction of the air mass movement and other meteorological factors has a bigger role in making the reflectance column of GOSAT over a country, and thus it cannot be considered the emission from a country [21]. The influence from atmospheric transport in the GOSAT XCO 2 signal is much stronger than anthropogenic sources such as power plants [18,24]. CO 2 flux from the terrestrial biosphere is much higher than that of anthropogenic emissions, which show high uncertainty [62]. In this regard, the NIR has significant limitations in explaining such background atmospheric concentration at the global or regional level required for trans-boundary comparison, which reflects the regional atmospheric stability governed by meteorological parameters such as wind speed, vertical mixing depth, including the effects of long-distance movements.
GOSAT XCO 2 is processed by the inverse methods of atmospheric transport [63]. To retrieve XCO 2 (GOSAT level 2) from GOSAT spectra (GOSAT level 1), the observed spectrum is used to minimize the difference between the raw data and surface/atmospheric radiative transfer model that simulating gas concentrations of the earth surface and atmospheric state. Mathematical details of these algorithms can be found in Yoshida et al. [64]. GOSAT level 2 XCO 2 data are processed with a NIES off-line global transport model [65] to simulate and reflect seasonal and spatial distributions of long-lived atmospheric constituents in the lower and mid-troposphere. In this regard, the GOSAT level 2 XCO 2 product has little difference from the inverse transport model-based CO 2 data. The differences between GOSAT level 2 XCO 2 and inverse transport model-based XCO 2 results are mostly within 0.5% globally [64]. The southern wind from the north European area increases the optical depths 0.26-0.43 in eastern Europe. The eastern wind flowing from Spain and Portugal to Belarus increases the optical depths 0.21-0.59 in France, Germany, Czech Republic, Poland, Belarus, 0.31-0.33 in Belgium and the Netherlands and 0.07-0.15 in Switzerland [66,67]. In particular, countries with a narrow land area are intensively concentrated in Europe, as shown below; Belgium (world land area ranking: 141, land area: 30,528 km 2 ), Switzerland (136, 41,284 km 2 ), the Netherlands (135, 41,543 km 2 ), Denmark (134, 43,094 km 2 ). In the case of a country with a small national territory, GOSAT does not measure the amount of carbon emitted from the country where the satellite platform is located but has a signal characteristic that is influenced by neighboring countries due to transboundary effects of atmospheric circulation associated with the asymmetric heating of land and sea. Therefore, it is not possible for the GOSAT data to quantify the amount of carbon emission and absorption caused by natural and anthropogenic carbon-generating activity within country boundaries. Since GOSAT data are processed using the inverse transport model, it has the advantage of measuring the cross-border atmospheric movement of natural and anthropogenic carbon. However, there are significant limitations to assessing the correlation between NIR reporting carbon emissions of individual countries in conjunction with GOSAT signals. Previous research suggests that the progress of actual emission-reducing actions presented in NIR is reliable only if it is verified by direct top-down methods that measure the actual CO 2 density (ppm) in the atmosphere [14]. It is essential to evaluate the CO 2 emissions calculated from various nation-specific emissions factors, activity data, and models by using top-down, standardized references in ppm. However, the IPCC guidelines improved through two revisions (1996,2006) still strongly lean toward bottom-up methods. We need to develop integrative Measuring, Reporting, and Verifications (MRV) with top-down and bottom-up methods to verify CO 2 emission mitigation actions presented in NIR.

Conclusions
The GOSAT data (top-down) provided a trans-boundary real atmospheric CO 2 density (unit: ppm) across 33 European countries from 2011 to 2016. It was particularly useful for assessing the spatial and temporal intensity of anthropogenic and natural CO 2 sources and sinks. When GOSAT data are evaluated in terms of information required for demonstrating trans-boundary objective evidence for national follow-up activities of UNFCCC, it was a much more powerful tool than bottom-up NIR (unit: ton) by quantifying the extent of trans-boundary carbon concentration (unit: ppm) in the European domain scale. However, GOSAT data have also shown limitations and the measurement error caused by natural phenomena, such as aerosol, diffusion radiations, and clouds. Current estimates of CO 2 emissions presented in NIR are based on the upscaling of emission activity data within defined country boundaries. In this regard, NIR may be a reasonable choice for analyzing anthropogenic activities. The results of this study confirmed that a carbon satellite could be a second-choice option when requiring the validation of NIR by assessing the spatial and temporal distribution of real atmospheric CO 2 density. The MRV problems are not solved entirely by any approach, as no single data-acquisition methodology can satisfy all monitoring needs. However, NIR, aided by carbon satellite, can complement the present national follow-up in a synergistic way.  Data Availability Statement: Publicly available datasets were analyzed in this study. This data can be found here: https://unfccc.int/ghg-inventories-annex-i-parties/2021, accessed on 20 December 2020, https://data2.gosat.nies.go.jp/GosatDataArchiveService/usr/download/ProductPage/view, accessed on 20 December 2020.