An 11-year record of XCO 2 estimates derived from GOSAT measurements using the NASA ACOS version 9 retrieval algorithm

. The Thermal And Near infrared Sensor for carbon Observation – Fourier Transform Spectrometer (TANSO-FTS) on the Japanese Greenhouse gases Observing SATellite (GOSAT) has been returning data since April 2009. The version 9 (v9) Atmospheric Carbon Observations from Space (ACOS) Level 2 Full Physics (L2FP) retrieval algorithm (Kiel et al., 2019) was used to derive estimates of carbon dioxide (CO 2 ) dry air mole fraction (XCO 2 ) from the TANSO-FTS measurements collected over its ﬁrst 11 years of operation. The bias correction and quality ﬁltering of the L2FP XCO 2 product were evaluated using estimates derived from the Total Carbon Column Observing Network (TCCON) as well as values simulated from a suite of global atmospheric inversion systems (models) which do not assimilate satellite-derived CO 2 . In addition, the v9 ACOS GOSAT XCO 2 results were compared with collocated XCO 2 estimates derived from NASA’s Orbiting Carbon Observatory-2 (OCO-2), using the version 10 (v10) ACOS L2FP algorithm. These tests indicate that the v9 ACOS GOSAT XCO 2 product has improved throughput, scatter, and bias, when compared to the earlier v7.3 ACOS GOSAT product, which extended through mid 2016. Of the 37 million soundings collected by GOSAT through June 2020, approximately 20 % were selected


Introduction
A new era of dedicated satellite observations of greenhouse gases began in 2009, with the successful launch of GOSAT (Kuze et al., 2009). Each day, GOSAT's Thermal And Near infrared Sensor for carbon  form Spectrometer (TANSO-FTS) acquires approximately 10 thousand high-spectral-resolution measurements of reflected sunlight ( 36.5×10 6 in 10 years). Soundings that are determined to be sufficiently clear of clouds and aerosols are processed by retrieval algorithms to produce estimates 10 of XCO 2 . Both the quality of the GOSAT TANSO-FTS spectra and the derived XCO 2 estimates have been continually refined over the past 12 years. While the official GOSAT L2 products are available from the National Institute for Environmental Studies (NIES; http://www.gosat.nies.go. 15 jp/en/about_5_products.html, last access: 10 January 2022; Yoshida et al., 2013) a number of independent research in-stitutes have developed their own products (e.g., Butz et al., 2011;Crisp et al., 2012;Cogan et al., 2012;Heymann et al., 2015).
One of these groups, the Atmospheric CO 2 Observations from Space (ACOS) team, used a Level 2 Full Physics (L2FP) retrieval algorithm developed for the NASA Orbiting Carbon Observatory (OCO) to derive estimates of XCO 2 from the GOSAT data (O'Dell et al., 2012;Crisp et al., 2012). Early XCO 2 estimates from these efforts had large biases and random errors when compared to XCO 2 estimates from 10 the Total Carbon Column Observing Network (TCCON) and other standards. For example, the v2.8 ACOS GOSAT L2FP product had biases of 7 to 8 ppm relative to TCCON (Crisp et al., 2012). These biases were reduced to 1-2 ppm in the v2.9 product. The next major release was v3.5 in 2014, which 15 spanned approximately 4 years. This data product showed additional reductions in bias and scatter against TCCON, as well as reasonable agreement in seasonal cycle phase and amplitude (Lindqvist et al., 2015;Kulawik et al., 2016).
These early space-based XCO 2 products were rapidly 20 adopted by the carbon cycle science community. Early studies based on GOSAT ACOS retrievals included Basu et al. (2013), Deng et al. (2014), Chevallier et al. (2014), andFeng et al. (2016). These studies provided the first comprehensive insights into regional flux estimates from space-based obser- 25 vations of carbon dioxide. Houweling et al. (2015) conducted an extensive inter-comparison of the early GOSAT-based atmospheric inversion system studies and reported a reduction in the global land sink for CO 2 and a shift in the terrestrial net uptake of carbon from the tropics to the extratropics. How- 30 ever, these studies also highlighted the role of spatiotemporal systematic errors in the satellite retrievals and the negative impact they can have on estimation of CO 2 sources and sinks using atmospheric inversion systems. Motivated by these early studies, as well as the launch of 35 the OCO-2 sensor in July 2014, the ACOS team continued to refine the L2FP retrieval algorithm. In 2016, the ACOS GOSAT v7.3 product was distributed. No formal results of the XCO 2 estimates were published by the algorithm team, although internal analysis showed small improvement over 40 v3.5, as well as an extension of the record to 7 years. A number of atmospheric inversion studies were published using the v7.3 product. For example, Chatterjee et al. (2017) and Liu et al. (2017)  Niño on the tropical carbon cycle. Palmer et al. (2019) used this data product in a global study, concluding that the tropical land regions were a net annual source of CO 2 emissions, including unexpectedly large net emissions from northern tropical Africa. Wang et al. (2019) found that the ACOS 50 GOSAT v7.3 XCO 2 yielded a stronger carbon land sink than the v7 OCO-2 product. Byrne et al. (2020) used the ACOS GOSAT 7.3 product to study interannual variability in the carbon cycle across North America, and Jiang et al. (2021) investigated interannual variability of the carbon cycle across 55 the globe with v7.3. Most recently, the v9 ACOS L2FP retrieval algorithm, first applied to OCO-2 (Kiel et al., 2019), was used to generate estimates of XCO 2 from an 11-year record of GOSAT measurements, spanning April 2009 through June 2020. This 60 both extends the time record over v7.3 and produces an ACOS GOSAT product that is more directly comparable to the newest OCO-2 product, which is now using version 10. The paper is organized as follows: Sect. 2 discusses the GOSAT TANSO-FTS instrument and measurements as re-65 lated to the ACOS XCO 2 estimates. In Sect. 3, updates to the ACOS v9 L2FP algorithm are detailed, and an assessment is given of the v9 XCO 2 data product volume. The XCO 2 quality filtering and bias correction procedures, specific to ACOS GOSAT v9, are also discussed. Section 4 provides an evalu-70 ation of the v9 XCO 2 product using estimates of XCO 2 from TCCON and from a suite of four atmospheric inversion systems (models). In addition, a comparison to collocated XCO 2 estimates derived from NASA's OCO-2 sensor is presented. A summary of the results is provided in Sect. 6. 75

The GOSAT instrument and measurements
The GOSAT mission is a joint project between the Japan Aerospace Exploration Agency (JAXA), the National Institute for Environmental Studies (NIES), and the Ministry of the Environment (MOE) (Kuze et al., 2009). GOSAT was 80 launched on 23 January 2009 into a sun-synchronous orbit with a local overpass time of approximately 12:49 and a 3 d ground repeat cycle. Its TANSO-FTS collects highresolution spectra of reflected sunlight that can be analyzed to yield estimates of carbon dioxide (CO 2 ) (Yoshida et al., 85 2011(Yoshida et al., 85 , 2013.

GOSAT TANSO-FTS instrument
TANSO-FTS collects high-resolution spectra of reflected sunlight in the near-infrared (NIR) and shortwave-infrared (SWIR) spectral ranges that include the oxygen A-band near 90 0.76 µm (ABO2 band) at approximately 0.36 cm −1 spectral resolution, and weak and strong CO 2 absorption features near 1.6 µm (WCO2 band) and 2.0 µm (SCO2 band), respectively, at 0.27 cm −1 spectral resolution. All three channels simultaneously measure two orthogonal components of po-95 larization approximately every 4.6 s.
Each GOSAT sounding has a circular ground footprint with a diameter of approximately 10.5 km when viewing the local nadir. An agile, two-axis pointing system allows cross-track and along-track motions of ± 35 • and ± 20 • , re-100 spectively. Before August 2010, a five-point cross-track scan was used, yielding footprints that were separated by approximately 150 km in both the down-track and along-track dimensions. Since that time, a three-point cross-track scan has 4 T. E. Taylor et al.: ACOS GOSAT v9 XCO 2 been used, yielding footprint separation of approximately 260 km (Kuze et al., 2016).
Over water, the TANSO-FTS scan mechanism targets the field of view to collect observations in the direction of the local glint spot, where sunlight is specularly reflected from the 5 surface. Early in the mission, glint observations were collected only within ± 20 • of the sub-solar latitude. In May 2013, to increase the latitudinal extent of the GOSAT ocean measurements, the scanning strategy was improved to better track the actual specular glint spot, which varies by lati- 10 tude and season. The latitude range for glint observation was further extended three times in increments of 3 • in September 2014, June 2015, and January 2016, by not only tracking the exact specular point but also tracking along the principal plane of the specular reflection when the glint spot was out 15 of range of the scan mechanism. In addition, more observations over fossil fuel emission target sites such as mega-cities and power plants have been made in recent years, allowing for detailed emission source studies (e.g., Kuze et al., 2020). Daily observation patterns can be found at https://www.eorc. 20 jaxa.jp/GOSAT/currentStatus_10.html (last access: 10 January 2022).
The TANSO-FTS detectors can be read out using independent medium-gain and high-gain signal chains. Most measurements over land use the instrument's high-gain signal 25 chain (H-gain), while brighter land surfaces are measured using the medium-gain signal chain (M-gain) to avoid saturating the detectors. Over oceans, which appear dark in the SWIR spectral bands, measurements are collected using the high-gain signal chain to maximize the signal. 30 During the first 7 years of GOSAT operations (2009)(2010)(2011)(2012)(2013)(2014)(2015), data acquisition was temporarily suspended due to one spacecraft and two instrument anomalies, as highlighted in Kuze et al. (2016). A rotation failure of a solar paddle in 2014 resulted in a data loss of 6 d. A switch from the primary 35 to secondary pointing mirror in January 2015 resulted in a data loss of approximately 6 weeks, while a temporary shutdown of the cryocooler in August 2015 resulted in a data loss of 13 d.
Since 2015, three additional anomalies interrupted data 40 acquisition. An unexpected shutdown of the instrument occurred in May 2018, resulting in the loss of a week of data. A failure of the second solar panel caused a significant loss of data spanning more than a month in November and December of 2018, and an anomaly of the FTS alignment laser 45 caused a loss of a week of data in June of 2020. In all these cases, the system was able to recover full functionality either through utilization of on-board back-up systems, or through mitigation strategies, and as of the summer of 2021, TANSO-FTS continues to collect science data. 50

ACOS GOSAT v9 L1b measurements
The JAXA L1b algorithm, which has been updated more than 10 times over the 11-year data record, produces an internally consistent set of geometrically, radiometrically, and spectrally calibrated TANSO-FTS radiances. The raw spectral 55 measurements are interferograms, which are calibrated and Fourier transformed to yield spectra. The version 205/210 Level 1b (L1b) geolocated and calibrated radiances provided by JAXA have been used for the ACOS v9 reprocessing. A list of L1b updates for v205/210 can be found in Table 3 of 60 the ACOS v9 Data Users Guide (DUG) (O'Dell et al., 2020).
Note that while the current L1b version is now 230, the only differences between this version and 205/210 are in the thermal infrared band (5.6-14.3 µm), which is not used in the ACOS XCO 2 retrieval.

65
After obtaining the calibrated L1b product from JAXA, the ACOS team converts the files to the format needed as input to the ACOS L2 algorithms. The L2FP algorithm uses a simple average of the S and P linear polarizations to produce an approximation of the total measured intensity. Due to co-70 operation agreements between JAXA and the California Institute of Technology, the distribution of the ACOS GOSAT L1b product is restricted and therefore not publicly available on the NASA DISC. However, the data may be procured by submitting a request to the GOSAT project.

The ACOS v9 L2FP XCO 2 retrieval algorithm
The ACOS Level 2 full physics (L2FP) retrieval algorithm is well documented, most recently in O'Dell et al. (2018) for v8 and in Kiel et al. (2019) for v9. A Bayesian optimal estimation framework is used to derive estimates of XCO 2 from 80 spectral measurements of reflected solar radiation. A postprocessing step assigns a simple good/bad quality flag (QF) to each XCO 2 value based on successful L2FP algorithm convergence and a series of empirically derived filters. An empirical bias correction (BC) to the estimated XCO 2 val-85 ues, derived from comparisons with TCCON-derived XCO 2 and CO 2 fields from a suite of atmospheric inversion systems, is included in the Lite File product. Here we provide a summary of the recent evolution of the ACOS algorithm and discuss retrieval parameters and setup specific to GOSAT.   Thompson et al., 2012) in ACOS v7 to ABSCO v5.0 (Oyafuso et al., 2017) in ACOS v8/9. The ACOS v9 L2FP algorithm is unmodified relative to v8 (Kiel et al., 2019). However, changes were made in v9 regarding the sampling of the meteorological prior, which does af-100 fect ACOS GOSAT estimates of XCO 2 . The source of the prior meteorology was switched from the European Center for Medium-range Weather Forecast (ECMWF) in ACOS v7, to the NASA Goddard Modeling and Assimilation Of-fice (GMAO) Goddard Earth Observing System (GEOS) Forward Processing -Instrument Team (FP-IT) product for ACOS v8/9. Both v7 and v8/9 used aerosol priors based on a simple monthly 1 • latitude by 1 • longitude climatology constructed from the output aerosol fields of the GMAO Modern-Era Retrospective analysis for Research and Applications (MERRA) product (Rienecker et al., 2011). However, between v7 and v8/9, an additional stratospheric aerosol layer was introduced, as described in Sect. 3.1.1 of O'Dell et al. (2018). In addition, the prior value of the aerosol opti-10 cal depth (AOD) for each retrieved aerosol type was lowered from 0.0375 in ACOS v7 to 0.0125 in ACOS v8/9 based on extensive testing. There was no change in the source of the CO 2 prior from ACOS v7 to v8/9; both versions adopted the prior developed by the TCCON team for use in the ggg2014 15 algorithm . An additional change from ACOS v7 to v8/9 was a switch from a purely Lambertian land surface model to a more sophisticated bi-directional reflectance distribution function (BRDF) model. (ZLO) is fit in the state vector to account for non-linearity in the ABO2 signal chain on GOSAT TANSO-FTS (Crisp et al., 2012).
To support comparisons of the ACOS GOSAT v9 XCO 2 product with the OCO-2 v10 product, Table 1 includes the 30 most recent updates to the ACOS v10 L2FP algorithm. For v10, the ABSCO tables were again updated from v5.0 to v5.1 (Payne et al., 2020). The aerosol prior was updated from the MERRA monthly climatology to daily GEOS-FT-IT values, with a tightened prior uncertainty (Nelson and O'Dell, 2019). 35 Finally, the CO 2 priors developed by the TCCON team for use in ggg2014 were updated to a revised set of priors developed for use in ggg2020.

ACOS GOSAT v9 L2FP sounding selection and convergence 40
GOSAT data from 20 April 2009 through 30 June 2020 were passed through the ACOS L2FP algorithm pipeline, which includes a series of stages where soundings can be rejected or selected for further processing. The throughput of each of these stages for ACOS GOSAT v9 is summarized in Table 2 and Fig. 1. The pipeline begins with a series of preprocessing steps, which reject corrupted spectra and screen the remainder to eliminate those with optically thick clouds and/or aerosols . From the full set of measurements (panel a of Fig. 1), the remaining soundings are ac-50 cepted by the L2FP algorithm (18.8 % of the 37.4×10 6 measured soundings contained in the ACOS GOSAT v9 record) (panel b of Fig. 1) and a retrieval of XCO 2 is attempted. The majority of the selected soundings successfully converge to a valid solution: 87 % for ACOS GOSAT v9 (16.4 % of the 55 total measured soundings). Soundings can fail to converge for a variety of reasons, including (i) producing non-physical values, such as negative gas mixing ratios or surface pressures (3.9 % of the selected), (ii) converging too slowly and exceeding a predefined number of iterations (3.2 % of the se-60 lected), or (iii) having more diverging steps than the predefined maximum (5.9 % of the selected). The 6.1 × 10 6 valid soundings were then run through the quality filtering and bias correction procedure discussed in the next section.
3.3 ACOS GOSAT v9 XCO 2 quality filtering and bias 65 correction All GOSAT soundings that converged to a valid XCO 2 value within the L2FP retrieval were input to the quality filtering and bias correction procedure. A modest fraction (4.5 % of the valid soundings) were removed from the final L2Lite 70 product based on screening via the IMAP-DOAS Preprocessor (IDP) CO 2 ratio, which indicated the presence of clouds or aerosols. Based on a series of screening criteria derived from comparisons with TCCON and modeled CO 2 fields, each sounding that converged within the L2FP is assigned 75 either a "good" (= 0) or "bad" (= 1) XCO 2 quality flag. Generally, for global or regional studies, it is recommended that users retain only the "good" quality soundings, as the soundings flagged as "bad" quality are likely to include biases that compromise their utility for some applications. A global map 80 of the ACOS GOSAT v9 "good" XCO 2 sounding density is provided in panels C and D of Fig. 1. A subset of data variables from the per-orbit L2Std files (OCO-2 Science Team et al., 2019b), along with the quality filter flag and biascorrected XCO 2 , are repackaged into the daily aggregated 85 L2Lite NetCDF files (OCO-2 Science Team et al., 2019a). A fundamental aspect of the quality filtering and bias correction procedures (QF/BC) is the need for XCO 2 truth metrics with which to compare the satellite-derived estimates (O'Dell et al., 2018). The development of ACOS GOSAT 90 v9 used XCO 2 truth metrics derived from both TCCON measurements and the median CO 2 distributions determined from a suite of four atmospheric inversion systems, which do not assimilate satellite CO 2 measurements.
TCCON is a well-established validation transfer stan-95 dard for space-based estimates of XCO 2 (Wunch et al., 2011a(Wunch et al., , 2017b. For the ACOS GOSAT v9 QF/BC, estimates of XCO 2 derived from TCCON measurements using the ggg2014 retrieval algorithm were used . Individual GOSAT soundings were compared to TC-100 CON daily mean XCO 2 values. TCCON data were included if (i) they were flagged as good (flag = 0), (ii) they fell within 3 standard deviations of a daily quadratic fit against time (to remove outliers, e.g., due to unscreened cloud), (iii) they covered at least 15 min within a given day, (iv) there were at 105 least three good soundings within the day, and (v) the stan-    dard deviation of the good soundings for the day was less than 3 ppm. In the GOSAT-TCCON comparisons described here, an averaging kernel correction was applied to each TC-CON XCO 2 estimate following Nguyen et al. (2014), prior to calculating the daily mean value.

5
Following the criteria defined in Wunch et al. (2017b), the spatial collocation criteria for GOSAT soundings were those falling within ±2.5 • latitude and ±5 • longitude of a TCCON station for most sites. For the Southern Hemisphere (SH) sites poleward of 25 • S latitude, where the variation of 10 CO 2 is low, the spatial box was increased to ±10 • latitude by 20 • longitude to increase the number of collocations. For the Edwards TCCON station, which lies in an arid region just north of the polluted Los Angeles metropolitan area, a very specific collocation box of [34.68, 37.46]  Gabriel Mountains, and regions too far outside of the Los Angeles basin with the Caltech TCCON data. Finally, only GOSAT soundings acquired within ±2 h of the mean TC-CON measurement time were considered. For the quality filtering and bias correction procedure, single sounding level 25 collocations are used to maximize the number of fit points.
Estimates of CO 2 from atmospheric inversion systems, or models, provide a useful metric for evaluating satellite-based estimates of XCO 2 (O'Dell et al., 2018). In this work, a suite of four models (CarbonTracker, CAMS, CarboScope, 30 and University of Edinburgh) were sampled at the GOSAT sounding times and locations. Brief descriptions of each, along with references, are provided in Table 3. The models use a variety of land biosphere prior fluxes, inverse solvers and transport models, and assimilate CO 2 data only from 35 flasks and continuous analyzers on a wide variety of platforms, e.g., observatories, towers, aircraft, and ships. Specifically, no data from GOSAT, OCO-2, or TCCON are assimilated. The CO 2 concentration fields of the models capture the known features of the global atmospheric CO 2 distri-40 bution, including seasonality, time trends, and inter-annual variability (IAV) due to El Niño-Southern Oscillation. For The fraction of the total soundings selected to run through the L2FP algorithm (b). The fraction of the total soundings that converged in the L2FP and were assigned a good L2FP QF (c). The sounding density of the good QF data per 2.5 • by 5 • latitude-longitude grid cell (d).
each GOSAT sounding, the vertical profiles of CO 2 from the corresponding grid box of each of the four models are spatiotemporally interpolated (linear in latitude, longitude, and time) to the GOSAT observation point, and the GOSAT averaging kernel is applied to each vertical profile to produce a 5 modeled XCO 2 as if viewed from the satellite.
For each GOSAT sounding, a multi-model median (MMM) XCO 2 was calculated from the models having a valid XCO 2 estimate for that location and time. Unless otherwise noted, the model XCO 2 is taken to be that which a 10 perfect OCO-2 would have observed, XCO 2,ak ; that is, an averaging kernel correction is applied to account for differences between the model profile of CO 2 and the ACOS prior in the unmeasured part of the profile: 15 where h i is the pressure weighting function on the i = 1. . .20 ACOS model levels, a is the normalized ACOS averaging kernel for CO 2 , u m is the model profile of CO 2 , and u a is the ACOS prior profile of CO 2 . To exclude outliers, models with XCO 2 that deviated more 20 than ±1.5 ppm from the initial MMM for that sounding were not included. The sounding was then rejected if more than one of the four models had been excluded, or if the standard deviation amongst the valid models was > 1 ppm. Approximately 85 % of the GOSAT v9 soundings with a good L2FP 25 quality flag had a valid MMM XCO 2 value for analysis. The regions with the highest fraction of rejections occur along the Southern Ocean (latitude −60 • ), the Amazon and Congo rain forests, and a broad region across northern Asia. Table 4 lists the model version numbers used for the QF/BC procedure, as 30 well as that used in the evaluation of the final good-quality XCO 2 product that will be presented later. Table 5 lists the quality filtering variables used for ACOS GOSAT v9 and their corresponding thresholds. Many of the same variables (18 out of 31) were also used in the OCO-2 v9 35 quality filtering, as seen in Table 5 of Kiel et al. (2019). This includes the IDP CO 2 and H 2 O ratios (Frankenberg et al., 2005), and the A-band preprocessor dP , i.e., the difference between the retrieved and prior surface pressure from the oxygen A band . Another common vari-40 able used for quality filtering is the perturbation in the L2FP CO 2 vertical profile relative to the prior, a quantity called "CO 2 grad del" (δ ∇ CO 2 ), as defined in Eq. (5) of O'Dell et al. (2018). A number of aerosol-related retrieval parameters are also used, similar to OCO-2 v9. Section 2.5 of the ACOS 45 GOSAT v9 DUG provides additional details on the quality filtering (O'Dell et al., 2020).
Spurious correlations in the estimates of XCO 2 with other retrieval variables due to inadequacies in the modeled physics motivate the application of a bias correction (Wunch 50 et al., 2011b;O'Dell et al., 2018). Generally such spurious correlations are found with state vector elements such as retrieved surface pressure, various aerosol parameters, and δ ∇ CO 2 . For each sensor there are also typically offsets by viewing mode, e.g., land glint versus ocean glint, which are 55 accounted for via the bias correction. A general discussion of the ACOS XCO 2 bias correction methodology is provided in Sect. 4 of O' Dell et al. (2018), where the fundamental equation is defined as  where C P is the mode-dependent parametric bias, C F is a footprint-dependent bias for footprints 1. . .8, and C 0 represents a mode-dependent global scaling factor. Note that for GOSAT there is no footprint-dependent bias correction term, as is necessary for OCO due to low-level calibration errors 5 across the detector frame. Further, to be consistent with previous ACOS GOSAT data versions, the global divisor is replaced by an additive offset, which is effectively the same because the range of XCO 2 variability (∼ 20 ppm) is small relative to the mean XCO 2 (∼ 400 ppm).

10
The explicit formula for application of the ACOS GOSAT v9 correction is provided in Sect. 2.5.6 of the DUG (O'Dell et al., 2020). For both land H-gain and M-gain, a set of five BC variables are used, while ocean H-gain uses only three variables. The difference between the H-gain and M-gain 15 bias correction over land is minor. New for ACOS GOSAT v9 is the use of a correction against time, which is made possible with an 11-year data record; the corrections are 0.05 ppm/yr over land and 0.10 ppm/yr over water. The source of this spurious drift in the bias-corrected XCO 2 is currently unclear 20 and is the subject of ongoing study. Although there is some commonality in the quality filtering and bias correction variables used for ACOS GOSAT v9 (compare Tables 5 and 6), they do differ somewhat, as is typically the case with each sensor and data version. Table 6 compares the bias correction variables used for ACOS GOSAT v9 with the variables used in the previous ACOS GOSAT v7.3, as well as with OCO-2 v9 and v10. The same few variables have appeared in all recent versions, including L2FP δ ∇ CO 2 , L2FP dP , and L2FP DWS for land 30 soundings. For ocean soundings the bias correction variables have evolved, with the only common one being δ ∇ CO 2 . Table 7 summarizes the effect of the quality filtering and bias correction on the ACOS GOSAT XCO 2 for v7.3 and v9. For ocean H-gain soundings, the v9 quality flag is substan-35 tially more restrictive compared to v7.3, i.e., 57 % pass rate compared to 78 %. This is mostly driven by the more extensive latitudinal coverage in the v9 record, which tends to include more soundings with high solar zenith angles (SZA) and low signal-to-noise ratio (SNR), which are more chal-40 lenging for the L2FP. For H-gain land observations, the two versions have quite similar QF pass rates ( 35 %-45 %). The QF pass rate for v9 M-gain land data is 39 % when compared against models but 56 % against TCCON. In all cases there is a significant reduction in the scatter of the 45 XCO 2 after application of the QF/BC: by a factor of 2 for ocean H-gain and land M-gain and a factor of 3 for land H- Table 5. ACOS GOSAT v9 L2FP quality filtering variables and thresholds. Descriptions of the variables can be found in the DUG (O'Dell et al., 2020). Soundings falling outside of the data ranges are assigned a bad XCO 2 quality flag. The second column identifies variables that were also used for OCO-2 v9 quality filtering, as taken from  gain. The QF/BC scatter is always slightly lower for v9 compared to v7.3, even though the number of soundings is greater by 1.5 to 10 times for the various scenarios. Figure 2 shows the relative magnitudes of the bias correction on the good-quality soundings by season, aggregated 5 to 2.5 • latitude by 5 • longitude. The global median bias of −1.8 ppm has been removed for clarity. This highlights gradients and contrasts in the bias correction, which are of importance as gradients in CO 2 concentrations are the primary driver of CO 2 fluxes in atmospheric inversion systems. 10 In general, the bias correction is necessary to remove spurious contrasts between land and ocean XCO 2 values. The strongest relative bias corrections are positive adjustments over the bright land surfaces in M-gain viewing mode, specifically the Sahara in DJF and JJA and Australia in DJF. The 15 land H-gain observations have a mix of relative bias correction values, ranging from mildly negative over high northern latitudes in JJA to moderately positive over northern midlatitudes in JJA in the western United States and the Middle East. Most of the ocean H-gain observations have a mildly 20 negative relative bias correction, with some mild positive values in the southern tropical oceans in DJF.

Evaluation of ACOS GOSAT v9 XCO 2
The ACOS GOSAT v9 XCO 2 record was characterized in five ways: (i) an analysis of the XCO 2 "good-quality" data 25 volume, (ii) a spatiotemporal analysis of the XCO 2 estimates, (iii) a validation against XCO 2 estimates from TCCON, (iv) a comparison to XCO 2 derived from models, and (v) a compar-   ison with collocated XCO 2 estimates from the OCO-2 v10 product.

ACOS GOSAT v9 "good-quality" data volume
It is instructive to compare the ACOS GOSAT v9 product to the earlier v7.3 product to highlight similarities and differ-5 ences in the quality filter screening. A time series histogram of the monthly throughput of the good-quality-filtered sound-ings for the v9 product compared to v7.3 is shown in Fig. 3. The soundings have been binned by month, with the three GOSAT observation modes displayed by color. The v7.3 10 product did not contain any land M-gain data in the L2Lite files (red in the figure) as the quality filtering and bias correction were not developed for that gain mode in v7.3 due to some unreconciled differences. An important feature of the v9 data record is the extension in time, which runs through 15 June 2020, compared to a termination date of June 2016 for  , there are some differences in the data volume for land H-gain and ocean H-gain observations. This is due to changes in both the details of the QF procedure, including changes in the variable thresholds used to assign 5 QF = good/bad, and to some differences in the convergence characteristics of the L2FP retrieval. Generally, v9 is producing up to 60 % more good-quality data than v7.3 near the end of the overlap period in 2016. There was a substantial increase in the number of good QF soundings from 2010 to 10 2019, due to the increased latitudinal range of the ocean observations as a result of improvements in the GOSAT pointing strategy, as well as improvements in the sounding selection for ACOS L2FP v9. Figure 4 shows sounding density Hovmöller plots compar- 15 ing ACOS GOSAT v7.3 (a) to v9 (b) with the three GOSAT observation modes combined. Again, the extended time period covered by v9 is evident. The increase in sounding density in the SH beginning in 2016 due to optimization of the GOSAT viewing strategy is prominent in the v9 product. This 20 feature is also seen in the spatial maps showing the fraction of good-quality soundings and the density per grid box, in panels (c) and (d) of Fig. 1, which was introduced in Sect. 3.2.
Persistently clear regions, such as the Sahara and the western part of Australia, have as many as 30 % of the observations 25 assigned a good-quality flag. Large regions of the tropical Pacific and Atlantic also contain a relatively high fraction of good-quality soundings. On the other hand, tropical forests and high latitudes in general have low yields of good-quality soundings. This is largely a combination of cloud contami-30 nation, dark surfaces at shortwave infrared wavelengths, and low solar illumination conditions, all three of which are problematic for retrieving CO 2 from space using reflected sunlight.
4.2 ACOS GOSAT v9 XCO 2 spatiotemporal analysis 35 There has been a steady increase in the atmospheric burden of CO 2 since the onset of the industrial age due mainly to the burning of fossil fuels (e.g., Keeling et al., 1995).  trend of approximately 2 % is understandable, given the significant differences in the spatiotemporal sampling of the two data sets. For the interested reader, a thorough comparison of satellite-and surface-derived growth rates in atmospheric CO 2 is given in Buchwitz et al. (2018).

5
The maps in Fig. 6 show the spatial distribution of XCO 2 at 2.5 • latitude by 5 • longitude resolution for 2010 (top) and 2019 (bottom), for DJF (left) and JJA (right). The dynamic range of the color scale in each case is 6 ppm. However, due to the secular increase in global CO 2 of 2.3 ppm per year, 10 the scale is centered 20 ppm higher in 2019 compared to 2010. The strong latitudinal gradients in XCO 2 are similar in these two seasons, while the zonal gradient tends to be weakest in MAM (not shown), just before the summer drawdown of CO 2 by the land biosphere begins. The increase in 15 the number of ocean H-gain soundings in the later part of the data record is also evident in these maps.
Qualitatively, the patterns in the maps look quite similar from 2010 to 2019, but with increased data coverage. In general, the highest concentrations of XCO 2 for the two selected 20 seasons are observed by GOSAT in the Northern Hemisphere (NH) during DJF, especially over northern tropical Africa (between 0 and 15 • N latitude), large portions of China, and the eastern United States. This stands to reason, as the atmospheric burden of CO 2 increases towards a peak during 25 NH winter due to inactivity of the land biosphere, coupled with strong anthropogenic CO 2 emissions. During DJF the ACOS GOSAT v9 XCO 2 exhibits relatively low concentrations across the entire SH, as would be expected if the Southern Ocean were a strong carbon sink (e.g., Gruber et al.,30 2019). In JJA, the XCO 2 is reduced over the mid-latitude and boreal forests, also expected behavior due to strong photosynthetic uptake of CO 2 during this season (e.g., Ciais et al., 2019).
A quantification of differences in the bias-corrected ACOS 35 GOSAT v9 XCO 2 data product relative to the v7.3 product is given in Fig. 7  , we see that the XCO 2 signal has an increasing tendency in time; i.e., the v9 XCO 2 increases more rapidly in time than v7.3. The cause of this effect is currently unknown, but it is partially due to the implementation in v9 of a time-dependent bias correction term of +0.05 ppm/yr for 60 land observations. This translates into an expected change of about 0.35 ppm in the v9 record over the 2009 to 2016 time span.
For the ocean H-gain observations at the single sounding level (panel d), the mean and standard deviation of the 65 XCO 2 are +0.28 and 0.79 ppm, respectively. When gridded and mapped at 2.5 • latitude by 5 • longitude resolution (panel e), the spatial distribution is fairly smooth, i.e., low variation in both latitude and longitude. Finally, when the data are gridded versus time and latitude (panel f), the mod-70 est variation in latitude is confirmed, but a substantial time dependence is observed, with the XCO 2 signal beginning negative in 2009 (v9 XCO 2 lower than v7.3), and switching to a positive XCO 2 signal by 2016 (v9 XCO 2 higher than v7.3). The time-dependent bias correction term for ocean H-75 gain observations was +0.1 ppm/yr. This translates into an expected change of about 0.7 ppm over the 2009 to 2016 time  span in the v9 record, accounting for some but not all of this time-dependent difference between v9 and v7.3. This direct comparison between the v9 and v7.3 XCO 2 product only allows for statements as to their differences. It does not allow one to deduce which is closer to truth. There-5 fore, an analysis of the v9 XCO 2 data product against truth metrics follows. Furthermore, it is difficult to accurately determine the effect that the new v9 XCO 2 product will have on atmospheric inversion system results relative to v7.3 without further detailed study.

ACOS GOSAT v9 XCO 2 versus TCCON
A list of TCCON stations used in this work, including basic physical information and data citations, is given in Table 8. For the evaluation against the ACOS v9 XCO 2 data, the single sounding collocations described in Sect. 3.3 were 15 aggregated into overpass mean values. Essentially the same TCCON data set was used for both the QF/BC procedure as for the evaluation, as no hold-over data were maintained. Also, as described in Sect. 3.3, an averaging kernel correction was applied to the TCCON data in order to fairly compare to 20 the satellite data. A one-to-one linear regression of the XCO 2 provides a simple quantification of the agreement, as shown in Fig. 8.
For ocean H-gain observations (Fig. 8a), the mean (µ) of the differences in XCO 2 ( XCO TCCON 2 = GOSAT − 25 TCCON) is essentially zero: 0.00 ppmCE1 for the singlesounding (SS) results and +0.01 ppm for the overpass mean (OPM) results. The corresponding standard deviations (σ ) are 1.08 and 0.82 ppm for the SS and OPM results, respectively. This indicates that roughly half of the SS er-30 ror variance is a result of instrument noise or other random high-frequency error sources (1.08 2 = 1.2 ppm versus 0.82 2 = 0.7 ppm). For land H-gain observations (Fig. 8b), µ = +0.10 ppm and +0.14 ppm for the SS and OPM, respectively. The land 5 H-gain σ values are higher than for ocean H-gain: 1.60 and 1.14 ppm for SS and OPM, respectively. Larger variations in XCO TCCON 2 are expected for land H-gain due to variability in topography and surface brightness, as well as higher likelihood of contamination by cloud and aerosol, all of which are 10 more challenging for the ACOS retrieval. Further, biology and atmospheric transport cause CO 2 signals to vary more over land regions, and in addition, instrument noise is higher because the SNRs tend to be lower.
Land M-gain observations have near-zero bias (µ = 15 −0.02 ppm and +0.02 ppm for SS and OPM, respectively) and scatter similar to that for ocean H-gain (σ = 1.09 and 0.84 ppm for SS and OPM, respectively), likely driven by lower variability in surface topography and brightness compared to land H-gain observations, as well as higher SNRs 20 over these bright land surfaces. The correlation in the XCO 2 between the data sets in all observation modes is high, with Pearson R 2 = 0.98, 0.98, and 0.99 for ocean H-gain, land H-gain, and land M-gain, respectively. Overall, these results indicate excellent agree-25 ment between the bias-corrected and quality-filtered ACOS GOSAT v9 XCO 2 product and collocated estimates from TC-CON. Figure 9 shows the mean absolute error (MAE) between the overpass mean collocated GOSAT and TCCON XCO 2 , 30 organized by latitude bins, season, and observation mode for v7.3 (panels a and b) and v9 (panels c, d, e). The calculation of the MAE and error bars follow the procedure reported in Chatterjee et al. (2013) (Eqs. 3 and 4). The error bars on the MAE represent the scatter around the mean. A smaller er-35 ror bar, or a lower scatter, implies that the MAE values are more consistent across a group of TCCON stations within a latitude band and season. The MAEs tend to be lower for v9 compared to v7.3, with smaller error bars and increased number of collocations. This is especially true for the SH ocean 40 H-gain data, where the MAE ranges from 0.4 to 0.7 ppm in v9 for all seasons, in contrast to v7.3, which had higher MAE ranging from 0.5 to 0.85 ppm in that region. In the v9 land Hgain data, the MAE is roughly a function of latitude, with the highest values ( 1.0 ppm) seen between 60-90 • N and the 45 lowest values ( 0.7 ppm) seen from 30-60 • S. This stands to reason as lower variability of XCO 2 in the SH tends to yield better agreement between satellite-and ground-based based CO 2 growth rates agreed generally better than 0.2 ppm per year. Here, we provide an update to those results using the 11-year ACOS GOSAT v9 XCO 2 data record. For this part of the analysis, a slightly more restrictive set of collocation criteria were implemented, compared to that described 25 in Sect. 3.3 for the BC/QF procedure and to that used to generate Fig. 8. The seasonal cycle analysis required that the TCCON record spanned at least one contiguous year (a full seasonal cycle) and that a minimum of 20 collocations with GOSAT occurred. In addition, the three GOSAT observation 30 modes (ocean H-gain, land M-gain, land H-gain) were combined for each site, and satellite overpass means of XCO 2 were aggregated into daily means. This resulted in approximately 7700 daily averages at 26 TCCON stations over the 11-year GOSAT data record. both GOSAT and TCCON. The mean seasonal amplitudes indicate a slight disagreement of a few tenths of a ppm, with TCCON showing a slightly higher fitted peak XCO 2 value during the spring maximum phase, compared to GOSAT. This is similar to the results for this site reported in Fig. 4   5 of Lindqvist et al. (2015). The time series of the calculated difference in satellite-and ground-based estimated XCO 2 (GOSAT − TCCON), shown in (c), highlights the magnitude of the scatter about the mean bias and suggests that there is no observable time drift in the data at this site. 10 A summary of the data from each station that met the seasonal cycle collocation criteria is provided in Table 9. In addition, the full complement of plots is presented in Appendix A. Overall, the seasonal cycle analysis at most sites is in agreement, to within the estimated uncertainties. The

ACOS GOSAT v9 XCO 2 versus models
The collocation and calculation of the multi-model-mean (MMM) is described in Sect. 3.3. Although the model data used for evaluation were very similar to those used in the QF/BC procedure, some minor version updates and exten-25 sions in time were included, as indicated in Table 4. It is important to be aware that there can be a considerable delay between performing the QF/BC procedure and the full generation of the final product, during which time the models are often updated. 30 Seasonal maps of XCO MMM 2 (GOSAT v9 minus MMM) are shown in Fig. 11 for the 11-year data record binned at 2.5 • latitude by 5 • longitude. Generally, the agreement between the model-derived values and the satellite estimates is    Table 9. Evaluation of the daily mean bias-corrected ACOS GOSAT v9 XCO 2 (all viewing modes combined) against collocated TCCON estimates for individual stations. There were 7547 d total for the 25 stations. The following sites/instruments were excluded from this part of the analysis due to inadequate time series or seasonal cycle coverage: Eureka, Four Corners, Indianapolis (Influx), JPL2007, Lauder1, Lauder3, Manaus, and Ny-Ålesund. The mean, standard deviation, and Pearson correlation coefficient (µ, σ , R 2 ) of the linear fit between GOSAT and TCCON are given in columns 3-5. The remaining columns quantify the seasonal cycle fit following the methodology described in Lindqvist et al. (2015). The bottom row provides mean summary statistics for the linear fit.  Ninõ produced an anomalously strong carbon release from tropical land regions due to higher temperature and below average precipitation (Liu et al., 2017). In contrast to the positive SH signal, negative XCO MMM 2 values (GOSAT lower 15 than MMM) have been observed in the v9 NH oceans since 2016. It is unclear why the satellite and models disagree over such large spatial and temporal scales, but recent work by Müller et al. (2021) suggests that the ACOS v7.3 (and to a lesser extent v9) XCO 2 values are in fact biased low by 20 approximately 1 to 1.5 ppm, as compared to a new independent evaluation data set generated from combined ship and aircraft measurements over the open oceans. Further investigation into the source of the ACOS GOSAT biases against models is warranted.  itive signal over northern tropical Africa in DJF. This feature was also observed in the OCO-2 v7 and v8 comparisons to a MMM (O'Dell et al., 2018) and in v10 XCO 2 anomaly maps (Hakkarainen et al., 2019).

ACOS GOSAT v9 XCO 2 versus OCO-2 5
NASA's Orbiting Carbon Observatory-2 (OCO-2) has been collecting science data since September, 2014 from a nearpolar low-Earth orbit (705 km altitude), with an afternoon Equator crossing time of 13:30 local time . Like GOSAT, OCO-2 takes measurements of reflected 10 solar radiation in the oxygen A-band (0.76 µm) and the weak and strong carbon dioxide bands (1.6 and 2.0 µm, respectively), which are used to estimate XCO 2 using the ACOS L2FP retrieval algorithm O'Dell et al., 2018). However, due to differences in the orbit parameters 15 of the two sensors, e.g., a 3 d repeat cycle for GOSAT versus a 16 d repeat cycle for OCO-2 (see Table 2 of Kataoka et al., 2017), the number of collocated soundings is somewhat limited. Therefore, some criteria must be defined in order to identify soundings that can be compared in a meaning-20 ful way. The underlying assumption of the collocation is that on scales of a few hundred kilometers and several hours, the natural variance in XCO 2 is not detectable in satellite-derived estimates from the ACOS L2FP algorithm.
For this study, the coincidence criteria to match OCO-2 25 soundings to individual GOSAT soundings were as follows: (i) falling within 2 • latitude and 3 • longitude, (ii) with a maximum spatial separation of 300 km, and (iii) acquired within ±2 h. Due to the dense nature of the OCO-2 soundings relative to the sparseness of the GOSAT soundings, there are 30 typically between zero and several hundred matched OCO-2 soundings per GOSAT footprint. A lower limit of 10 and an upper limit of 100 (randomly selected) OCO-2 soundings that meet the coincidence criteria were set in order to retain the GOSAT sounding for analysis. The individual L2FP  quality flags are applied for both GOSAT and OCO-2 during the collocation procedure, and then the mean value of XCO 2 from the 10 to 100 collocated OCO-2 soundings is calculated and subtracted from the corresponding GOSAT XCO 2 to produce XCO OCO−2 2 .

5
Here we compare ACOS GOSAT v9 against OCO-2 v10 (rather than to the deprecated v9), since we assume that science users will adopt the newest OCO-2 product. Major updates to the version 10 ACOS L2FP algorithm (Osterman et al., 2020) are discussed in Sect. 3.1. 10 One complexity in comparing ACOS GOSAT v9 and OCO-2 v10 is the fact that the two versions of the algorithm used different CO 2 priors. Typically, models which assimilate satellite CO 2 data take into account the unmeasured part of the prior CO 2 profile (specified via the retrieval's averag- 15 ing kernel) via an averaging kernel correction, as given in Eq. (1). Therefore, in order to fairly compare these two data sets as models would assimilate them, we need to remove their difference due to the unmeasured part of the CO 2 profile, as follows: where h is the XCO 2 pressure weighting function, a is the normalized XCO 2 averaging kernel, u a is the ACOS v9 CO 2 prior profile used for GOSAT, and u a is the ACOS v10 CO 2 prior profile used for OCO-2. The summation takes place 25 over the 20 vertical levels defined in the ACOS code. In summary, the total adjustment to the ACOS GOSAT XCO 2 value is calculated as the contribution of the difference in the vertical CO 2 priors at each level weighted by the one minus the averaging kernel at that level. The global mean adjustment 30 due to the CO 2 prior correction was approximately 0.2 ppm, with 95 % of corrections between −0.1 and +0.5 ppm. Spatial maps of XCO OCO−2 2 (GOSAT v9 -OCO-2 v10) for the prior-corrected, collocated soundings are shown in Figs. 14 and 15 for land and ocean, respectively. In each 35 figure the maps are shown by season at 2.5 • latitude by 5 • longitude resolution. In all seasons, higher scatter in XCO OCO−2 2 is observed over land ( 1 ppm) than over ocean (< 0.7 ppm), likely due to variability of land surface features and/or lower signal-to-noise ratios of the radiance 40 measurements. The annual global mean XCO OCO−2 2 for land is near zero (0.06 ppm) and exhibits little variation with season. For ocean H-gain, the global mean XCO OCO−2 2 is larger (−0.40 ppm) and varies more significantly by season from −0.2 (DJF) to −0.6 ppm (JJA). The disagreements in 45 the ocean H-gain data tend to be spatially coherent, with a notably large negative difference in the NH in most seasons. Currently, the underlying cause of these disagreements is unknown and could stem from instrument calibration or sampling-related issues, differences in retrieval algorithm 50 versions, or even collocation issues.
The disagreement in XCO 2 for ocean H-gain between ACOS GOSAT v9 and OCO-2 v10 is highlighted in panel (a) of  although there is a persistent low difference in the NH for the remainder of the record. Figure 17 shows the XCO OCO−2 2 data for the combined land H-gain and M-gain data, similar to Fig. 16. The main feature here is that the overall variability is larger compared 5 to the ocean H-gain data, which we attribute to biases introduced by variations in both topography and surface albedo. A slightly positive (red) signal is observed during the September to December months in the SH, especially in 2014, 2018, and 2019. Although additional investigation into such signals 10 is warranted, it is beyond the scope of the current work.
A set of summary statistics for the ACOS GOSAT v9 versus OCO-2 v10 XCO 2 product is given in Table 10. The values reported here are on the individual collocations by year and season, rather than the spatially gridded averages as 15 given in Figs. 14 and 15. For the land observations, there has been a very slight upward trend in time of the XCO OCO−2 2 to slightly more positive values (GOSAT v9 larger than OCO-2 v10 XCO 2 ). On the other hand, for ocean H-gain observations, the general trend has been an increasingly more 20 negative XCO OCO−2 2 in time, as is seen in Fig. 16. Additional investigation will be required to determine the root cause(s) of these differences.

Data availability
The ACOS GOSAT v9 XCO 2 data are available on the NASA Goddard Earth Science Data and Information Services Center (GES-DISC) in both the perorbit full format (OCO-2 Science Team et al., 2019b, 5

Summary
The v9 ACOS GOSAT XCO 2 product, spanning February 2009 through June 2020, has been compared to XCO 2 estimates from TCCON, a suite of atmospheric inversion systems (models), and with collocated OCO-2 v10 data. The 30 ACOS GOSAT v9 product is an improvement over ACOS GOSAT v7.3 relative to these standards. The v9 product provides a significant extension of the data record and contains data in M-gain viewing mode over bright land surfaces. Of the 37.4 × 10 6 estimates of XCO 2 contained in the 35 ACOS GOSAT v9 data record, approximately 80 % were prefiltered due to contamination by cloud and/or aerosol, or due to insufficient SNR. Of the 7.0×10 6 that were selected to run through the ACOS L2FP algorithm, approximately 6.1 × 10 6 returned valid estimates of XCO 2 . However, only 2 × 10 6 40 of those were identified as being of "good" quality. This represents 5.4 % of the total recorded soundings. The quality filtering and bias correction variables used for ACOS GOSAT v9 were similar to those used in previous product versions, and similar to those used for OCO-2 v9 and v10, but include 45 Table 10. A set of summary statistics for the comparison of the ACOS GOSAT v9 XCO 2 to the OCO-2 v10 product. Individual collocations for each year and season are given by N , while the mean XCO 2 and the standard deviation from the mean are given by µ and σ , respectively, both in units ppm. The top portion of the  Specifically, the standard deviation of the mean station bias for the 26 sites is 0.41 ppm for the ACOS GOSAT v9 record, compared to 0.51 ppm at 23 stations for ACOS GOSAT v7.3.
Comparisons with collocated XCO 2 derived from a suite of four atmospheric inversion systems (models) suggest 15 annual global mean differences of 0.15 ppm and standard deviation of 0.5 ppm. Hemispherical differences in XCO 2 estimates over oceans were observed, as well as robust subcontinental-scale land features. Results indicate better agreement with models in the ACOS v9 20 product (µ = −0.20 ppm, σ = 0.8 ppm) compared to v7.3 (µ = −0.54 ppm, σ = 1.0 ppm) for the overlapping period April 2009 through June 2016, but further investigation is required to explain the remaining disagreement over large spatial and temporal scales. 25 Comparisons with collocated OCO-2 v10 XCO 2 data show low bias but relatively high scatter for land observations (µ = 0.06 ppm, σ = 1.0 ppm, when averaged across seasons). Increased scatter over land is expected due to XCO 2 bias introduced by variability in topography and sur-30 face albedo. However, for ocean H-gain observations, although the XCO 2 scatter is lower than that for land as expected (σ = 0.7 ppm), the global mean bias is relatively high (µ = −0.4 ppm, when averaged across seasons). These are issues that must be resolved in order for GOSAT v9 and 35 OCO-2 v10 data to provide consistent information to atmospheric inversion systems for assessing fluxes of CO 2 .
Global estimates of CO 2 derived from satellite measurements provide coverage in traditionally data-sparse regions where ground-based measurements are difficult. The assim-40 ilation of satellite XCO 2 into atmospheric inversion systems to quantify the spatiotemporal variations of carbon fluxes is a promising, but challenging, area of research. This research continues to benefit from various improvements in transport models, atmospheric inversion systems, and satellite re-45 trievals. The role of the GOSAT record in this field remains unique due to its exceptional 11-year length and its coverage of nearly 5.5 years of the carbon cycle prior to the launch of OCO-2. The ACOS GOSAT v9 L2Std and L2Lite file products are both available on the NASA GES DISC (OCO-2 Sci-50 ence Team et al., 2019b, a). , with the mean difference (horizontal solid black line) and ±1 standard deviation (gray shading). In these plots, the three GOSAT observation modes have been combined in order to provide the maximum number of collocations possible for the seasonal fits.                   Competing interests. The contact author has declared that neither they nor their co-authors have any competing interests.