A ten-year global monthly averaged terrestrial NEE inferred from the ACOS GOSAT v9 XCO2 retrievals (GCAS2021)

A global gridded Net Ecosystem Exchange (NEE) of CO2 dataset is vital in global and regional carbon cycle studies. Top-down atmospheric inversion is one of the major methods to estimate the global NEE, however, the existing global NEE datasets generated through inversion from conventional CO2 observations have large uncertainties in places where observational data are sparse. Here, by assimilating the GOSAT ACOS v9 XCO2 product, we generate a ten-year (2010–2019) 15 global monthly terrestrial NEE dataset using the Global Carbon Assimilation System, version 2 (GCASv2), which is named as GCAS2021. It includes gridded (1°×1°), globally, latitudinally, and regionally aggregated prior and posterior NEE and ocean (OCN) fluxes, and prescribed wildfire (FIRE) and fossil fuel and cement (FFC) carbon emissions. Globally, the decadal mean NEE is -3.73±0.52 PgC yr-1, with interannual amplitude of 2.73 PgC yr-1. Combining the OCN flux, and FIRE and FFC emissions, the net biosphere flux (NBE) and atmospheric growth rate (AGR) as well as their inter-annual variabilities (IAVs) 20 agree well with the estimates of Global Carbon Budget 2020. Regionally, our dateset shows that eastern North America, Amazon, Congo Basin, Europe, boreal forests, southern China and Southeast Asia are carbon sinks, while western US, African grasslands, Brazilian plateaus and parts of South Asia are carbon sources. In the TRANSCOM land regions, the NBEs of temperate N. America, northern Africa and boreal Asia are between the estimates of CMS-Flux NBE 2020 and CT2019B, and those in temperate Asia, Europe, and Southeast Asia are consistent with CMS-Flux NBE 2020 but significantly different from 25 CT2019B. In the RECCAP2 regions, except for Africa and South Asia, the NBEs are comparable with the latest bottom-up estimate of Ciais et al. (2021). Compared with previous studies, the IAVs and seasonal cycles of NEE of this dataset could clearly reflect the impacts of extreme climates and large-scale climate anomalies on the carbon flux. The evaluations also show that the posterior CO2 concentrations at remote sites and in regional scale, as well as on vertical CO2 profiles in the AsiaPacific region and the Amazon basin, are all consistent with independent CO2 measurements from surface flask and aircraft 30 CO2 observations, indicating that this dataset captures surface carbon fluxes well. We believe that this dataset will contribute to regional or national-scale carbon cycle and carbon neutrality assessment, and carbon dynamics research. The dataset can be https://doi.org/10.5194/essd-2022-15 O pe n A cc es s Earth System Science Data D icu ssio n s Preprint. Discussion started: 4 February 2022 c © Author(s) 2022. CC BY 4.0 License.

Southeast and South Asia, therefore, the CO2 vertical profiles near 8 cities over the Asia-Pacific region are selected in this study. The 8 cities are Hong Kong, Singapore, Jakarta, Bangkok, Sydney, New Delhi, Shanghai, and Tokyo. The HIPPO programme completed aircraft measurements spanning the Pacific from 85 ° N to 67 ° S during the periods of March to April 2010, and June to September 2011, with vertical profiles every approximately 2.2 ° of latitude (Wofsy et al., 2011). The During the evaluation of this study, TAB and TEF are combined as one site of TAB_TEF. At each site, 1-3 spiral profiles from approximately 4420 m to about 300 m a.s.l. were observed in each month. It is worth noting that OBSPACKv6 only provides ALF, RBA, SAN and TAB observations from 2010 to 2012, the rest data were downloaded from Gatti et al. (2021). For the 190 CONTRAIL vertical profiles, the observations between the heights of 2 and 6 km are used, because the data measured below 2000 m are highly affected by local emissions (Jiang et al., 2014) due to the frequently ascending and descending of aircrafts.
And for the HIPPO and CARBAM observations, the data above 1 km are adopted. members could be used for calculating the posterior uncertainties based on user defined regional masks. We also provide a Fortran program for the calculation of posterior uncertainties. The gridded data are in NETCDF-3 format, while the regional aggregated data are in xlsx format. Table 1 presents the year-by-year and decadal averaged posterior global carbon budgets during 2010 ~ 2019 of this study. The global annual NEE is in the range of -2.51±0.53 to -5.24±0.50 PgC yr -1 (negative means absorbing CO2 from the atmosphere, and positive means releasing CO2 to the atmosphere). The year of 2011 has the largest land sink in the decade, while the year of 2016 has the weakest one, with interannual amplitude reaching 2.73 PgC yr -1 . On average, the decadal mean NEE is -3.73±0.52 PgC yr -1 . The OCN flux has an overall increase trend from 2010 to 2009, with a mean of -2.64±0.16 PgC yr -1 . Figure  220 4 presents a comparison between the estimates of this study and GCP2020 (Friedlingstein et al., 2020). There are large differences for the land-use and land-cover change (LULLC) carbon emissions between this study and GCP2020, we directly use the FIRE emission from GFED 4.1s as prescribed land-use emission, while GCP2020 uses an average of three bookkeeping models (Houghton et al., 2017;Hansis et al., 2015;Gasser et al., 2020), which account for changes in all carbon pools affected by LULCC. Therefore, we compared the NBE (excluding FFC emissions) and atmospheric growth rate (AGR) between this 225 study and GCP2020. The interannual changes of global NBE and AGR of this study match well with the estimates of GCP2020, with CORR of 0.75 and 0.88, BIAS (GCAS2021 minus GCP2020) of 0.15 and 0.25 PgC yr -1 , and MAE of 0.51 and 0.40 PgC yr -1 , respectively. For the prior fluxes, the CORR, BIAS, and MAE of NBE and AGR compared against the GCP2020 estimates are 0.16 and 0.49, -0.51 and 0.09 PgC yr -1 , and 0.63 and 1.10 PgC yr -1 ( Figure S2). These indicate that the estimate of global carbon budgets has been significantly improved after constrained by the GOSAT retrievals. of NEE constraint with GOSAT XCO2 agrees well with a recent regional inversion using surface CO2 and 14 CO2 measurements, which also shown significant sources over western US and sinks over central and eastern US (Basu et al., 2020). By using the Community Land Model (CLM5.0) and a Data Assimilation Research Testbed (DART) that assimilated with remotely sensed 240 observations of leaf area and above-ground biomass, Raczka et al. (2021) simulated the NEE over western US and also found that there are large areas with carbon release. The western US is dominated by natural lands, which is particularly vulnerable to forest mortality from droughts, insect attacks, and wildfires, Ghimire et al. (2015) found large carbon release legacy from bark beetle outbreaks across western US. In addition, the ageing and decline of forest may be another reason for the carbon release in western US (Sleeter et al., 2018). The significant sources of NEE in the grasslands of Africa are consistent with 245 previous top-down estimates based on satellite retrievals (Palmer et al., 2019) and surface CO2 measurements (Valentini et al., 2014). Many observations based on the eddy covariance also reported carbon sources of NEE in the savanna grassland of West and South Africa (e.g., Veenendaal et al., 2004;Räsänen et al., 2017;Quansah et al., 2015). The significant increase of carbon release in the grasslands of Africa may be related to the underestimates of carbon emissions from small fires in GFED 4s. The FIRE emission in GFED 4s was estimated based on global burned area, which were from coarse spatial-resolution sensors. 250 Ramo et al. (2021) shown that coarse sensers are unsuitable for detecting small fires that burn only a fraction of a satellite pixel, and pointed out that the FIRE emission of Africa in GFED 4s was underestimated by about 31% in 2016. Table 2 lists the aggregated mean posterior annual NEE, NBE and FIRE emissions during the 1-years for the 11 TRANSCOM regions and the 10 RECCAP2 regions. Compared with the prior NEE, the absolute relative changes in most TRANSCOM regions are greater than 50% ( Figure S4) after constrained with GOSAT data. In all regions, the aggregated 255 posterior NEE are negative, indicating a carbon sink in each region. For the 11 TRANSCOM regions, we estimate that Europe has the strongest NEE, followed by boreal Asia, tropical S. America, and northern Africa has the weakest NEE. Among the 10 RECCAP2 regions, Russia's NEE is the strongest, followed by N. America and Europe, and West Asia's NEE is the weakest.

Global carbon budgets 215
It is worth noting that the Europe's NEE in the TRANSCOM region is twice that in RECCAP2. This is because the coverage of Europe is different in TRANSCOM and RECCAP2, the former includes the entire European continent, while the latter does 260 not include European Russia. which was calculated by taking the sum of the carbon-stock change and lateral carbon fluxes from crop and wood trade, and riverine-carbon export to the ocean. Figure 6b shows a comparison between this study and Ciais et al. (2021). Although the inverted NBE is not completely equivalent to the land sink obtained by the bottom-up method, generally, to reconcile topdown and bottom-up results, the inverted NBE should be adjusted with the lateral transport of reduced carbon compounds (RCC) and carbon release from net imported products (Ciais et al., 2008;Jiang et al., 2016). Overall, except for Africa and 285 South Asia, the NBE estimated in this study and Ciais et al. (2021) -Luijkx et al., 2015;320 Doughty et al., 2015;Gatti et al., 2014). In 2019, the decrease of NEE may be related to the Indian Ocean Dipole event, which has significantly reduced the carbon uptakes over southern China, Indo-China peninsula, and Australia (Wang et al., 2021b).

Interannual variations and seasonal cycles
In SL, due to the small land area, its NEE is an order of magnitude lower than the other two regions. It could be found that there is a continuous decreasing trend. On average, the NEE in NL, TL, and SL during this decade are -2.33 ± 0.35, -1.25 ± 0.38, and -0.05 ± 0.07 PgC yr -1 , which account for 62.6%, 33.4% and 1.4% of the global total land NEE, respectively, indicating 325 that the global land NEE is dominated by the NEE in NL. However, the correlation coefficients for the IAVs of NEE between these three regions and the global land are 0.57, 0.86, and 0.37, respectively, indicating that the IAV of global NEE is dominated by its inter-annual changes in TL.
In Figure 8, we further present the IAVs and seasonal cycles of NEE in the 11 TRANSCOM regions. Since there are some overlaps between the TRANSCOM and RECCAP2 regions, for example, the N. America region in RECCAP2 is almost the anthropogenic CO2 emissions, which may affect the performance of the MOZART model. In the Amazon basin, the simulated CO2 profiles also agree well with the observations, with BIAS and MAE basically lower than 1 and 1.5 ppm, respectively.
Except for the lowest level at ALF, all BIAS are positive, with a total average of 0.2 ppm at 1000 ~ 1500 m heights, indicating that the strong carbon sink in tropical S. America estimated in this study are reasonable. Figure 13 shows the comparisons against the HIPPO observations at different heights and latitudes. Overall, most BIAS 430 are within ±0.5 ppm, showing a good agreement between the simulations and observations. Relatively large BIAS occurs over northern high latitudes, which is consistent with the comparisons against the surface observations as shown in Figure 10, and also reveals an overestimation of carbon releases at high latitudes.

Summary
A global NEE dataset is essential for estimating the regional terrestrial carbon budget and understanding the responses of 435 carbon fluxes on extreme climates. Here, by assimilating the GOSAT ACOS v9 XCO2 product, we generate a ten-year global monthly terrestrial NEE dataset from 2010 to 2019 (GCAS2021) using the GCASv2 system. GCAS2021 includes monthly and annual gridded (1°×1°) prior and posterior NEE and OCN flux, and prescribed FIRE and FFC emissions, and globally, latitudinally, and regionally aggregated fluxes and their uncertainties. Globally, the decadal mean NBE and AGR as well as their IAVs match well with the estimates of GCP2020. Regionally, our product shows carbon sinks over eastern N. America, 440 Amazon, Congo Basin, Europe, boreal forests, southern China, and southeast Asia, and carbon sources over western US, African grasslands, Brazilian plateau, and parts of South Asia. In the 11 TRANSCOM land regions, the NBEs of temperate N.
America, northern Africa and boreal Asia are between the results of CMS-Flux NBE 2020 and CT2019B, and those in temperate Asia, Europe, and tropical Asia are very close to CMS-Flux NBE 2020 but significantly different from CT2019B.
In the RECCAP2 regions, except for Africa and South Asia, the NBEs are comparable with the latest bottom-up estimate of 445 Ciais et al. (2021). The IAVs and seasonal cycles of NEE could clearly reflect the impact of extreme climates or large-scale climate anomalies. We also qualitatively evaluate the NEE estimates by comparing posterior CO2 concentrations with independent CO2 measurements from surface flask and aircraft CO2 observations, and the results show that the simulated remote site and regional average CO2 concentrations, as well as the vertical CO2 profiles, are all consistent with the observations. We believe that this dataset will be useful in the estimates of regional or national-scale terrestrial carbon budgets, 450 the study of carbon sink evolution mechanisms, the evaluation of ecosystem models, and the assessments of carbon neutrality strategies.