The Sentinel-5 Precursor (S5P) satellite was successfully launched on 13 October 2017, from Plesetsk in northern Russia with the Tropospheric Monitoring Instrument (TROPOMI) aboard. The instrument is a grating spectrometer which measures sunlight reflected by the Earth's atmosphere and its surface from the ultraviolet to the shortwave infrared (SWIR) with daily global coverage, a spatial resolution of about 7×7 km2, and a high radiometric accuracy to infer the carbon monoxide (CO) total column over dark vegetation surfaces with a precision of <10 % . One of the primary targets of the mission is to monitor the atmospheric concentration of CO. This trace gas is emitted to the atmosphere by incomplete combustion, e.g., by traffic, industrial production, and biomass burning. Its major sink is the reaction with the OH radical . With a typical background concentration of ca. 80 ppb (in the Northern Hemisphere) and an atmospheric residence time from days to months , the trace gas is established as a tracer of how pollution is transported, redistributed, and depleted in the atmosphere.

The S5P mission builds upon the heritage of SCIAMACHY (Scanning Imaging Absorption Spectrometer for Atmospheric Chartography; ), which provided atmospheric CO total column concentrations from the same spectral range . Measurements of SCIAMACHY in the SWIR have a spatial resolution of about 30 km × 120 km (along-track × across-track) for an integration time of 0.5 s with a global coverage cycle of 3 days. Most importantly, the SCIAMACHY noise error of single CO retrievals can exceed 100 % for dark scenes. Therefore, spatial and temporal averaging of single CO measurements is required , which limits the data interpretation of SCIAMACHY CO data.

From space, CO is also measured by other satellite instruments with global coverage, e.g., MOPITT (Measurements of Pollution in the Troposphere; ), AIRS (Atmospheric Infrared Sounder; ), TES (Tropospheric Emission Spectrometer; ), IASI (Infrared Atmospheric Sounding Interferometer; ). The S5-P mission is the first of a sequence of the European Space Agency's (ESA) atmospheric composition satellites, which also comprises the Sentinel-5 mission, a series of spectrometers with the first launch in the 2021–2023 time frame.

In this study, we use the Shortwave Infrared CO Retrieval Algorithm (SICOR). It is developed by SRON (the Netherlands Institute for Space Research), for the operational processing of TROPOMI data and also serves as algorithm baseline for the data processing of the successor mission Sentinel 5. The algorithm infers the vertical column concentration of CO (the vertically integrated amount of CO above the surface) simultaneously with effective cloud parameters from TROPOMI's 2.3 µm spectra . A first comparison of the TROPOMI CO data product with CO fields from the European Center for Medium-Range Weather Forecast (ECMWF) was performed by . Based on this, our study deploys SICOR on TROPOMI measurements taken during the early months of the instrument in orbit, and show the capability of the instrument to detect and monitor the air pollution from hotspots like larger cities and industrial regions. Investigating the temporal evolution of CO enhancements over the Atlantic and Indian oceans shows the capability of the instrument to track the atmospheric transport of pollution on a day-to-day basis in agreement with co-located ground-based measurements. Moreover, a validation with collocated ground-based Fourier-transform spectrometer (FTS) measurements at nine TCCON sites, indicates the TROPOMI CO data quality. The paper is structured as follows: sect.  describes the dataset and methodology and Sect. 3 presents our analysis of the TROPOMI CO data product comprising a first validation with TCCON ground-based measurements, the detection of CO hotspots and the transport of CO pollution over the oceans. Section  gives the conclusions of the study and finally the data availability is described in the data availability statement.

For this study, we used nominal Earth radiance and Solar irradiance measurements of TROPOMI during the commissioning phase of the instrument from 9 November 2017 to 4 January 2018. We deployed the SICOR algorithm on the 2.3 µm spectra of TROPOMI and retrieved the total column density of CO simultaneously with interfering trace gases and effective cloud parameters (cloud height, z, and optical thickness, τ) describing the cloud contamination of the ground scene . The retrieval approach is based on the profile scaling method and the implementation and retrieval settings are discussed in detail by . The reference profile of CO that is scaled during the retrieval is taken from simulations of the global chemical transport model TM5 and monthly averaged over 3∘×2∘ latitude and longitude grid boxes. Therefore, the retrieval result is the total column density of CO [molec cm-2]. To compare it with other measurements we also represent the data product as a dry column mixing ratio XCO [ppb] by dividing the CO total column density by the dry air column density derived from co-located ECMWF pressure fields.

For the data analysis, we performed a posteriori quality filtering of the TROPOMI data. To this end, we used retrievals with a solar zenith angle <80∘ and discarded the two most westward ground pixels of the swath due to a performance issue that is still under investigation. Furthermore, we distinguished between retrievals under clear-sky (τ<0.5 and z<5 km, over land) and cloudy condition (τ>0.5 and z<5 km, over land and ocean). The remaining retrievals are not considered in this study. Clear-sky observations over ocean, which is a dark surface in the SWIR, cannot be used for data interpretation because of a too low signal.

For the total column of CO, the vertical sensitivity of the retrieval is described by the total column averaging kernel , which is illustrated in Fig.  for TROPOMI data of one particular day, 10 November 2017. It shows the dependence of the averaging kernel on the cloudiness of the scene, where the standard deviation indicates its variation due to different observation and atmospheric parameters, e.g., solar zenith angle, viewing zenith angle and ground reflectivity. For very strict cloud clearing of the data (with z<5 km and τ<0.01), the total column averaging kernel is close to 1 for all altitudes with little variation, meaning that the derived column can be interpreted as an estimate of vertically integrated amount of CO. Filtering the data less strict using the clear-sky filter from above (z<5 km and τ<0.5) results in a slightly reduced sensitivity with a moderate standard deviation and concluded that those measurements are usually clear-sky equivalent for remote regions without local pollution sources and the induced errors due to the choice of the reference profile to be scaled by the inversion to be on a percentage level . The presence of clouds changes significantly the vertical sensitivity of the retrieval. Figure  shows the column averaging kernel when filtering for optical thick clouds at 5 km altitude. The sensitivity below the cloud is significantly reduced (values lower than 1) due to cloud shielding, and the retrieval estimates a CO total column mainly based on the measurement sensitivity to CO above the cloud (values higher than 1).

Consequently, the direct comparison of reference measurements with the retrieved CO columns from cloud contaminated TROPOMI measurements can lead to errors >30 % . This so called smoothing error is due to imperfect knowledge of the vertical profile of CO. However, the TROPOMI CO dataset provides total column averaging kernels acol for each retrieval. To compare a vertical profile ρ, e.g., from airborne in situ measurements or model simulations with the TROPOMI CO product a total column concentration c=acolρ can be calculated from ρ. This can be directly compared with the retrieval result as it is affected in the same way by the reduced sensitivity as the retrieval . When the reference measurement is not a vertical profile the application of the total column averaging kernel becomes more difficult. In that case, the TROPOMI CO dataset can be filtered for retrievals under clear-sky conditions to avoid misinterpretations. Alternatively, an approach as presented by can be followed who quantified expected differences in GOSAT/TCCON CO2 retrievals due to averaging kernel differences using the GEOS-Chem model to simulate a realistic range of CO2 profiles.

The TROPOMI instrument is still in the early phase of the mission and the performance of the CO retrieval is expected to improve in the future. For example, single overpasses show stripes of erroneous CO in flight direction, probably due to calibration issues of TROPOMI. Considering high-frequency variations of CO measurements across flight direction per orbit, we infer the stripe pattern by median filtering of the detected features in flight direction per orbit. Figure  provides an example, where the average the average of the stripe pattern in cross flight direction is -0.03 ppb with a standard deviation of 1.1 ppb. Some stripes can reach values higher than 5 ppb. Therefore, the stripe pattern can be removed from the data a posteriori to the retrieval and its removal is indicated accordingly in the remainder of the paper. suggested a similar approach to improve the quality of the NO2 data product of the Ozone Monitoring Instrument (OMI, ).

The CO observations of TROPOMI represent a major step forward for the monitoring of air pollution from space. In this study we investigated the quality of the CO data product applying the operational retrieval software to the first two months of data of the S5-P mission in space. As only a small fraction of all TROPOMI observations are expected to be cloud free , it is essential to account for clouds in the retrieval. Building on previous work , an important feature of the TROPOMI CO retrieval is to infer cloud parameters and trace gas columns at the same time to achieve a good yield from the data processing. Thereby, we account for the vertical sensitivity of cloudy measurements and enhance the data yield both for observations over land and ocean . On average, 8 % of all measurements are clear-sky, 22 % cloudy-sky observations over land, and 51 % cloud-sky over the oceans.

A first validation of the TROPOMI CO data product with collocated TCCON observations at nine selected measurement sites showed good agreement with a mean bias of about 6 ppb for both clear-sky and cloudy observations and a mean standard deviation of 3.9 and 2.4 ppb, respectively, demonstrating a good repeatability of the observations. Here, the standard deviation of the station biases is 3.8 and 4.0 ppb for both types of measurements. Additionally, a comparison with ground-based FTS at Mexico shows that the CO pollution hotspot can be observed with high accuracy. For this study, only a limited amount of TCCON data were available with confined spatial and temporal coverage. The Sentinel 5 Precursor as an operational mission requires a continuous monitoring of the CO data quality, which will be performed as part of the operational validation activities. In this context, future work will consider the validation of the TROPOMI CO data for longer time scales including additional TCCON and NDACC stations to improve the significance of the product validation.

Due to the good data quality, in combination with the spatial resolution of 7×7 km2, TROPOMI can capture the variability of CO due to atmospheric transport of pollution on a day-to-day basis, demonstrated for daily overpasses at Ascension Island and Reunion in agreement with TCCON observations at these sites. Based on these preliminary results, we conclude that the TROPOMI CO product already fulfills the mission requirements on precision (<10 %) and accuracy (<15 %) , assuming a background concentration of 100 ppb.