The EM27/SUN FTS (Bruker Optics, Germany) is the standard instrument of the COCCON network for land-based observations of XCO2, XCH4, and XCO . Modifications and performance demonstrations for ship-based deployments are described in , , and . Summarizing the most important features here, the EM27/SUN has an optical resolution of 0.5 cm⁻¹ and a semi-field-of-view of 2.36 mrad . The latest EM27/SUN version contains two detectors. The first InGaAs photodetector covers the spectral range between 5500 and 11 000 cm⁻¹ , where O2, CO2, and CH4 absorption lines are present. The second InGaAs-detector with spectral extension towards the infrared covers the wavelength range between 4000 and 5500 cm⁻¹ , where CO absorption features can be found. The EM27/SUN FTS demonstrated its versatility and robustness under various climatic conditions and in different settings , making it an ideal instrument for campaign deployments. In addition, demonstrated its suitability in terms of stability and comparability for network applications such as those implemented through COCCON.

Our EM27/SUN is operated and characterized according to COCCON protocols, with two modifications. First, to avoid data loss in the case of tracker instability, we record individual interferograms for each forward-backward scan rather than relying on automatic co-adding of 10. Second, we use an interferogram sampling rate of 20 kHz (instead of 10 kHz). This leads to one forward-backward scan being completed approximately every 15 s. In addition to this, the solar tracker of the standard EM27/SUN is replaced by a custom-built system to be able to compensate for the rather quickly changing attitude and orientation of a ship platform. The high precision of our custom-built tracking system was demonstrated for shipborne deployments in , and for balloonborne deployments in , where more details on the tracking system can also be found. The instrument is housed in an aluminum box to protect it from harsh weather. The entire box is ventilated and equipped with various ancillary sensors for pressure and temperature measurements, and its position is tracked via GPS. Mats have been placed underneath the instrument to damp it from vibrations caused by the engine of the ship. Figure  shows the box with the instruments on board the commercial vessel Nichiyu Maru (Ocean Link, Ltd.) during the deployment described in Sect. .

To supplement the setup with measurement capabilities for NO2, we co-mounted a grating spectrometer measuring solar absorption spectra in the visible (VIS) spectral range. Using differential optical absorption spectroscopy (DOAS) , NO2 atmospheric column abundances are accessible. The DOAS instrument uses part of the light beam from the solar tracker, which is not needed for the FTS. This is possible since the beam from the solar tracker has a diameter approximately twice as large as the aperture of the FTS. The DOAS itself needs less than 0.15 % of the total beam diameter. The aperture of both instruments is sufficiently far away from the fringes of the beam of the tracker mirror. The part of the solar beam needed for the DOAS is coupled into an optical fiber (Laser Components FVP-400) using a custom-built telescope assembly similar to the one used in and . The assembly consists of the following components, as schematically shown in Fig. : A prism (Thorlabs PS910) positioned under total-reflection relays the incoming beam into a lens tube with a bandpass filter centered at 460 nm (Edmund Optics Hoya B460). The field of view of the telescope is limited to 1.15° by focusing the light onto an 800 µm aperture (Edmund Optics 34-445) using a bi-convex lens (Thorlabs LB1027-A). Furthermore, the light is diffused before entering the fiber by a polytetrafluorethylene reflector plate mounted at a 45° angle with respect to the optical axis. The fiber is mounted so that only light from the reflector plate can enter it. The fiber is connected to an Ocean Optics QE-Pro spectrometer operating in a Czerny-Turner configuration in the visible spectral range.

The technical data of the VIS spectrometer and the settings used during the deployment described in this work are listed in Table . The spectrometer entrance slit with a width of 100 µm defines the light throughput and the width of the spectral response function (SRF). The spectral sampling amounts to 5 detector pixels per full-width-at-half-maximum (FWHM) of the SRF, corresponding to 0.7 nm, as determined by measuring individual spectral emission lines from mercury and krypton emission lamps. The DOAS spectrometer is configured such that it records spectra in a wavelength range between 400.7 and 495.9 nm. Two different exposure times are used to maximize the signal but prevent saturation of the detector. At high solar zenith angles (SZA) during the morning and evening, spectra are recorded with an exposure time of 700 ms. At low SZA, the instrument is switched automatically to an exposure time of 350 ms.

A two-stage cooling is used to keep the DOAS spectrometer and its detector at constant temperatures. The first stage is the QE-Pro spectrometer's internal thermoelectric cooler (TEC). It can cool the detector down to 20 K below the ambient temperature. In addition, the spectrometer is housed inside a custom-built enclosure, which is sealed so that water vapor is prevented from entering by allowing exchange with the ambient air only through a silica gel tube. A second TEC is used to regulate the temperature inside this enclosure to an accuracy of ±0.1 K. This is, however, only possible for a temperature range of 2 K below and 15 K above the outside temperature. Since outside temperatures in the range between 0 and 40 °C occurred during the deployment, the external TEC was operated at temperatures between 10 and 38 °C, and the internal TEC at temperatures of up to 27 °C, which is not ideal in terms of noise reduction but ensures temperature stability throughout the deployment. An upgrade of the external TEC, facilitating stabilization at lower temperatures, is planned for the future. The current setup, however, already leads to a temperature stability of the DOAS detector better than ±0.0015 K on all days on which a proper temperature range with respect to the ambient conditions was selected.

For the campaign reported here, the SRF of the DOAS spectrometer was recorded by measuring the emission lines of krypton or mercury lamps. We conducted these lamp measurements before and after the campaign deployment and approximately every six weeks while the instrument was onboard the ship. All SRFs show shape deviations of less than 4 % relative to the peak height of the first SRF measurement. In addition, it was verified by re-running the analysis (see Sect. ) with several different SRFs measured throughout the campaign that changing the SRF used in the retrieval leads to changes of the result for the vertical column density (VCD) of NO2 of up to 0.75 %, which is much smaller than the overall retrieval error between 2 % and 14 % (depending on weather conditions). This leads to the conclusion that the grating spectrometer remained stable throughout the deployment. Since the remarkable ensemble performance of DOAS instruments was already demonstrated, e.g., in , and the stability and high accuracy of QE Pro spectrometers were demonstrated during long-term ground-based deployments, as well as during aircraft-based and balloonborne measurements , additional performance tests, as done for the FTS (see Sect. ), are not needed for this instrument.

During previous campaigns, the instrument required an operator onboard the vessel (e.g., ). For the present campaign, the instrument was upgraded to operate fully automatically and to allow for remote access. To this end, custom-built software packages were used to automatically operate the two spectrometers, the temperature stabilization, all ancillary sensors, and the solar tracker. A maintenance crew only needed to go on board the vessel approximately every six weeks to exchange the hard drives used to store the data and to perform characterization measurements. In order to allow for remote access for quick instrument checks and to notify the maintenance crew in case of instrument errors, access via the mobile phone network using an MC Technologies MC100 network switch with a Quectel EG21-1 module was installed. A 1.5 m long 790–960 MHz multiband antenna from MC Technologies with 6 dBi antenna gain was mounted on the ship's railing. Since the ship was traveling along the coast of Japan (see Sect. ), this antenna was large enough to connect to the instrument on about 80 % of the route.

In a first step, the interferograms recorded by the EM27/SUN FTS need to be Fourier transformed into spectra. To this end, as in our previous work, the preprocessor of the PROFFAST retrieval software is used, which is also employed for EM27/SUN data processing within the COCCON network . We use a spectral window of 3500 to 14 000 cm⁻¹ for channel 1 and 3000 to 5200 cm⁻¹ for channel 2 of the FTS, a phase correction, and a Norton-Beer medium apodization throughout this study. Before the Fourier transform, the direct current (DC) component of the interferograms is used to filter out scenes with low brightness or large brightness fluctuations following and . Large brightness fluctuations indicate clouds passing by or the tracker not fully compensating for the ship's motion. Following , the DC-variability of an interferogram can be defined as 1DCvar=max(IDC)min(IDC)-1 where IDC is the intensity of the DC-component of the interferogram. All interferograms with a DC-variability greater than 5 % throughout the integration time of the measurement are filtered out. In addition, a low average DC value can indicate clouds passing in front of the sun. Therefore, interferograms with an average DC value lower than 5 % are excluded from any further analysis.

The algorithm RemoTeC performs the actual spectral retrieval. It was chosen here since the standard algorithm of the COCCON network, PROFFAST, currently does not yet accept moving instruments. The RemoTeC algorithm was originally developed for the retrieval of gas abundances from the GOSAT satellite but has since been applied to various satellites, including OCO-2 and TROPOMI . Adapted versions for ground-based applications exist and have been applied to a variety of settings (e.g., ). For ground-based direct sun applications, RemoTeC neglects scattering, reducing the radiative transfer equation to Beer-Lambert's law. A priori profiles for the retrieval are calculated based on meteorological data from NCEP , in situ ground-based pressure and temperature measurements, standard profiles for CH4 and CO2 from tracer model 4 (TM4) and CarbonTracker as well as standard profiles for CO based on tracer model 5 (TM5) . Spectral line parameters are taken from HITRAN 2016 . The solar top-of-the-atmosphere reference spectrum is provided by Geoffrey Toon (2016, personal communication).

Formally, RemoTeC retrieves partial column densities for six altitude layers, which are equidistant in pressure, employing a Phillips-Tikhonov regularization scheme . Each target species' vertical column density (VCD) is then calculated by summing up the partial columns in these six layers. The forward model itself is calculated for 72 atmospheric layers based on Voigt line shapes. The forward model convolves the line-by-line spectra by the instrument SRF (often also called instrument lineshape (ILS) in the context of FTS) to simulate the observed spectra. The SRF was recorded on 23 August 2023, following the procedures of the COCCON protocol . Table  lists the spectral windows used for the retrieval of CO2, CH4, and CO as well as for molecular oxygen O2. Water vapor H2O is an interfering absorber.

We use the retrieved O2 and H2O column densities to calculate the total pressure at the instrument level and call it “spectroscopic” pressure pspec since it is derived from spectroscopic measurements. For quality assurance, the spectroscopic pressure can be compared to the in situ surface pressure pin-situ measured by a co-deployed pressure sensor (cf. also ). Deviations (on top of trivial offset factors between spectroscopic and in situ data) indicate potential error sources of our setup, such as pointing errors of the solar tracker, drifts of the optical alignment, or spectroscopic parameter errors. Therefore, we use spectroscopic pressure as a quality filter as follows:

We remove outliers with 1-pspecpin-situ<0.05.

Since, in contrast to CO2 or CH4 absorption in the near-infrared, absorption optical thicknesses of NO2 are typically minor in the visible spectral range, we can apply the DOAS technique to infer NO2 column densities from the measurements of the VIS spectrometer. The DOAS retrieval consists of several steps. A dark current, offset, and non-linearity correction is performed during preprocessing. Dark current and offset spectra are measured every night at 02:00 a.m. local time throughout the campaign and are then used for the respective correction. A non-linearity correction is performed using a ninth-degree polynomial, which was determined in the laboratory based on measurements of a halogen light source recorded at various exposures. Afterwards, typically 100 spectra (corresponding to between 30 and 90 s of measurements, depending on the exposure time) are co-added. The co-addition is stopped at less than 100 spectra if the exposure time changes, the spectra have a saturation of less than 8 % (which is typical for measurements during sunrise and sunset or during the presence of larger clouds), large saturation fluctuations are detected (which relates to perturbed sun-tracking), the time between two subsequent spectra is longer than 70 s, or spectra are missing between subsequent measurements. All co-added spectra consisting of fewer than 50 measurements are discarded.

After the preprocessing, the DOAS fit is performed using the software package QDOAS . QDOAS recalibrates the initial pixel-to-wavelength mapping based on the solar Fraunhofer lines as a first step. Then, the actual DOAS fit, based on Beer-Lambert's law, finds the differential slant column densities (dSCDs) of NO2 and interfering absorbing gases together with ancillary parameters such as spectral shifts, a background polynomial, an intensity offset parameter accounting for instrument straylight, as well as a center-to-limb-darkening (CLD) and Ring effect correction required for direct-sun measurements in the presence of scattering due to high-altitude clouds. The latter is a typical weather condition during the deployment of the instrument. All DOAS fit parameters are detailed in Table , and a list of all absorption cross-sections used can be found in Table .

One of the key aspects of DOAS is that the retrieval finds dSCDs with respect to a reference spectrum I0 measured by the very same spectrometer. In our case, this reference spectrum, taken for the entire campaign dataset, is recorded around noon under minimal airmass and cloud-free conditions on 17 October 2023. To get the total slant column densities (SCDs) throughout the atmosphere, we need to add the absorber slant column density contained in the reference spectrum SCD_ref to the dSCDs found by the DOAS fit. To find SCD_ref, we use an approach similar to the procedures of the PANDONIA network and the bootstrap method by . The approach builds on Langley's method , which assumes the following linear relation for absorber i 4dSCDi=VCDi×AMFi−SCDref where AMFi is the air-mass factor, which can be approximated as AMF=1cosΘ as long as SZA Θ<70° . Thus, SCD_ref is the ordinate intercept of a plot of the dSCDs versus the AMF.

Langley's method is only applicable for background measurements where the gas concentrations do not change over the range of SZA used for the plot, i.e., over the course of a day. Since our NO2 measurements are conducted in the vicinity of local NO2 sources and NO2 is variable due to photochemistry and meteorological transport, our measurements do not fully comply with this assumption. However, since we use a single reference spectrum for the entire campaign period, we can assemble a composite Langley plot for NO2 spanning many days as shown in Fig. . The composite Langley plot reveals a sharp, well-defined lower boundary, defined by background conditions composed of many individual days. To find SCD_ref, we define background conditions as the 20 % lowest dSCDs in each 0.1 AMF bin (orange data in Fig. ) and fit the linear relationship given by Eq. () to this background data. For uncertainty estimation, we use a bootstrap method, repeating the fit procedure multiple times with randomly chosen subsets of the background data and taking the maximum and minimum of the found SCD_ref as the error range. The error introduced by arbitrarily choosing 20 % as a threshold was estimated by re-running the whole analysis with different thresholds between 5 % and 40 %. The total error is then the Gaussian sum of the error of the bootstrap realizations (0.51×1015 molec. cm⁻²) and the error determined by changing the threshold (0.68×1015 molec. cm⁻²). Based on this method, we find SCDref=(-7.02±0.85)×1015 molec. cm⁻² for NO2, which is used to calculate total SCDs and together with the AMF the respective VCDs of NO2.

Finally, quality filtering is applied to the NO2 VCDs. Measurements with SZA >70° are filtered out since the cosine approximation used to calculate the AMF is not valid due to Earth's curvature becoming non-negligible . Also, all measurements with a root-mean-square spectral fitting error (RMSE) of more than five times the average RMSE of the respective measurement day are discarded since a large RMSE indicates brightness fluctuations during the measurement, e.g., caused by variable cloud cover disturbing the solar tracking. Finally, for calculating ratios with respect to gases measured by the EM27/SUN, the NO2 VCDs are also downsampled to the frequency of the EM27/SUN measurements by averaging the DOAS measurements to the time intervals of the EM27/SUN dataset. This averaging results in one datapoint every 2.5 to 3 min.

Shipborne deployments are a major challenge for FTS. Especially shocks during the transport of the instrument to the ship or constant vibrations on board, e.g., caused by the engine of the ship, may lead to misalignments or instrumental drift. For future operation of our instrument, for example, within the COCCON, it is crucial to verify that it remains stable throughout the deployment and can be operated within the error margin of land-based spectrometers. The important criteria for benchmarking the stability of an EM27/SUN FTS are the change of the spectral response function (SRF, often also referred to as instrument line shape (ILS)) and the drift of the spectrometer with respect to an instrument of a high-resolution, ground-based FTS network such as the Total Column Carbon Observing Network (TCCON). These two criteria were monitored throughout the deployment and are discussed in the following. For additional performance tests, including an evaluation of the precision of the tracking system while the ship was in motion, the reader is referred to and .

SRF measurements with the FTS were conducted before and after the deployment, as well as during the ship's maintenance break. These measurements were conducted following the procedure described in under controlled laboratory conditions. SRF measurements are impossible on the ship itself since the lamp assembly cannot be reliably set up onboard. All SRF measurements are evaluated with the retrieval software LINEFIT 14.6 as described in and to be consistent with SRF measurements from previous campaigns.

Panels a and c of Fig.  show the modulation efficiency (ME) and the phase error (PE), respectively, of all SRF measurements conducted during the current deployment with the FTS. Since the instrument was not re-aligned since 2019, previous SRF measurements already reported in and , as well as previously unpublished SRF results from a deployment in 2022, are included in the figure. It should be noted that an abrupt SRF change reported in , which was at that time attributed to a shock during the transport from Heidelberg to Vancouver, was, in the meantime, attributed to a malfunction of the ancillary humidity sensor. After correction of this malfunction, the previously reported change disappears, as can be seen in Fig. .

During the current, latest deployment, our instrument has a ME of 0.9908±0.0023 and a PE of 0.0013±0.0013. If the whole data since 2019 is considered, our instrument has a ME of 0.9895±0.0034 and a PE of 0.0025±0.0016. These numbers for ME and PE and the respective errors are similar to those reported by for the COCCON travel standard (ME: 0.9805±0.0027, PE: -0.0020±0.0006) and the reference instrument SN-37 of the COCCON network (ME: 0.9836±0.0027, PE: 0.0015±0.0012). Our results for ME and PE are also within the range known from other COCCON instruments as reported by , and our standard deviation for ME for the current campaign is smaller than the error budget of 0.29 % derived by for the SRF retrieval procedure used in our study. Based on this, we conclude that despite frequent transports by aircraft and potential rough handling during craning, the SRF of our instrument has remained stable, and its changes are within the error margin of other instruments in the COCCON network. Thus, the same SRF measurement is used to evaluate the whole deployment (see Sect. ).

As recommended by , the SRF of the second channel used for the retrieval of CO is also evaluated and compared to a SRF retrieval from the main detector channel based on water vapor absorption in the 5275 to 5400 cm⁻¹ spectral window, where both detectors are sensitive. The results are shown in panels b and d of Fig. . The deviations between the ME and the PE of the two channels of our EM27/SUN are similar to those reported by for other instruments of the COCCON network. In addition, changes between the different measurements are similar to those for the standard SRF retrieval shown in panels a and c of Fig. . Only in May 2024, the ME of channel 2 (black dots in Fig. ) is found somewhat higher than before and is also higher than the ME for detector channel 1. It is assumed that this change is caused by an incident where the crane bumped the spectrometer against parts of the ship when it was lifted to the upper deck in February 2024, leading to a slight misalignment of the second detector channel. Since we consider the SRF measurement based on channel 1 more reliable, we have used an SRF from that channel for the final retrieval described in Sect. .

In addition to monitoring the SRF, our instrument has been regularly compared to an instrument of the COCCON network (SN-147). Roughly every six weeks, we operated the two instruments side-by-side in the harbor of Kawasaki (Tokyo metropolitan area). For this, both EM27/SUN instruments were placed on the deck of the vessel Nichiyu Maru. Additional comparison measurements in the courtyard of NIES in Tsukuba were performed before and after the campaign, as well as during a maintenance break in February 2024. The COCCON instrument itself is regularly compared to the TCCON station in Tsukuba to ensure a good quality of the COCCON instrument.

To evaluate the stability of our EM27/SUN with respect to the COCCON instrument, average ratios between the two instruments are calculated for the VCDs and XGAS for all individual days when side-by-side measurements were conducted. The data reduction for measurements of the shipborne EM27/SUN follows Sect. . The measurements of the COCCON instrument are processed according to the standard COCCON protocol , relying on the retrieval software PROFFAST.

Table  lists the daily ratios between the gas retrievals from the COCCON measurements and those of the shipborne EM27/SUN. Based on the seven comparison days, average ratios of 1.0287±0.0006 for XCO2, of 1.0269±0.0012 for XCH4, and of 1.0591±0.0231 for XCO are found. The overall ratios are larger than the ones between different COCCON instruments reported by , which is related to differences in RemoTeC and PROFFAST. Most importantly, PROFFAST corrects for a spectroscopic error in the O2-column, which accounts for a bias of about 2 % , and already has an overall scaling factor with respect to the standards of the World Meteorological Organization (WMO) included, while RemoTeC leaves these corrections to the post-processing. In addition, our RemoTeC retrieval uses a different spectroscopic database (see Sect. ) than PROFFAST, which leads to well-known offsets . A correction for the difference between RemoTeC and PROFFAST is later applied through an overall scaling factor with respect to TCCON in the retrieval as described in Sect. .

Inter-day differences are small for XCO2 and XCH4 as well as for the VCDs of O2, CO2, and CH4, as can be inferred from the standard deviations listed in Table . Inter-day differences are larger for XCO and the VCDs of CO, but the overall accuracy and precision requirement is much less stringent than for CO2 and CH4. For CO, intra-day differences show an SZA dependency on the same order of magnitude as the inter-day variability.

Our standard deviations for XCO2 and XCH4 listed in Table  have a similar order of magnitude as the ones reported by for the reference instrument of COCCON with respect to TCCON (0.0015 for XCO2, 0.0024 for XCH4), which are, however, based on data from a much longer time series. Thus, our assessment suggests that our shipborne EM27/SUN complies with the standards of the COCCON network. The overall scaling factors for our instrument with respect to TCCON were calculated by multiplying our scaling factors from the comparison measurement in Kawasaki on 28 October 2023 with average scaling factors from comparison measurements between the COCCON instrument and the TCCON station in Tsukuba made in 2023 (see Table ). The TCCON data is retrieved with the standard GGG2020 algorithm .