Shipborne measurements of XCO2, XCH4, and XCO above the Pacific Ocean and comparison to CAMS atmospheric analyses and S5P/TROPOMI

Measurements of atmospheric column-averaged dry-air mole fractions of carbon dioxide (XCO2), methane (XCH4), and carbon monoxide (XCO) have been collected across the Pacific Ocean during the Measuring Ocean REferences 2 (MORE-2) campaign in June 2019. We deployed a shipborne variant of the EM27/SUN Fourier transform spectrometer (FTS) on board the German R/V Sonne which, during MORE-2, crossed the Pacific Ocean from Vancouver, Canada, to Singapore. Equipped with a specially manufactured fast solar tracker, the FTS operated in direct-sun viewing geometry during the ship cruise reliably delivering solar absorption spectra in the shortwave infrared spectral range (4000 to 11000 cm-1). After filtering and bias correcting the dataset, we report on XCO2 , XCH4 , and XCO measurements for 22 d along a trajectory that largely aligns with 30 ∘ N of latitude between 140 ∘ W and 120 ∘ E of longitude. The dataset has been scaled to the Total Carbon Column Observing Network (TCCON) station in Karlsruhe, Germany, before and after the MORE-2 campaign through side-by-side measurements. The 1σ repeatability of hourly means of XCO2 , XCH4 , and XCO is found to be 0.24 ppm , 1.1 ppb , and 0.75 ppb , respectively. The Copernicus Atmosphere Monitoring Service (CAMS) models gridded concentration fields of the atmospheric composition using assimilated satellite observations, which show excellent agreement of 0.52±0.31 ppm for XCO2 , 0.9±4.1 ppb for XCH4 , and 3.2±3.4 ppb for XCO (mean difference ± SD, standard deviation, of differences for entire record) with our observations. Likewise, we find excellent agreement to within 2.2±6.6 ppb with the XCO observations of the TROPOspheric MOnitoring Instrument (TROPOMI) on the Sentinel-5 Precursor satellite (S5P). The shipborne measurements are accessible at 10.1594/PANGAEA.917240.

The greenhouse gases carbon dioxide ( ${\mathrm{CO}}_{\mathrm{2}}$ ) and methane ( ${\mathrm{CH}}_{\mathrm{4}}$ ) and the air pollutant carbon monoxide ( $\mathrm{CO}$ ) are the target constituents of a range of currently orbiting and planned Earth observing satellite missions [36], [17]. The latest addition to the fleet of spaceborne sensors is the Sentinel-5 Precursor (S5P) satellite with its TROPOspheric Monitoring Instrument (TROPOMI) in orbit since October 2017 [45]. TROPOMI retrieves, among other constituents, the column-averaged dry-air mole fractions of ${\mathrm{CH}}_{\mathrm{4}}$ ( ${\mathrm{XCH}}_{\mathrm{4}}$ ) and $\mathrm{CO}$ ( $\mathrm{XCO}$ ) from spectra of backscattered sunlight in the shortwave-infrared (SWIR) spectral range [[5], [6], [29], [14]]. In parallel to the expansion of the fleet of greenhouse gas sensors in orbit, the European Centre for Medium Range Weather Forecasts (ECMWF) operates the Copernicus Atmosphere Monitoring Service (CAMS) on behalf of the European Commission. CAMS assimilates satellite measurements of atmospheric composition to forecast the global ${\mathrm{CO}}_{\mathrm{2}}$ , ${\mathrm{CH}}_{\mathrm{4}}$ , and $\mathrm{CO}$ concentrations at high spatial and temporal resolution using the Integrated Forecasting System (IFS). During our campaign in June 2019, CAMS assimilated ${\mathrm{XCO}}_{\mathrm{2}}$ and ${\mathrm{XCH}}_{\mathrm{4}}$ measurements from the Greenhouse gases Observing SATellite (GOSAT) [36], ${\mathrm{CH}}_{\mathrm{4}}$ and $\mathrm{CO}$ measurements from the Infrared Atmospheric Sounding Interferometer (IASI) [13], and $\mathrm{CO}$  measurements from the Measurement of Pollution in the Troposphere (MOPITT) [16] instrument. The ultimate goal is to monitor the mitigation of anthropogenic greenhouse gas emissions and air pollution from global to regional scales [[39], [40], [30], [31], [2], [32]]. Validation of the ${\mathrm{XCO}}_{\mathrm{2}}$ , ${\mathrm{XCH}}_{\mathrm{4}}$ , and $\mathrm{XCO}$ satellite data mostly relies on ground-based direct-sun spectroscopic observations conducted by the Total Carbon Column Observing Network (TCCON) [53] supplemented by the emerging Collaborative Carbon Column Observing Network (COCCON) [20] that measure the column-averaged dry-air mole fractions with similar column sensitivity to the satellites. Likewise, the CAMS model uses TCCON as an evaluation tool [2]. Most of the observatories of the TCCON and COCCON are located at continental sites. Thus, validation of the satellites and models over the oceans is limited to a few island and coastal observatories (in particular Ascension, Reunion, Tenerife, Japan, California). For the satellite retrievals, ocean–land biases [3] can occur since the ocean surface is dark, and thus, satellites typically have to resort to glint geometry [8] or retrievals above clouds [47], [42], which impose difficulties different from the typical clear-sky nadir observations above land.

To enable the evaluation of satellites and models over the oceans, [33] developed a shipborne prototype of the EM27/SUN Fourier transform spectrometer (FTS) [[21], [22]] which is the instrument used within the COCCON [20]. The EM27/SUN has proven to be a reliable instrument for ${\mathrm{XCO}}_{\mathrm{2}}$ and ${\mathrm{XCH}}_{\mathrm{4}}$ measurements in various studies ranging from ad hoc networks covering a larger region of interest [[26], [12], [44], [46], [48], [15]] to mobile deployments [[9], [38]] for the quantification of localized ${\mathrm{CO}}_{\mathrm{2}}$ and ${\mathrm{CH}}_{\mathrm{4}}$ sources. The latest variant of the EM27/SUN incorporates a second spectral detector channel that enables $\mathrm{XCO}$ measurements to be conducted simultaneously with observations of ${\mathrm{XCO}}_{\mathrm{2}}$ and ${\mathrm{XCH}}_{\mathrm{4}}$ [27]. The shipborne observations by [33] were conducted on board the R/V Polarstern during a cruise from Cape Town, South Africa, to Bremerhaven, Germany, in March and April 2014. These measurements were used for evaluating ${\mathrm{XCO}}_{\mathrm{2}}$ and ${\mathrm{XCH}}_{\mathrm{4}}$ observations of the Greenhouse Gases Observing Satellite (GOSAT) and for improving the interhemispheric gradient modeled by the IFS for the CAMS ${\mathrm{CO}}_{\mathrm{2}}$ and ${\mathrm{CH}}_{\mathrm{4}}$ analysis and forecasting system [1].

Here, we report on the further developments of the shipborne EM27/SUN prototype toward routine use as a validation tool over the open oceans. To demonstrate the performance and robustness of the instrumentation and its suitability for satellite and model validation, we deployed the instrument on the German R/V Sonne during the MORE-2 (Measuring Oceanic REferences 2) campaign which led from Vancouver, Canada, to Singapore between 30 May and 5 July 2019. Figure  shows the track of the research vessel over the Pacific Ocean. We report on technical developments (Sect. ), the data processing chain and data quality assessment (Sect. ), and comparisons to TROPOMI's $\mathrm{XCO}$ measurements and CAMS' analysis fields of ${\mathrm{XCO}}_{\mathrm{2}}$ , ${\mathrm{XCH}}_{\mathrm{4}}$ , and $\mathrm{XCO}$ (Sect. ) over the Pacific Ocean. The data collected are publicly available at https://doi.org/10.1594/PANGAEA.917240 for evaluating other datasets, and the shipborne instrument is recommended for routine deployment on ships.

MAP: Figure 1 Track of the R/V Sonne (gray line) during the MORE-2 campaign starting from Vancouver, Canada, on 30 May 2019 and entering port in Singapore on 5 July 2019. The blue dots are the locations of all quality-assured EM27/SUN measurements. The map is provided by [49].

The EM27/SUN is a commercially available FTS which was developed in cooperation by Bruker Optics and the Karlsruhe Institute of Technology (KIT) [22]. The spectrometer has the dimensions $\mathrm{42}\times \mathrm{27}\times \mathrm{35}$   ${\mathrm{cm}}^{\mathrm{3}}$ and weighs about 25  $\mathrm{kg}$ . The EM27/SUN uses a ${\mathrm{CaF}}_{\mathrm{2}}$ beam splitter and a RockSolid™ pendulum interferometer with two cube corner mirrors. The maximum optical path difference of 1.8  $\mathrm{cm}$ supports a spectral resolution of 0.50  ${\mathrm{cm}}^{-\mathrm{1}}$ . After the sunlight passed the interferometer, a parabolic off-axis mirror focuses it on an InGaAs photodetector with a spectral range of 5500–11 000  ${\mathrm{cm}}^{-\mathrm{1}}$ , further called the SWIR-1 channel. Another mirror decouples about 40  $\mathrm{\%}$ of the beam on a second spectrally extended InGaAs photodetector covering the spectral range of 4000–5500  ${\mathrm{cm}}^{-\mathrm{1}}$ [27], called the SWIR-3 channel. Typical exposure times are on the order of 6  $\mathrm{s}$ for a single interferogram. Spectra have been generated from raw DC-coupled interferograms using the preprocessor used by the COCCON network, which has been developed in the framework of the ESA project COCCON-PROCEEDS [[24], [43]]. As suggested by [19], we use water vapor absorption lines to measure the instrumental line shape (ILS).

The EM27/SUN is mechanically robust, but it would not withstand precipitation or sea spray. For the shipborne variant, we assembled a small container that houses the EM27/SUN FTS with its solar tracker, a laptop, and several ancillary sensors (GPS, pressure, temperature) similar to [28]. During the entire MORE-2 campaign, we placed the container outside on the port side of the observation deck of the R/V Sonne, which was the uppermost continuously accessible deck available. We chose this spot to avoid obstruction of the direct light path from the sun to the instrument by ship structures. The container is a white lacquered K470 Zarges aluminum box (IP65 waterproof) with dimensions of $\mathrm{95}\times \mathrm{69}\times \mathrm{48}$   ${\mathrm{cm}}^{\mathrm{3}}$ and mass of 13.4  $\mathrm{kg}$ when empty. Figure shows a photograph of the container deployed on the ship. The solar tracker is a modified version of the custom-built setup used by [33] consisting of a mirror assembly on two perpendicular rotation stages that allow for pointing to any azimuth and elevation position of the sun in the overhead sky. For the ship deployment, we covered the solar tracker with a protective housing that has a fused silica wedged window transmitting the incoming sunlight. The solar tracker housing is positioned on top of the box and attached to the rotation stages moving with the azimuth and elevation rotations. The precision required for the pointing of the solar tracker is 0.05 ${}^{\circ }$ relative to the center of the sun [21] to keep mole fraction uncertainties due to pointing errors below 0.1 %. Our tracking system satisfied this requirement for 79 % of the measurements for which the sun was within the field of view (FOV) of the solar tracker. We observed the largest pointing deviations when high cirrus clouds were present and when the sun was close to the zenith where the azimuth rotation has a singularity. Our filter criteria reliably remove such observations (see Sect. ).

In addition to the main solar tracker, we mounted a f-theta fisheye lens (Fujinon FE185C057HA-1) with a field of view of $\mathrm{185}{}^{\circ }\times \mathrm{185}{}^{\circ }$ under a protective acrylic glass dome onto the lid of the box. A camera (IDS UI-3280CP-M-GL Rev.2) observes the sky through the fisheye lens and provides the position of the sun with an accuracy better than 2 ${}^{\circ }$ when the sun is not within the FOV of the solar tracker. The ambient pressure and temperature sensors, as well as the GPS antenna, are mounted on the lid as well. The box is equipped with a Pfannenberg PF 66000 fan for ventilation on one side and an air outlet on the other side to prevent the box from overheating, both of which are covered by protective lids against sea spray and precipitation. During the whole MORE-2 campaign, the box interior temperature never exceeded 40  ${}^{\circ }\mathrm{C}$ . Inside the box, a Raspberry Pi 3 Model B is the central control unit. It allows for remote access via a network to connect to the laptop controlling the EM27/SUN, an Advantech Ark-2150 embedded PC running the solar tracking software [33], and a central storage unit (Synology DS2018). The Raspberry Pi continuously reads the ancillary sensors for the box interior temperature, the ambient pressure and temperature, and the GPS position of the instrument. The electronics runs on 24  $\mathrm{V}$ DC provided by an AC C-TEC 2410-10 uninterrupted power supply (UPS). In case of a power cut, the Raspberry Pi securely shuts down all devices within the approximately 60  $\mathrm{s}$ backup time of the UPS. The whole container weighs about 80  $\mathrm{kg}$ and consumes 190  $\mathrm{W}$ via a regular 230  $\mathrm{V}$ AC line if the measurement electronics is running at full power. The ventilation consumes an additional 160  $\mathrm{W}$ if switched on, which was necessary throughout the campaign.

Graph: Figure 2 Photograph of the instrument container on board R/V Sonne(a) and the solar tracker housing (b). The solar tracker housing (A in a), the ambient sensors (B in a), and the fisheye camera (C in a) are mounted on top of the box. The solar tracker housing (b) consists of a cylinder which rotates around a horizontal axis in the elevation direction. The cylinder is mounted on a cube which is able to rotate around a vertical axis in the azimuth direction. Sunlight enters the tracker through a fused silica wedged window.

The quality assessment and the retrievals of ${\mathrm{XCO}}_{\mathrm{2}}$ , ${\mathrm{XCH}}_{\mathrm{4}}$ , and $\mathrm{XCO}$ largely follow [33]. Therefore, we summarize the methods here, mostly highlighting the differences and new aspects compared to our precursor study.

The spectral retrieval of the targeted gas concentrations from direct-sun absorption spectra is based on forward modeling of the spectra given a priori concentrations of the molecular absorbers and then iteratively adjusting the concentrations to optimally (in a least squares sense) fit the measured spectra. The spectral retrieval calculates total column number densities of the target gases [ $\mathrm{GAS}$ ] which are a posteriori ratioed by the total column number density of oxygen [ ${\mathrm{O}}_{\mathrm{2}}$ ] to yield the column-averaged dry-air mole fraction ${\mathrm{X}}_{\mathrm{GAS}}$ of the target gas according to

1 $${\mathrm{X}}_{\mathrm{GAS}}={\displaystyle \frac{\left[\mathrm{GAS}\right]}{\left[{\mathrm{O}}_{\mathrm{2}}\right]}}\cdot \mathrm{0.2094},$$

where 0.2094 is the constant column-averaged dry-air mole fraction of molecular oxygen. Referencing the target gas column [ $\mathrm{GAS}$ ] to the oxygen column cancels out instrument- and retrieval-related errors common to both retrievals.

For the spectral retrieval, we use a variant of the RemoTeC algorithm [7] which is in use for satellite observation from GOSAT [[7], [50]], the Orbiting Carbon Observatory (OCO-2) [51], and TROPOMI [29]. We have adapted RemoTeC for transmittance calculations applicable to ground-based direct-sun measurements such as those conducted here. Since the spectral resolution of the EM27/SUN is insufficient to extract profile information from the absorption line shapes, RemoTeC retrieves a scaling parameter on the a priori absorber profiles. The spectral retrieval windows for ${\mathrm{CO}}_{\mathrm{2}}$ , ${\mathrm{CH}}_{\mathrm{4}}$ , and ${\mathrm{O}}_{\mathrm{2}}$ are located in the SWIR-1 channel and almost identical to the ones used by [33]. The retrieval window for $\mathrm{CO}$ is located in the new SWIR-3 channel. Table collects the information on window selection and interfering absorbers. The absorption cross sections of all species are generated from the HITRAN2016 database [23]. Meteorological parameters such as pressure and temperature profiles are taken from the National Centres for Environmental Prediction (NCEP) available at [41]. These NCEP FNL (Final) Operational Model Global Tropospheric Analyses fields are from the Global Data Assimilation System (GDAS) and have a spatial resolution of $\mathrm{1}{}^{\circ }\times \mathrm{1}{}^{\circ }$ and a temporal resolution of 6 h. The a priori profiles for ${\mathrm{CO}}_{\mathrm{2}}$ and ${\mathrm{CH}}_{\mathrm{4}}$ are taken from CAMS greenhouse gas analysis [[39], [40]] and for $\mathrm{CO}$ from the near-real-time operational analysis [[30], [31]]. The profiles are interpolated to the time and location of each individual EM27/SUN measurement. CAMS provides $\mathrm{CO}$ profiles with 60 model levels on a $\mathrm{0.4}{}^{\circ }\times \mathrm{0.4}{}^{\circ }$ grid and 6 h temporal resolution for the campaign period in June 2019. The CAMS ${\mathrm{CO}}_{\mathrm{2}}$ and ${\mathrm{CH}}_{\mathrm{4}}$ profiles have 136 model levels on a $\mathrm{0.25}{}^{\circ }\times \mathrm{0.25}{}^{\circ }$ horizontal grid with 6 h temporal resolution.

Table 1 Spectral windows with target and interfering absorbers. CIA refers to collision-induced absorption.

The measurements collected during the MORE-2 campaign require quality filtering since cloudy or partially cloudy scenes need to be screened and we want to avoid measurements that are contaminated by the exhaust plume of the ship. To this end, [33] suggested a cascade of three criteria: a filter based on the DC part of the recorded interferograms, a filter based on the deviation between spectroscopically derived surface pressure and surface pressure measured in situ, and a filter based on the visual identification of steep slopes in the ${\mathrm{XCO}}_{\mathrm{2}}$ time series.

During the MORE-2 campaign, our FTS recorded the interferograms with the slowly varying DC part included. The DC part is indicative of the overall incoming radiance, and thus, it can be used to track clouds that obstruct the direct-sun view. The DC filter criterion screens measurements either if the DC part $I_{\text{DC}}$ is too small to be direct sunlight or if the fluctuation DC ${}_{\text{fluc}}$ , defined as

2 $${\text{DC}}_{\text{fluc}}\equiv {\displaystyle \frac{max\left(I_{\text{DC}}\right)}{min\left(I_{\text{DC}}\right)}}-\mathrm{1},$$

exceeds 5 %. The DC filter removes 8.39 % of the dataset.

The surface pressure filter compares the surface pressure measured in situ by the ship's meteorological station and the surface pressure calculated from the spectroscopic measurements, as suggested by [53]. We calculate the spectroscopic pressure from the measured total column number densities of [ ${\mathrm{O}}_{\mathrm{2}}$ ] and [ ${\mathrm{H}}_{\mathrm{2}}\mathrm{O}$ ] with

$$\begin{array}{}\text{3}& {\displaystyle }{p}_{\text{dry}}& {\displaystyle }=\left[{\mathrm{O}}_{\mathrm{2}}\right]\cdot {\displaystyle \frac{M_{{\mathrm{O}}_{\mathrm{2}}}}{N_{\mathrm{A}}\cdot {\mathit{\xi }}_{{\mathrm{O}}_{\mathrm{2}}}}}\cdot g,\text{4}& {\displaystyle }{p}_{{\mathrm{H}}_{\mathrm{2}}\mathrm{O}}& {\displaystyle }=\left[{\mathrm{H}}_{\mathrm{2}}\mathrm{O}\right]\cdot {\displaystyle \frac{M_{{\mathrm{H}}_{\mathrm{2}}\mathrm{O}}}{N_{\mathrm{A}}}}\cdot g,\end{array}$$

where $M_{\mathrm{GAS}}$ is the molar mass of the gas molecule, $N_{\mathrm{A}}$ Avogadro's constant, ${\mathit{\xi }}_{{\mathrm{O}}_{\mathrm{2}}}=\mathrm{0.2314}$ (the dry-air mass mixing ratio of oxygen), and $g$ the gravitational constant.

We scale the spectroscopic pressure to the in situ pressure with a factor of 0.9693 such that the ratio

5 $$R_{\text{psf}}\equiv \mathrm{0.9693}\cdot {\displaystyle \frac{p_{\text{dry}}+p_{{\mathrm{H}}_{\mathrm{2}}\mathrm{O}}}{p_{\text{in situ}}}}$$

scatters around unity within the measurement noise under unperturbed conditions. Any measurement for which $R_{\text{psf}}$ deviates by more than 0.3 % from unity is excluded from further processing, which led to a rejection for 6.2 % of the data.

The third quality filter screens the (rare) situation when our EM27/SUN measurements detected the ship's exhaust plume. During the MORE-2 campaign, this happened in the morning of 8 June 2019 when the ship's exhaust plume crossed the light path. The observations show a steep increase of 2  $\mathrm{ppm}$ (parts per million) in ${\mathrm{XCO}}_{\mathrm{2}}$ , while ${\mathrm{XCH}}_{\mathrm{4}}$ and $\mathrm{XCO}$ show no increase. We removed 179 measurements (corresponding to 55 min). After applying all filters, a total of 32 859 (84.38 %) direct-sun measurements pass the filter process and are considered for further processing.

After retrieving and quality filtering the ${\mathrm{XCO}}_{\mathrm{2}}$ , ${\mathrm{XCH}}_{\mathrm{4}}$ , and $\mathrm{XCO}$ concentrations, the records require bias correction for a spurious dependency on the solar zenith angle (SZA) causing an artificial diurnal cycle [53] and for species-dependent scaling factors that adjust our spectroscopic measurements to observations of the TCCON whose stations have been compared and scaled to standards of the World Meteorology Organization [52]. We determine the SZA dependency for each species from observations above the Pacific in background air where the columns are expected to be constant and the scaling factors to TCCON via the ratio of side-by-side measurements.

The spurious dependency on SZA causes an underestimation of the column-averaged dry-air mole fractions at high SZAs for each of the target species. [53] suggested an empirical correction as a function of SZA $\mathrm{\Theta }$ according to

6 $${\mathrm{X}}_{\text{corr, GAS}}\left(\mathrm{\Theta }\right)={\displaystyle \frac{{\mathrm{X}}_{\text{GAS}}\left(\mathrm{\Theta }\right)}{{\mathit{\chi }}_{\text{GAS}}\left(\mathrm{\Theta }\right)}},$$

where ${\mathrm{X}}_{\text{corr, GAS}}\left(\mathrm{\Theta }\right)$ is the corrected column-averaged dry-air mole fraction of the gas under consideration and ${\mathit{\chi }}_{\text{GAS}}\left(\mathrm{\Theta }\right)$ is a third-order polynomial of the form

7 $${\mathit{\chi }}_{\text{GAS}}\left(\mathrm{\Theta }\right)=a{\mathrm{\Theta }}^{\mathrm{3}}+b\mathrm{\Theta }+c,$$

with $a$ , $b$ , and $c$ being free fitting parameters. We perform the fit by splitting our time series into morning and afternoon parts at the lowest SZA of the day and consider only those half days which contain a measurement at $\text{SZA}=(\mathrm{45.0}\pm \mathrm{0.5}){}^{\circ }$ , which was the case for 26 half days. We reference each measurement to the observation closest to $\text{SZA}=\mathrm{45}{}^{\circ }$ , i.e., we choose ${\mathit{\chi }}_{\mathrm{GAS}}\left({\mathrm{45}}^{\circ }\right)=\mathrm{1}$ . Furthermore, we identified half days which showed actual atmospheric variability by fitting the SZA dependency in a first attempt and calculating the standard deviation (SD) of the fit residuum. We removed a half day if less than 97 % of its observations were within $\mathrm{2}\mathit{\sigma }$ of this fit since this indicates actual atmospheric variability which we do not want to misinterpret as spurious SZA dependency. This results in removing 2, 9, and 9 half days from the ${\mathrm{XCO}}_{\mathrm{2}}$ , ${\mathrm{XCH}}_{\mathrm{4}}$ , and $\mathrm{XCO}$ records for the fit of the correction polynomial, respectively. Figure  shows the corresponding data and the fitted correction polynomials for each target species, and Table  lists the parameters defined in Eq. () and the coefficients of determination for each fit. The lowest coefficient of determination is found for $\mathrm{CO}$ most likely due to the stronger atmospheric variability compared to ${\mathrm{CO}}_{\mathrm{2}}$ and ${\mathrm{CH}}_{\mathrm{4}}$ .

Table 2 Parameters $a$ , $b$ , and $c$ (and coefficients of determination $R^{\mathrm{2}}$ ) used for correcting the spurious SZA dependency of the measurements.

Graph: Figure 3 SZA dependency of the retrieved XCO2 (upper panel), XCH4 (middle panel), and XCO (lower panel) and the inferred correction polynomial (solid line). Negative SZAs denote morning and positive SZAs afternoon measurements.

After correcting the SZA dependency, we adjust our measurements to those of the TCCON station at KIT, Karlsruhe, to ensure traceability to WMO standards [25]. To this end, we performed side-by-side measurements at Karlsruhe for a day before (30 April 2019) and after (23 July 2019) the ship campaign. Figure shows the TCCON measurements alongside the EM27/SUN measurements before and after scaling. Following [33], we calculate the scaling factor as the mean ratio ${\mathit{\gamma }}_{\text{GAS}}$ between the TCCON and the EM27/SUN 1 h means for both days according to

8 $${\mathit{\gamma }}_{\text{GAS}}={〈{\displaystyle \frac{\langle {\text{X}}_{\mathrm{GAS}}^{\mathrm{TCCON}}{\rangle }_{\mathrm{h}}}{\langle {\text{X}}_{\mathrm{GAS}}^{\mathrm{EM}\mathrm{27}}{\rangle }_{\mathrm{h}}}}〉}_{\text{day}}$$

for each of the target species. Table  lists the scaling factors ${\mathit{\gamma }}_{\text{GAS}}$ and their error bars, which we calculate as the standard error ${\mathit{\sigma }}_{\mathit{\gamma }}$ of the mean using the hourly mean variances ${\mathit{\sigma }}_{\mathit{\gamma },\mathrm{h}}^{\mathrm{2}}$ ;

$$\begin{array}{}\text{9}& {\displaystyle }{\mathit{\sigma }}_{\mathit{\gamma }}=\sqrt{{\displaystyle \frac{{\sum }_{\mathrm{h}}^N{\mathit{\sigma }}_{\mathit{\gamma },\mathrm{h}}^{\mathrm{2}}}{N^{\mathrm{2}}}}},\text{10}& {\displaystyle }{\left({\displaystyle \frac{{\mathit{\sigma }}_{\mathit{\gamma },\mathrm{h}}}{{\mathit{\gamma }}_{\mathrm{h}}}}\right)}^{\mathrm{2}}={\left({\displaystyle \frac{\mathit{\sigma }\left(\langle {\text{X}}_{\mathrm{GAS}}^{\mathrm{TCCON}}{\rangle }_{\mathrm{h}}\right)}{\langle {\text{X}}_{\mathrm{GAS}}^{\mathrm{TCCON}}{\rangle }_{\mathrm{h}}}}\right)}^{\mathrm{2}}+{\left({\displaystyle \frac{\mathit{\sigma }\left(\langle {\text{X}}_{\mathrm{GAS}}^{\mathrm{EM}\mathrm{27}}{\rangle }_{\mathrm{h}}\right)}{\langle {\text{X}}_{\mathrm{GAS}}^{\mathrm{EM}\mathrm{27}}{\rangle }_{\mathrm{h}}}}\right)}^{\mathrm{2}},\end{array}$$

where $\mathit{\sigma }\left(\right)$ is the SD of the hourly mean and $N$ the total number of hours of side-by-side observations. For ${\mathrm{CO}}_{\mathrm{2}}$ , the scaling factors are consistent within the error bars, and for ${\mathrm{CH}}_{\mathrm{4}}$ , the scaling factors before and after the campaign are consistent at roughly 3 ‰, though the differences are larger than the combined error bars. For $\mathrm{CO}$ , the scaling factors before and after the campaign differ by roughly 2 %, which is substantially more than the error bars. We identify a change in the ILS as the most likely candidate for this difference in the scaling factors. The ILS was measured under laboratory conditions before and after the campaign. We also performed ILS measurements at the beginning of the campaign on 2 June 2019, yet it was impossible to assure laboratory conditions there since the measurements were conducted on deck. The pre-campaign ILS differs from the one taken on board the R/V Sonne most likely due to rough handling during the transport from Germany to Vancouver and, in consequence, a slight change in the optical alignment. We could not detect any change in the ILS after shipment back to Germany from Singapore. Thus, we adjust our measurements with the factors derived from the TCCON side-by-side measurements on 23 July 2019. Even after applying the scaling factor, Fig. shows that there is some residual differences between the EM27/SUN and TCCON data growing towards the afternoon. At present, the origin of these differences is unclear. [54] suggest further investigations of the TCCON $\mathrm{XCO}$ scaling factor based on comparisons of the TCCON to the Network for the Detection of Atmospheric Composition Change (NDACC) and AirCore measurements. Should the TCCON scaling factor be updated in the future, our $\mathrm{XCO}$ data will be scaled accordingly.

Graph: Figure 4 Side-by-side measurements of XCO2 (a, b), XCH4 (c, d), and XCO (e, f) by the shipborne EM27/SUN and the TCCON station at Karlsruhe on 30 April 2019 (a, c, e) and 7 July 2019 (b, d, f). EM27/SUN measurements before and after scaling are shown in orange and blue, and TCCON records are shown in red.

Table 3 Scaling factors $\mathit{\gamma }$ for EM27/SUN ${\mathrm{XCO}}_{\mathrm{2}}$ , ${\mathrm{XCH}}_{\mathrm{4}}$ , and $\mathrm{XCO}$ observations as derived from the side-by-side measurements at the TCCON station Karlsruhe (Eq. ). The uncertainty is the standard error of the mean of the hourly data (Eq. ).

Figure  shows the quality-filtered and bias-corrected ${\mathrm{XCO}}_{\mathrm{2}}$ , ${\mathrm{XCH}}_{\mathrm{4}}$ , and $\mathrm{XCO}$ records for our cruise over the Pacific Ocean. The trajectory largely follows 30 ${}^{\circ }$  N of latitude between 140 ${}^{\circ }$  W and 120 ${}^{\circ }$  E of longitude, crossing the date line on 14 June 2019. For statistical analysis, we calculate hourly means of our records and use the campaign averaged SD of the hourly means as a measure for our precision, which amounts to 0.24  $\mathrm{ppm}$ (0.06 %), 1.1  $\mathrm{ppb}$ (0.06 %; parts per billion), and 0.75  $\mathrm{ppb}$ (1.03 %) for ${\mathrm{XCO}}_{\mathrm{2}}$ , ${\mathrm{XCH}}_{\mathrm{4}}$ , and $\mathrm{XCO}$ , respectively. The data records clearly show that we sampled background air masses for most of the time. We calculate a campaign mean and SD throughout the longitudinal section for each species, finding means and SDs as little as $\mathrm{411.6}\pm \mathrm{0.6}$   $\mathrm{ppm}$ for ${\mathrm{XCO}}_{\mathrm{2}}$ , $\mathrm{1835}\pm \mathrm{7}$   $\mathrm{ppb}$ for ${\mathrm{XCH}}_{\mathrm{4}}$ , and $\mathrm{71}\pm \mathrm{5}$   $\mathrm{ppb}$ for $\mathrm{XCO}$ .

Graph: Figure 5 XCO2(a), XCH4(b), and XCO(c) measured by the shipborne EM27/SUN (blue) above the Pacific alongside coincident CAMS atmospheric composition analysis data (green) and coincident XCO satellite observations by TROPOMI (orange). EM27/SUN measurements and CAMS data are hourly averages, while TROPOMI observations are averaged per overflight. Single TROPOMI measurements are marked small in the background.

Figure  compares our observations to column-averaged dry-air mole fractions we calculated from vertical profiles of the CAMS atmospheric composition analyses. These profiles are the same as those we use as a priori for our retrieval. Therefore, the differences between our retrievals and CAMS have no contribution from the a priori profiles being different from CAMS. Furthermore, Fig.  shows TROPOMI $\mathrm{XCO}$ observations for which we apply coincidence criteria of 0.5 ${}^{\circ }$ radius and 4 h time span. The TROPOMI $\mathrm{XCO}$ data are available at [18] and have been retrieved by the SICOR algorithm [37] which allows for retrievals above clouds. The latter capability is important over the ocean since the ocean is dark implying large noise unless the ocean-glint spot is observed or clouds offer a bright reflection target. After filtering with TROPOMI's quality flag (i.e., the internal quality descriptor bounded by 0 and 1 must be larger than 0.5), we find 1783 coincident TROPOMI $\mathrm{XCO}$ measurements distributed among 19 d. Although TROPOMI has ${\mathrm{XCH}}_{\mathrm{4}}$ measurement capabilities, we do not discuss these here since there is currently no ocean data available. Likewise, we do not show any OCO-2 or GOSAT data since the number of coincidences was 43 and 9, respectively, and limited to individual days, which we consider too few for a robust statistical analysis. The SICOR algorithm uses the global chemistry transport model TM5 [35] as an a priori source, which introduces a difference in the comparison to our EM27/SUN $\mathrm{CO}$ measurements with a priori profiles from CAMS [4]. We calculate the difference due to the a priori profiles for each EM27/SUN observation and find it to be $\mathrm{0.11}\pm \mathrm{0.40}$   $\mathrm{ppb}$ (campaign mean $\pm$ SD) with a maximum of 0.92  $\mathrm{ppb}$ . This contribution is small but not entirely negligible compared to the differences we find between our data and TROPOMI CO.

Graph: Figure 6 Differences between CAMS and our shipborne EM27/SUN (green) for XCO2(a), XCH4(b), and XCO(c) and between TROPOMI XCO and our EM27/SUN (orange). The EM27/SUN measurements were subtracted from the CAMS data and coincident TROPOMI observations.

Figure  depicts the differences between CAMS and our data and between TROPOMI and our data. We average the differences to CAMS over the entire campaign and calculate the standard deviations of the differences, which amount to $\mathrm{0.52}\pm \mathrm{0.31}$   $\mathrm{ppm}$ for ${\mathrm{XCO}}_{\mathrm{2}}$ , $\mathrm{0.9}\pm \mathrm{4.1}$   $\mathrm{ppb}$ for ${\mathrm{XCH}}_{\mathrm{4}}$ , and $\mathrm{3.2}\pm \mathrm{3.4}$   $\mathrm{ppb}$ for $\mathrm{XCO}$ (see also Table ). Thus, CAMS shows excellent agreement with the shipborne measurements within $\mathrm{1}\mathit{\sigma }$ for ${\mathrm{CH}}_{\mathrm{4}}$ and $\mathrm{CO}$ and $\mathrm{2}\mathit{\sigma }$ for ${\mathrm{CO}}_{\mathrm{2}}$ . Maximum differences between the hourly means of CAMS and EM27/SUN observations amount to $\mathrm{1.0}\pm \mathrm{0.2}$   $\mathrm{ppm}$ ${\mathrm{XCO}}_{\mathrm{2}}$ , $\mathrm{13.8}\pm \mathrm{1.3}$   $\mathrm{ppb}$ ${\mathrm{XCH}}_{\mathrm{4}}$ , and $\mathrm{10.3}\pm \mathrm{0.5}$   $\mathrm{ppb}$ $\mathrm{XCO}$ , in which the range is the propagated error using the SDs of the hourly mean values. For ${\mathrm{XCH}}_{\mathrm{4}}$ , CAMS tentatively shows an underestimation by a few parts per billion around the date line and an overestimation around 160 ${}^{\circ }$  E. Further, the intra-day variability of ${\mathrm{XCH}}_{\mathrm{4}}$ shows a systematic difference on the order of a few parts per billion. However, there is no consistent intra-day pattern that fits all the campaign days. Likewise for $\mathrm{XCO}$ , there is an intra-day residual pattern on the order of a few parts per billion but no consistency that informs us of potential model errors or shortcomings of the shipborne measurements.

Table 4 Comparison of the EM27/SUN observations to CAMS model data and coincident TROPOMI $\mathrm{XCO}$ measurements. The data indicate the mean differences $\pm$ the SD of the differences for the entire record.

Graph: Figure 7 Absolute differences between TROPOMI XCO observations and our EM27/SUN measurements plotted as a function of the SICOR/TROPOMI retrieval parameters scattering layer height (a), scattering optical thickness (b), interfering CH4 (c, from weak two-band total column), and interfering H2O(d). The color code indicates (logarithmic) occurrence, and the red line is a linear fit with its coefficient of determination R2 given in the upper right of each plot.

TROPOMI $\mathrm{XCO}$ also shows very good agreement with our data. The mean difference and standard deviation among the entire campaign record is $\mathrm{2.2}\pm \mathrm{6.6}$   $\mathrm{ppb}$ without any systematic pattern correlating with the position in the Pacific Ocean. [6] improved the SICOR/TROPOMI $\mathrm{CO}$ data product in comparison to TCCON observations by adjusting the spectroscopic database, decreasing the global mean bias below 1  $\mathrm{ppb}$ compared to TCCON station records with an SD of 2.6  $\mathrm{ppb}$ and a TCCON station-to-station bias variation of 1.8  $\mathrm{ppb}$ . We investigated whether the small residual differences correlate with the cloud parameters or interfering absorber abundances that SICOR/TROPOMI retrieves simultaneously with $\mathrm{XCO}$ . Figure shows the correlations of the differences with the layer height and the scattering optical thickness of the cloud layer, as well as the atmospheric methane and water column retrieved by SICOR/TROPOMI. [37] find a retrieval bias in the case of $\mathrm{CO}$ enhancements in combination with clouds, which we can not assess from our background concentration observations. Furthermore, we find no correlation of the $\mathrm{XCO}$ differences with the atmospheric methane and water vapor columns retrieved by SICOR/TROPOMI. These two species show spectroscopic interferences with the $\mathrm{CO}$ absorption lines in the SWIR-3 band, and thus, they could be candidates for inducing retrieval errors. Overall, our evaluation suggests that SICOR/TROPOMI provides robust results over clouds and for ocean-glint observations.

The ${\mathrm{XCO}}_{\mathrm{2}}$ , ${\mathrm{XCH}}_{\mathrm{4}}$ , and $\mathrm{XCO}$ records are available for download on PANGAEA at 10.1594/PANGAEA.917240[34]. The CAMS ${\mathrm{CO}}_{\mathrm{2}}$ and ${\mathrm{CH}}_{\mathrm{4}}$ data used in the paper is the official CAMS GHG analysis (10.24380/654b-gm83, [10]). The data for ${\mathrm{CO}}_{\mathrm{2}}$ and ${\mathrm{CH}}_{\mathrm{4}}$ is available via request to Copernicus Service Desk by emailing to copernicus-support@ecmwf.int or via the CAMS enquiry portal in https://atmosphere.copernicus.eu/help-and-support (last access: 29 January 2020). The CO data is from the CAMS NRT analysis available for download at 10.24380/hhra-8c27[11].

We deployed an EM27/SUN FTS on board the German R/V Sonne on the MORE-2 campaign cruise from Vancouver to Singapore leaving port on 30 May 2019 and arriving on 5 July 2019. Compared to our precursor study [33], our instrument setup was able to withstand environmental conditions, and it ran largely without requiring on-site operating personnel. Plus, the sun-viewing FTS was augmented by another detector to collect solar absorption spectra of $\mathrm{CO}$ in addition to ${\mathrm{CO}}_{\mathrm{2}}$ and ${\mathrm{CH}}_{\mathrm{4}}$ [27]. We provide records of the column-averaged dry-air mole fractions ${\mathrm{XCO}}_{\mathrm{2}}$ , ${\mathrm{XCH}}_{\mathrm{4}}$ , and $\mathrm{XCO}$ for 22 d of measurements on the Pacific Ocean largely following 30 ${}^{\circ }$  N of latitude. Our observations are representative of global background conditions; thus, they are useful for assessing the performance of atmospheric models and satellite measurements without perturbations due to local atmospheric variability, and they add to the largely land-based validation data provided by the TCCON and the COCCON. Our measurements show an overall precision (hourly SDs averaged for the whole campaign) of 0.24  $\mathrm{ppm}$ for ${\mathrm{CO}}_{\mathrm{2}}$ , 1.1  $\mathrm{ppb}$ for ${\mathrm{CH}}_{\mathrm{4}}$ , and 0.75  $\mathrm{ppb}$ for $\mathrm{CO}$ . Systematic errors due to residual pointing uncertainties, sampling of the ship's exhaust plume, and a spurious dependency on the SZA are treated by filtering flawed data and by empirical corrections. We made our observations compatible with the TCCON through side-by-side measurements before and after the campaign at the TCCON station Karlsruhe. Through comparisons to our data, we evaluate the performance of the CAMS model for ${\mathrm{XCO}}_{\mathrm{2}}$ , ${\mathrm{XCH}}_{\mathrm{4}}$ , and $\mathrm{XCO}$ and the performance of $\mathrm{XCO}$ measurements by the TROPOMI instrument on board the Sentinel-5 Precursor satellite. Averaged over the entire campaign, the differences to CAMS amount to $\mathrm{0.52}\pm \mathrm{0.31}$   $\mathrm{ppm}$ for ${\mathrm{XCO}}_{\mathrm{2}}$ , $\mathrm{0.9}\pm \mathrm{4.1}$   $\mathrm{ppb}$ for ${\mathrm{XCH}}_{\mathrm{4}}$ , and $\mathrm{3.2}\pm \mathrm{3.4}$   $\mathrm{ppb}$ for $\mathrm{XCO}$ . Furthermore, we find the TROPOMI $\mathrm{XCO}$ of $\mathrm{2.2}\pm \mathrm{6.6}$   $\mathrm{ppb}$ to be in excellent agreement with the ground-based observations. The instrument is a valuable asset for the validation of satellite observations over sea. In the future, we plan to fully automate our instrument design for routine deployment on ships to enrich validation opportunities over the open oceans where other opportunities are sparse.

We acknowledge funding for the MORE-2 campaign by BMBF (German Federal Ministry of Education and Research). The development of the COCCON preprocessing tool has been supported by ESA in the framework of the COCCON-PROCEEDS project. The Copernicus Atmosphere Monitoring Service is operated by the European Centre for Medium-Range Weather Forecasts on behalf of the European Commission as part of the Copernicus program (, last access: 8 January 2021).