Satellite and Ground-based Measurements of X CO 2 in a Remote Semi-Arid Region of Australia

In this study, we present ground-based measurements of column-averaged dry-air mole fractions (DMFs) of CO2 (or XCO2 ) from an EM27/SUN portable spectrometer, equipped with an automated clam shell cover, taken in a semi-arid region of Australia. We compared these measurements to space-based XCO2 retrievals from the Greenhouse Gases Observing Satellite (GOSAT). Side-by-side measurements of EM27/SUN with the Total Carbon Column Observing Network (TCCON) instrument at the University of Wollongong were conducted in 2015-2016 to derive an XCO2 calibration factor of 0.9954 5 relative to TCCON. Although we found a slight drift of .0125% per month in the calibration curve of the EM27/SUN vs TCCON XCO2 , the alignment of the EM27/SUN proved stable enough for a campaign, keeping the retrieved Xair values, a measure of stability, to within 0.5% and the modulation efficiency to within 2%. From the measurements in Alice Springs, we confirm a small bias of around 2 ppm in the GOSAT M-gain to H-gain XCO2 retrievals, as reported by the NIES GOSAT validation team. Based on the reported random errors from GOSAT, we estimate the required duration of a future campaign in 10 order to improve the estimated bias between the EM27/SUN and GOSAT. The dataset from the Alice Springs measurements is accessible at http://dx.doi.org/10.4225/48/5b21f16ce69bc (Velazco et al., 2018).


Introduction
The Greenhouse Gases Observing Satellite (GOSAT), launched on January 23, 2009, is the first satellite in orbit dedicated to the measurements of the two major anthropogenic greenhouse gases that cause global warming: CO 2 and CH 4 (Yokota et al., 2009;Kuze et al., 2009).Since its launch, GOSAT data have been used to study and determine the global distributions of CO 2 and CH 4 (Reuter et al., 2014;Houweling et al., 2015;Parker et al., 2015;Turner et al., 2015).The spectrometer on board GOSAT, called Thermal and Near infrared Sensor for Carbon Observation-Fourier Transform Spectrometer (TANSO-FTS), is able to simultaneously detect short wave infrared (SWIR: bands 1 at 0.76 µm, 2 at 1.6 µm and 3 at 2.0 µm) and thermal infrared (TIR: band 4 from 5.5 to 14.3 µm).The TANSO-FTS has a footprint of about 10.5 km in diameter on the ground viewed at nadir, which is observed for 4 seconds.Bands 1, 2, and 3 are measured in two linear polarizations simultaneously with three selectable gains; high (H), medium (M) and low (L) (Suto et al., 2013).GOSAT employs the different gain settings 1 Earth Syst.Sci.Data Discuss., https://doi.org/10.5194/essd-2018-161 to compensate for the different signal levels due to the reflective properties (albedo) of the Earth's surface, which also depend on wavelength.For the majority of the soundings over land, H-gain is used.GOSAT M-gain retrievals over land are used over surfaces that are bright in the SWIR such as deserts and semi-arid regions.However, a bias between GOSAT M-and H-gain retrievals of X CO2 has been reported, along with a lack of M-gain validation with TCCON (Yoshida et al., 2013).
Fig. 1 shows the locations of GOSAT M-gain soundings over a period of one year obtained from the NIES (National Institute for Environmental Studies, Japan) version 2.72 (white squares) and the previous version 2.60 (black dots) for reference.
Satellite-based retrievals of X CO2 are validated by the Total Carbon Column Observing Network (TCCON), a ground-based network dedicated to the precise measurements of greenhouse gases (Wunch et al., 2011).TCCON data are used to validate satellite instruments such as GOSAT (Morino et al., 2011), GOSAT-2 (Matsunaga et al., 2018), OCO-2 (Crisp et al., 2017) and Sentinel 5P (Borsdorff et al., 2018).However, apart from locations in the western U.S.A. (Dryden and Los Alamos TCCON sites), there are no operational TCCON stations (magenta stars in Fig. 1) that are ideally located in a clean desert or semi-arid region not influenced by nearby anthropogenic emissions and have plenty of nearby GOSAT M-gain soundings.The Dryden TCCON station at the Armstrong Flight Research Center (AFRC), Edwards, CA is located in the Mojave desert at 34.960 • N, 117.881 • W (700 m a.s.l).Although not as densely populated, AFRC is only approximately 100 km north Los Angeles (L.A. county pop.c. 10.16 million) and 100 km east of Bakersfield,CA (pop. c. 376,380).
Although significantly less populated, the world's deserts and semi-arid regions encompass large areas that are mostly undisturbed by anthropogenic emissions and are important for understanding the carbon cycle.Recently, the importance of semi-arid regions in the carbon cycle inter-annual variability has been highlighted (Poulter et al., 2014;Ahlström et al., 2015;Trudinger et al., 2016).Therefore, measurements over M-gain regions are needed by the satellite community (Yoshida et al., 2013) because understanding and minimizing biases in satellite retrievals over semi-arid regions are necessary for carbon cycle studies.This highlights the significance of this study.Central Australia has a large semi-arid region that is relatively easy to access.There are plenty of uninhabited, vast, homogeneous areas that can accommodate large footprints of satellite-based sensors such as GOSAT, GOSAT-2, OCO-2, and Sentinel 5P.Central Australia is also an ideal place to measure undisturbed atmospheric conditions that can serve as calibration point for satellite retrievals of atmospheric composition.At the same time, the desert environment provides high surface reflectivity, which challenges satellite retrievals and thus leads to their improvement.Despite the importance of desert locations like Central Australia in remote sensing, infrastructure support is not available and accessibility could be a challenge.Benchmark measurements and pilot studies for desert sites are needed.We address this need by utilizing a well-established portable spectrometer, an EM27/SUN by Bruker Optics GmbH (Gisi et al., 2012;Frey et al., 2015;Hase et al., 2015;Hedelius et al., 2016;Frey et al., 2018), which we retrofitted for operations in a harsh environment.The instrument can measure spectra covering the spectral bands in the near infrared necessary to derive column-averaged dry-air mole fractions of CO 2 , CH 4 and CO with sufficient stability for short-term campaigns (Hedelius et al., 2016).The instrument was transported to Alice Springs in Central Australia with the primary objective of validating the GOSAT X CO2 signal and making benchmark measurements in the region.The Alice Springs measurements are unique considering that they have been collected from a clean desert environment and can be directly compared to GOSAT retrievals using the M-gain measurements.
In this paper, we focus on X CO2 measurements from the EM27 and GOSAT.We compare NIES GOSAT retrieval versions 2.60 and the new version 2.72, which are not bias corrected, with the retrievals from the EM27/SUN.This manuscript is organized as follows.A description of Alice Springs and desert Australia is given in Section 2. In Section 3, we briefly discuss We provide our conclusions in Section 5.

Alice Springs Australia Site Description
We conducted a measurement campaign with the retrofitted EM27/SUN system at the Australian Bureau of Meteorology (BOM) facility in Alice Springs (23.7 • S, 133.87 • E) near the end of September to the beginning of October 2016, with the primary objective of validating the GOSAT X CO2 signal in this semi-arid region.Alice Springs is located in Central Australia, also called the "Red Centre".Next to Darwin and Palmerston, it is the third-largest town in the Northern Territory of Australia with a population of 23,726 in the urban area (2016 census).The vegetation around Alice Springs is composed mostly of dry scrubby grassland.The Alice Springs terrain, consisting mostly of sandy plains with some areas of rocky highland, is bounded by several deserts; the Tanami desert to the north, Simpson desert to the east and southeast, the Great Victoria desert to the south and the Gibson desert to the west.In September, statistical data from the Australian Bureau of Meteorology (BOM) show an average monthly rainfall of 8.7 mm, average high and low temperatures of 27.3 • C and 10.3 • C respectively and mean monthly sunshine of 300 hours.Collected GOSAT M-gain soundings over land for one whole year (black dots for version 2.6 and white squares for version 2.72) shown in Fig. 1 include a large part of this "Red Centre" including Alice Springs.Apart from remote regions in North Africa, the Middle East and near-densely-populated areas in California USA, Central Australia is the only reasonably accessible region that provides an abundant amount of M-gain soundings.3 Instruments and Methods

EM27/SUN with Automated Clamshell Cover
The EM27/SUN system, which we will refer to as "EM27" from hereon and its characterization are thoroughly described in the works of Gisi et al. (2012); Frey et al. (2015); Hase et al. (2015) and Hedelius et al. (2016).For measurements of CO 2 , CH 4 H 2 O and O 2 spectra in the near infrared, the instrument is fitted with an Indium Gallium Arsenide (InGaAs) detector 5 dedicated to 5,000-12,000 cm −1 .In addition, a second extended InGaAs detector has subsequently been added for frequencies below 5000 cm −1 allows for measurements of CO and N 2 O.This also enables measurements of spectra covering the O 2 bands necessary to derive column-averaged dry-air mole fractions of CO 2 and CH 4 similar to the method used by TCCON.
However, in contrast to TCCON, we used the typical maximum optical path difference (MOPD) of 1.8 cm, corresponding to a spectral resolution of 0.5 cm −1 .We equipped the Bruker EM27 with a weather station-controlled automated clamshell cover and fairing (Fig. 2).Our objective was to achieve autonomous and remote measurements of greenhouse gases in harsh environments, in particular, to perform much needed measurements of greenhouse gas (CO 2 and CH 4 ) total columns, in desert Australia.The design and more details of this construction can be obtained by contacting the authors.Recently, Hedelius et al.
(2016) published a long-term assessment of errors and biases in retrievals of several TCCON gases from the EM27.In spite of a reported drift, they found that the stability of the EM27 is sufficient for short-term campaigns.Therefore, before deployment to Alice Springs, we have operated the EM27 from December 2015 to September 2016 and tested the automated clamshell cover at the University of Wollongong, where the TCCON instrument (Griffith et al., 2014) is also located.Results of the collocated measurements are in Section 4.

Retrieval of X CO2 and X air
To retrieve X CO2 , we use the same retrieval software used by TCCON called GGG2014 (Wunch et al., 2017), which was also used by Hedelius et al. (2016) for their EM27.Column-averaged dry-air mole fractions (DMFs) of gases (X gas ) are retrieved from the EM27 measurement as in Wunch et al. (2010): where V C gas is the vertical column of the gas and V C dryair is the dry pressure column of air.The DMF of air ("X air ") is also retrieved using the oxygen column but uses the measured surface pressure to obtain the vertical column of air: where mH 2 O and m dry air are the mean molecular masses of water (18.02 g mol −1 ) and dry air (28.964 g mol −1 ), X H2O is the retrieved DMF of water vapour and V C air is calculated from the surface pressure, P s : where g is the column-averaged acceleration due to gravity and N a is Avogadro's constant.X air can be a good indicator of instrument stability and changes in spectrometer alignment because V C air is calculated using the surface pressure, which can be measured to better than 0.3 hPa (Wunch et al., 2011) and accurate over long periods with current instrumentation.

GOSAT Specific Point Observations
Specific point observations over Alice Springs were requested from July 2016, in preparation for the campaign in September.
Five locations within 100 km of the center of Alice Springs were targeted (see Fig. 3).We only used the specific point observation data to compare with the EM27.However, to construct a time-series shown in Fig. 7, we used all available GOSAT data version V2.72 from NIES spanning the years 2010 to 2017.Daily averages within 1000 km of Alice Springs are calculated for 4 Results

Comparison with Wollongong TCCON Station
In this subsection, we present a side-by-side comparison of retrievals of X CO2 between the EM27 and the Wollongong TCCON station.Apart from the work by Hedelius et al. (2016), studies on long-term comparisons of the EM27 with TCCON are rare.
Here, we focus on comparisons of X air and X CO2 from measurements spanning almost one year under varied environmental 5 conditions.

X air
From Equation 2, we can see that the theoretical value of X air would be 1.0 for a perfect instrument with a perfect retrieval.Figure 4 shows the normalized probability distribution functions (PDF) of X air values retrieved by both instruments in 2016.One manifestation of instrumental differences between the Wollongong TCCON and the EM27 is the difference in the X air retrieved from both instruments.The corresponding mean values of X air from both instruments are 0.9824 and 0.9849 for TCCON and EM27 respectively.The TCCON X air PDF exhibits a more Gaussian pattern, while the EM27 X air PDF is slightly skewed.Figure 5 shows the mean X air values calculated for each month in 2016 for TCCON (red) and EM27 (blue).
The error bars represent the spread (one standard deviation) in the monthly mean values of X air .This should give us a measure of how much the X air changes and therefore another measure of instrument stability.In Fig. 5, we see that the mean X air values from the EM27 are slightly larger compared to the TCCON instrument but a slight seasonal dependence can be seen in both instruments.The difference in the mean X air values does not appear to be purely related to sampling differently across that seasonal dependence.

X CO2 Comparison with Wollongong TCCON and Apparent Drift
From the EM27 and TCCON measurements in 2016, we derived a calibration factor for the EM27.We calibrated the EM27 measurements to Wollongong TCCON data because the Wollongong X CO2 has been calibrated against aircraft profile measurements that are traceable to the WMO in-situ scale (Wunch et al., 2010).From this exercise, we arrive at a calibration factor of EM27=0.9954*TCCON(Fig. 6).Frey et al. (2015) reported a comparable calibration factor of 0.9951*TCCON for X CO2 using collocated measurements of five EM27 instruments with the TCCON instrument at the Karlsruhe Institute of Technology in Germany.However, Hedelius et al. (2016) reported a smaller bias of +0.03% between EM27 and TCCON.We think that the slightly larger bias in this work (-0.46%) is probably due to an imperfect alignment of our EM27, with a modulation efficiency (ME) of 96% at maximum optical path difference (OPD).The ME calculation was done using water lines measured at ambient laboratory/room air (Frey et al., 2015).The ILS was monitored before, during and after the campaign and the ME remained stable within 94%-96%.Therefore, we are confident that the calibration factor relative to TCCON was consistent during the campaign with that derived relative to Wollongong TCCON.Hedelius et al. (2016) reported a noticeable drift in the EM27 measurements over several months.We also observed a very small drift in the EM27 vs TCCON X CO2 values shown in Fig. 6, where the calibration line has been derived.For clarity, we color-coded the EM27 vs TCCON X CO2 values in Fig. 6 according to days after Jan 1, 2016.The colors progress with time, i.e. from dark blue (oldest) to dark brown (most recent).We found that the drift within three months before the campaign is very small, i.e. 0.0125% per month as calculated from the EM27/TCCON calibration factors from July to mid September.

9
Earth Syst.Sci.Data Discuss., https://doi.org/10.5194/essd-2018-161 Open Access Further work and more measurements may be necessary to better understand the cause of this phenomenon but this is beyond the scope of this study.

Measurements in Alice Springs
Results of the the ground based X CO2 measurements during the Alice Springs campaign are shown in Fig. 7.The measurements started on September 29, 2016 with interruptions due to cloudy and/or rainy weather.In total, we gathered 6 days of good 5 ground-based measurements of X CO2 under good weather conditions, which coincided with 9 GOSAT specific point soundings within 100 km of the site.A slight airmass dependence in the X CO2 retrievals results to lower X CO2 values at low sun elevation, as can be observed in Fig. 7.This airmass dependence is well-known and also discussed in Wunch et al. (2011).But for the purpose of GOSAT comparisons, this airmass dependence is negligible because GOSAT passes the site at around 13:05 local time, corresponding to a high sun elevation (see Table 1 for the dates).

Comparisons with GOSAT Measurements in Alice Springs
To demonstrate the variability of X CO2 in the region, we plot the time series of interpolated GOSAT soundings near Alice Springs in Fig. 8.The daily mean soundings were selected and calculated within a 1000-km radius of the BOM facility where EM27 measurements were made.Vertical gray lines represent the corresponding standard deviation for each daily mean.The H-gain retrievals (red triangles) and M-gain retrievals (blue squares) were separated in this plot.Weighted linear least squares fitted straight lines were calculated for the M-gain X CO2 (cyan) and H-gain X CO2 data (magenta).We confirm that the M-gain retrievals are biased high (around 2 ppm) compared to the H-gain retrievals.Compared to the TCCON sites in Wollongong and Darwin (Deutscher et al., 2014), the X CO2 signal in Alice Springs is relatively smooth and undisturbed, which is expected because Alice Springs is in the middle of the Australian continent, a desert environment with no large sources to interfere with the X CO2 signal.The terrestrial biosphere is the largest driver of variability in the Southern Hemisphere column X CO2 (Deutscher et al., 2014).Moreover, Deutscher et al. (2014) have shown that the magnitude of the seasonal variability in the column-average dry-air mole fraction of X CO2 is comparable in magnitude to the annual increase.In Alice Springs, only a slight seasonal cycle in the measured X CO2 can be seen.This slight cycle is probably less driven by the surrounding vegetation because the centre of the Australian arid zone is not affected by seasonality and has low aggregate vegetation (Lawley et al., 2011) but more by meridional effect.

11
Earth Syst.Sci.Data Discuss., https://doi.org/10.5194/essd-2018-161 Open Access GOSAT specific point observations that were scheduled during the campaign are shown on Table 1.M and H-gain retrievals (6 days apart) were obtained during the short campaign.We averaged and compared the GOSAT retrievals directly with the coincident measurements from the EM27 (± 0.5 hour) and derived a calibration factor of 0.9927*EM27 for GOSAT H-gain and 0.9983*EM27 for the GOSAT M-gain, as shown in Fig. 9.This is only slightly different from the previous version 2.60 of the data (H-gain: 0.9935*EM27 and M-gain: 0.9997*EM27).Indeed, there is a slight bias in the M and H gain retrievals from GOSAT, i.e.M gain retrievals are biased slightly low and H-gain retrievals are very close compared to the EM27.However, the M and H-gain bias relative to each other has been improved in the version 2.72 data release.

Recommendations for a Future Alice Springs Campaign: Simulating Error Reduction in Bias Estimates
In this subsection, we attempt to estimate how long a measurement campaign in Alice Springs should last in order to improve the statistical errors in the estimated bias between GOSAT and the EM27.In order to estimate the required measurements from the EM27, we make the following assumptions and conditions: 13 Earth Syst.Sci.Data Discuss., https://doi.org/10.5194/essd-2018-161 Open Fig. 10 shows a plot of how the statistical errors in the estimated bias improve with the number of weeks in the campaign.A two-week campaign would already reduce the statistical error between GOSAT M-gain and EM27 measurements to less than the reported GOSAT M-Gain standard deviation (0.25 ppm).To achieve a statistical error of 0.1 ppm between GOSAT M-gain and EM27 measurements, it would take about 6.5 weeks of measurements or more.On the other hand, 8.5 weeks for H-gain is needed because the H-gain retrievals have a slightly higher standard deviation.The ideal time window to perform this would be from March to November (Autumn-Winter-Spring).It is best to avoid the summer months (Dec-Feb) because of the desert heat in Alice Springs, which could be physically demanding for the operator/s and may affect the operation of an EM27 without a temperature-regulated cover.A temperature-regulated cover would improve stability and make the EM27 suitable for summer operations even in the high heat (>40 • C) of Alice Springs.Note however that our error estimates are conservative and this implies the assumption that there are no drifts in the M-and H-gain measurements relative to one another that would require long-term validation. 14

Conclusions
Alice Springs would be a welcome addition to ground sites dedicated to environmental satellite sensor calibration and validation.With the exception of Lauder, New Zealand, there are no permanently dedicated ground targets for satellite calibration and validation in the Southern Hemisphere.Any addition from the southern region would be highly valued, especially the type of addition that Alice Springs could offer.Currently, vicarious radiometric calibration of passive optical instruments from satel-5 lites is done on the large homogeneous Railroad Valley playa (RRV) in Nevada, USA.However plans for future lithium mining threaten to disrupt the playa floor, which could render this critical asset useless for large footprint sensors on satellites like GOSAT, GOSAT-2, OCO-2, OCO-3 and on other future satellites.Alice Springs would offer unique opportunities particularly in the context of satellite surface products and associated emissivity assessments because the environment of Alice Springs, provides ideal conditions to measure undisturbed atmospheric conditions that will serve as calibration point for satellite re-10 trievals of atmospheric composition.At the same time, the desert environment, like RRV, provides high surface reflectivity, which is a challenge to satellite retrievals and thus this data is especially needed for validation purposes.In a future study, we will investigate GOSAT methane retrievals using the same approach employed here. 15 the instruments and methods, which are already well established.Results of measurement comparisons with the TCCON station in Wollongong and comparisons with the GOSAT M-gain and H-gain soundings in Alice Springs are shown and discussed under Section 4. Statistical calculations and recommendations for a future Alice Springs campaign are also under Section 4.

Figure 1 .
Figure 1.GOSAT M-gain retrievals of CO2, CH4 for 2016.Black dots are from version 2.60 and the white squares are from version 2.72.Magenta stars indicate the location of operational TCCON sites as of October 2018.The center of the magenta cross marks the location of Alice Springs in Australia and the magenta circle has a radius of approximately 1000 km centered on Alice Springs.The Darwin and Wollongong (W'gong) sites in Australia are labeled.
review for journal Earth Syst.Sci.Data Discussion started: 9 January 2019 c Author(s) 2019.CC BY 4.0 License.

Figure 2 .
Figure 2. Drawing of the Wollongong EM27 automated clamshell cover and fairing (by Steve Selby, UOW).

5MFigure 3 .
Figure 3.A Google Earth map showing the GOSAT specific point observations around Alice Springs (cyan circles labeled "Obs#") .The ground-based site is located at the airport, c. 14 km south of the town center.The magenta circle shows a 60-km radius from the site, for visual reference.
However, inaccuracies in the O 2 spectroscopy lead to an X air value of approximately 0.98 for all TCCON sites.This value varies by approximately 1% with solar zenith angle, indicating an airmass dependence in the O 2 retrievals(Pollard et al.,   10   2017).Deviations from the characteristic values for X air are generally indicative of erroneous behavior in the measurement and retrieval system such as interferometer misalignment, tracking errors or fitting to an incorrect airmass due to timing errors as recently reported inPollard et al. (2017).7EarthSyst.Sci.Data Discuss., https://doi.org/10.5194/essd-2018forjournal Earth Syst.Sci.Data Discussion started: 9 January 2019 c Author(s) 2019.CC BY 4.0 License. 10

Figure 4 .
Figure 4. Normalized probability distribution functions (PDF) of Xair from TCCON (right panel) and Xair from the EM27 (left panel) in Wollongong for 2016.The PDFs for each month are separated by colors and the mean Xair values for each month are indicated by the month number in a year.The mean Xair for the EM27 is 0.9849 and for TCCON is 0.9824 for the whole year.The clamshell cover and fairing were installed on the EM27 in February, this may have affected the alignment resulting in a shift in the Xair.

Figure 5 .
Figure 5.A measure of the spread of the of Xair values from TCCON (red) and EM27 (blue) by taking the mean (central markers) and corresponding standard deviations (bars) for each of the monthly probability distribution functions shown in Fig. 4.
review for journal Earth Syst.Sci.Data Discussion started: 9 January 2019 c Author(s) 2019.CC BY 4.0 License.

Figure 7 .
Figure 7. Retrievals CO2 from the EM27 in Alice Springs, compressed along the x-axis (days).The measurement days on the x-axis is not continuous due to interruption from bad weather.See Section 6 for data availability.
review for journal Earth Syst.Sci.Data Discussion started: 9 January 2019 c Author(s) 2019.CC BY 4.0 License.However, H-gain retrievals seem to have diminished around 2011, coinciding with the dry months starting from April 2011.H-gain soundings from April-August 2011 were 23% less compared to the April-August mean derived from the years 2010-2017.This could be due to stunted vegetation growth or mortality in the absence of rain in the region, which led to brighter surfaces.Bright surfaces mean more M-gain measurements and this may explain the 74% increase in M-gain soundings in April-August 2011 compared to the April-August 2010-2017 average.Rainfall records indicate that April-August 5 2011 had the least amount of rainfall on record in Alice Springs after 2002, which was only broken by 2015, then 2017 (http://www.bom.gov.au/climate/data/,station number 015590).

Figure 8 .
Figure 8.Time series XCO 2 from GOSAT M-gain and H-gain retrievals in Alice Springs from the NIES version 2.72 product.The timing of the Alice Springs campaign from Sep. 27 to Oct. 6 is indicated by the yellow arrow.Gray lines represent one standard deviation associated with each daily mean.The inset shows a map projection with the approximate locations of the soundings used for this plot.Only data within a 1000-km radius from Alice Springs (yellow X on the map) were considered.

Figure 9 .
Figure 9. Mean XCO 2 from the EM27 vs GOSAT XCO 2 version 2.60 (left panel) and version 2.72 (right panel) in Alice Springs.The calculated means are weighted by the1-σ uncertainty of the individual measurements.The scaling factor derived from the TCCON comparisons, EM27=0.9954*TCCON,has been taken into account.
review for journal Earth Syst.Sci.Data Discussion started: 9 January 2019 c Author(s) 2019.CC BY 4.0 License. 1.At least one specific point overpass per week (M-gain or H-gain), similar to this campaign.2. 1-σ standard deviation of all GOSAT specific point observations is: 0.25 ppm (M-gain) and 0.29 ppm (H-gain).3. 1-σ standard deviation of all EM27 measurements is 0.05 ppm at an averaging time of 1-hour.Chen et al. (2016) reported a 1-σ standard deviation of 0.04 ppm at 10 minutes averaging.4. The error in the estimated bias is calculated as σ T = σ 2 GOSAT + σ 2 EM 27 , which then improves by the number of weeks (N), i.e. σ 2 T /N 5. Any long-term drifts in the satellite measurements are negligible over the whole duration of the campaign.

Figure 10 .
Figure 10.The statistical errors in the estimated bias of XCO 2 improve with the number of weeks in the campaign.