3.1 MAX-DOAS measurements and instrument setup on the vessel

The MAX-DOAS instrument used in this study consisted of an Avantes spectrometer (AvaSpec-ULS2048x64) with a wavelength range from 288 to 500 nm, a spectral resolution of 0.6 nm full width at half maximum (FWHM), and a telescope unit which was installed aboard RV Maria S. Merian. The instrument was continuously scanning a vertical plane pointing portside perpendicular to the vessel's direction, in order to generally point towards the African continent. Each scan lasted about 10 min. In the vertical plane, elevation angles from −3 to 8^∘ were measured in 1^∘ steps as well as 10, 15, 30, and 90^∘. A maximum exposure time of 0.1 s was used in order to be able to account for the ship's movement (roll angle) during data analysis. After correcting the angles for the roll of the ship, the individual measurements within a 30 s period were sorted by viewing angle and binned into 1^∘ intervals. Zenith sky measurements were averaged over 60 s. For the analysis, elevation angles from 0 to 90^∘ above the horizon were used. Observations were taken at solar zenith angles (SZAs) smaller than 96^∘ for zenith sky measurements and for the off-axis at SZAs smaller than 90^∘.

The ship sailed at nearly constant speed over ground with ∼ 12–14 kn. The absolute wind direction was primarily from the south. Consequently, headwinds were dominant during the campaign, shifting either side of the bow (Fig. 3). Measurements with a relative wind direction between 90 and 270^∘ (grey-shaded wind directions; Fig. 3) are excluded from the analysis, because of possible contamination by the vessel's plume. Furthermore, an intensity filter was used to exclude poor viewing conditions during heavy rain events (i.e. very dark scenes) to avoid the entailed high uncertainties.

Figure 3Relative (a) wind speed and (b) wind direction (0^∘ indicates headwinds) during the COPMAR campaign. Grey shaded measurements are excluded from the analysis in order to avoid possible contamination by the vessel's plume.

Download

3.2 Satellite measurements

The measurement data obtained during the COPMAR project are compared with satellite measurements of absorber vertical column densities (VCDs) from the Ozone Monitoring Instrument (OMI; Levelt et al., 2006) and Global Ozone Monitoring Experiment 2A/B (GOME-2A/B; Callies et al., 2000; Munro et al., 2016) instruments.

For the comparison with stratospheric NO₂, data from satellite instruments (GOME-2A and GOME-2B) are used. To retrieve the latitudinal dependency of stratospheric NO₂, a monthly mean for all measurements having SZAs smaller than 85^∘ is calculated for October 2016 using the DOAS fit settings from Richter et al. (2011). The satellite data are averaged over longitudes between 10 and 40^∘ W and divided by stratospheric air mass factors (AMFs). Additionally, satellite observations for HCHO and CHOCHO are taken into account which are described in Alvarado et al. (2018) and Alvarado et al. (2019).

The measurement accuracy for satellite data is described in Boersma et al. (2004) and Lorente et al. (2017) as well as in Alvarado et al. (2014, 2019). Generally, over remote ocean areas, the uncertainty of tropospheric columns is dominated by errors in the stratospheric column (for NO₂) and errors in the spectral fitting (Boersma et al., 2004). Boersma et al. (2004) found a relative uncertainty of up to 100 % for NO₂ in these areas. Thus, the satellite measurements over remote oceanic areas are close to the detection limit, making it impossible to compare single satellite measurements to the COPMAR results. Consequently, monthly means are used for the comparison. For spatial collocation of the measurements, the satellite pixels are averaged within a 200 km radius of the midday position of the ship.

3.3 AERONET data

The Maritime Aerosol Network (MAN; Smirnov et al., 2009) is a part of the AErosol RObotic NETwork (AERONET) project (Holben et al., 1998) and provides ship-borne aerosol optical depth (AOD) measurements from Microtops Sun photometers. These are handheld direct Sun measuring devices which are separated into five spectral channels (Smirnov et al., 2009). In the dataset of cruise MSM58/2, AODs at 380, 440, 500, 675, and 870 nm are available. The Ångström exponent is calculated from the channels at 440 and 870 nm (Smirnov et al., 2009). Details of the Microtops handheld Sun photometers and their uncertainties can be found in Ichoku et al. (2002), Morys et al. (2001), Porter et al. (2001), and Knobelspiesse et al. (2003, 2004). In the following, the AOD at 440 nm and the Ångström exponent are used for comparison with our data.

3.4 MOZART-4 CTM

The 4-D fields of HCHO and CHOCHO concentrations are needed as a priori information for the calculation of VCDs for both MAX-DOAS and satellite data (Sect. 3.7). These a priori information have been taken from the Model for OZone and Related chemical Tracers version 4 (MOZART-4; Emmons et al., 2010) model output (available at https://www.acom.ucar.edu/wrf-chem/mozart.shtml, last access: 17 April 2018). The model data have 6-hourly temporal and ∼ 1.9^∘ × 2.5^∘ horizontal resolution. We further used these model data for the interpretation of our results (Sects. 4.2 and 4.3). For the comparison, the model data are linearly interpolated in time on the cruise track.

3.5 FLEXPART

FLEXPART is a Lagrangian dispersion and transport model which can be used to simulate atmospheric transport of air parcels (https://www.flexpart.eu/, last access: 29 September 2018; Stohl et al., 1998, 2005; Stohl and Thomson, 1999). Here, backward-in-time simulations (Seibert and Frank, 2004) are used to determine potential source regions of trace gas enhancements which were observed during 3 d during the cruise. The simulations were driven by ECMWF IFS (version CY41R2) wind fields at a horizontal resolution of 0.2^∘ and a temporal resolution of 1 h. Backward simulations started at altitudes of 20, 500, 1000, 1500, 2000, 2500, 3000, 3500, 4000, 4500, and 5000 m above the ship's location at hourly intervals between ∼ 11:00 and ∼ 15:00 UTC, coinciding with the measurement times. For each model run, 2 million individual air parcels were followed backwards in time for 2–4 d; the simulated air tracer did not undergo any deposition or other loss processes.

3.6 FINN

The Fire INventory from NCAR (National Center for Atmospheric Research; FINN) is a dataset providing daily global emissions of trace gases and particles from biomass burning at 1 km resolution (Wiedinmyer et al., 2011). For this inventory, satellite observations of active fires from MODIS are used. The emissions are calculated by using land cover type, emission factors, and the estimated fuel loadings. In our analysis, we consider fires on the days before and after the day with high FLEXPART emission sensitivity to the surface, in order to account for uncertainties in fire detection.

3.7 DOAS analysis

The spectra measured by the Avantes instrument (see Sect. 3.1) have been analysed using the differential optical absorption spectroscopy (DOAS; Platt and Stutz, 2008; Burrows et al., 2011) technique which is based on the Beer–Lambert law and describes the spectral attenuation of the initial intensity (I₀) of light due to extinction along the light path s:

Here, λ is the wavelength, σ the absorption cross section, I the reduced intensity, and ρ the absorber concentration. This method can be used for ground-based as well as for satellite data, and for wavelengths in the ultraviolet and visible spectral ranges. The main result of the DOAS analysis is the integrated concentration of trace gases ρ(s) along the effective light path s, the so-called slant column density (SCD; Platt and Stutz, 2008; Burrows et al., 2011):

This quantity depends on the light path which is influenced by the SZA, the relative azimuth angle, the viewing direction of the instrument, and the viewing conditions. Therefore, a rough estimation of the absorber altitude is possible, using SCDs from different elevation angles due to the scan angle dependency of the light path. For pollution close to the ground, the highest SCDs are expected for the lowest elevation angle, whereas for absorbers in elevated layers, the highest SCDs are expected at a higher elevation angle. For better interpretation, the SCDs are converted into VCDs, which are the integral of the trace gas concentration from the surface to the top of the atmosphere (TOA) along the altitude z (Platt and Stutz, 2008; Burrows et al., 2011):

For MAX-DOAS measurements, I₀ is usually a zenith sky measurement. Therefore, the derived quantities are differential values (differential SCD, dSCD), which has to be considered in the analysis. The sensitivity of the measurement to an absorber varies with altitude; this is expressed by the so-called box air mass factor (BAMF; Burrows et al., 2011), which is defined as ${\mathrm{BAMF}}_i={\mathrm{SCD}}_i/{\mathrm{VCD}}_i$ for an atmospheric layer i. BAMFs are calculated by radiative transfer models (here SCIATRAN; Rozanov et al., 2014), which take into account the viewing geometry and environmental effects (Platt and Stutz, 2008). The total AMF can be retrieved by using the BAMF and the trace gas concentration in the individual layers:

VCD_i is the vertical distribution of the trace gases taken from the MOZART-4 model. The differential AMF (dAMF) is the difference between the AMF of the individual measurement and the AMF of the zenith sky reference measurement ($\mathrm{dAMF}={\mathrm{AMF}}_{\mathrm{meas}}-{\mathrm{AMF}}_{\mathrm{ref}}$). By applying the dAMF, the dSCD can be converted to VCD:

3.8 DOAS-fit settings

3.8.2 Stratospheric NO₂

For the evaluation of the instrument setup, stratospheric NO₂ is used, which is well known from previous studies such as Peters et al. (2012), Kreher et al. (1995), and Senne et al. (1996). For this NO₂ retrieval, a fitting window from 450 to 490 nm is used. Absorption cross sections for NO₂ (Vandaele et al., 1998) at 220 K and orthogonalised to this cross section at 298 K are used to account for the stratosphere and the troposphere, respectively. Furthermore, an O₃ (Serdyuchenko et al., 2014; Gorshelev et al., 2014) cross section at 223 K with an I₀ correction of 10²⁰ molec cm⁻² (Platt et al., 1997; Richter, 1997) is included in the fit. Additionally, cross sections for O₄ (Thalman and Volkamer, 2013) and H₂O (Rothman et al., 2010), and a cross section to correct the Ring effect (Vountas et al., 1998) are included in the analysis, along with a fifth-order polynomial. Figure 4 shows an example fit for stratospheric NO₂.

Figure 4Example of a stratospheric NO₂ fit for 14 October 2016 at 06:58 UTC (latitude: ∼ 7.4^∘ N; longitude: ∼ 17.8^∘ W). Both are shown, the reference scaled with the SCD as well as the scaled reference plus the residual. Zenith sky measurement with a SCD of 3.99 × 10¹⁶ molec cm⁻² and a rms of 6.9 × 10⁻⁴ at 90.9^∘ SZA.

Download

A fixed reference spectrum measured on 17 October 2016 at 12:20 UTC and 4.03^∘ SZA is used to analyse the data. On this day, the ship was close to the Equator (∼ 9.6^∘ S) with negligible tropospheric NO₂ (known from MAX-DOAS measurements analysed with a sequential zenith sky reference spectrum and satellite observations) and good weather conditions. The retrieved dSCDs for NO₂ are converted into VCDs by using AMF calculated at a wavelength of 470 nm. To analyse stratospheric NO₂, only zenith sky measurements with a SZA smaller than 92^∘ are used.

For the conversion into total VCD, a reference VCD (VCD_ref) is calculated as follows:

$$\begin{array}{}\text{(6)}& \begin{array}{rl}{\mathrm{VCD}}_{\mathrm{ref}}& ={\displaystyle \frac{{\mathrm{VCD}}_{\mathrm{AM}}+{\mathrm{VCD}}_{\mathrm{PM}}}{\mathrm{2}}}\\ & ={\displaystyle \frac{{\mathrm{dSCD}}_{\mathrm{AM}}+{\mathrm{dSCD}}_{\mathrm{PM}}}{\mathrm{2}\times ({\mathrm{AMF}}_{\mathrm{88}{}^{\circ }}-{\mathrm{AMF}}_{\mathrm{ref}})}}\\ & \approx \mathrm{1.97}\times {\mathrm{10}}^{\mathrm{15}}\mathrm{molec}{\mathrm{cm}}^{-\mathrm{2}}\end{array},\end{array}$$

with the assumption, that the reference spectrum is at noon and VCD_ref is the mean value of the morning (VCD_AM) and evening value (VCD_PM), both taken at SZA = 88^∘. The AMFs are calculated by using the radiative transfer model SCIATRAN (Rozanov et al., 2014): ${\text{AMF}}_{\mathrm{ref}}=\text{AMF}\left(\mathrm{10}{}^{\circ }\right)=\mathrm{1.11}$ and $\text{AMF}\left(\mathrm{88}{}^{\circ }\right)=\mathrm{12.99}$. Afterwards, a reference SCD (SCD_ref) can be calculated:

$$\begin{array}{}\text{(7)}& {\mathrm{SCD}}_{\mathrm{ref}}={\mathrm{AMF}}_{\mathrm{ref}}\times {\mathrm{VCD}}_{\mathrm{ref}}\approx \mathrm{2.14}\times {\mathrm{10}}^{\mathrm{15}}\mathrm{molec}{\mathrm{cm}}^{-\mathrm{2}},\end{array}$$

which is added to the dSCD to retrieve total SCD:

$$\begin{array}{}\text{(8)}& \mathrm{SCD}=\mathrm{dSCD}+{\mathrm{SCD}}_{\mathrm{ref}}.\end{array}$$

After the conversion into total VCDs, the latitudinal dependency and the diurnal cycle of stratospheric NO₂ are calculated. The longest light path in the stratosphere, and thus the highest sensitivity to stratospheric NO₂, can be found during twilight. Therefore, measurements between 88^∘ and 92^∘ SZA are used for both morning and evening measurements for the analysis of the latitudinal dependency of stratospheric NO₂. For the calculation of the stratospheric diurnal cycle, half-hour binned values are calculated, because of the higher signal-to-noise ratio compared to the single measurements.

The observed stratospheric NO₂ values are well above the detection limit. The mean rms is 3.5 × 10⁻⁴ and the strongest convoluted differential peak of the NO₂ cross section (1.4 × 10⁻¹⁹) which leads to an estimated detection limit of 4.8 × 10¹⁵ molec cm⁻² for dSCDs (Sect. 3.7). For large SZAs with an AMF of around 15, the detection limit for VCDs is therefore 3.2 × 10¹⁴ molec cm⁻².

3.8.3 Formaldehyde

For the HCHO retrieval, a fitting window from 336.5 to 359 nm is used. Absorption cross sections for HCHO at 297 K (Meller and Moortgat, 2000), O₃ (Serdyuchenko et al., 2014) at 223 K and orthogonalised 243 K, as well as NO₂ at 298 K (Vandaele et al., 1998) are used. For the latter two gases, an I₀ correction (Aliwell et al., 2002) of 10²⁰ and 10¹⁷ molec cm⁻² is applied, respectively. Additional cross sections are O₄ (Thalman and Volkamer, 2013), BrO at 223 K (Fleischmann et al., 2004), as well as a cross section to correct the Ring effect (Vountas et al., 1998) and a polynomial of degree 5. Furthermore, the mean residual of the whole cruise is included as additional cross sections to improve the fit results. By including this additional cross section, the fit rms reduces by approximately 50 % (not shown). Figure 5 shows an example fit for HCHO.

Figure 5Example of a HCHO fit for 13 October 2016 at 14:38 UTC (latitude: ∼ 10.1^∘ N; longitude: ∼ 19.8^∘ W). The SCD is 2.52 × 10¹⁶ molec cm⁻² with a rms of 2.2 × 10⁻⁴ at 29.46^∘ SZA and a viewing angle of 15^∘.

Download

As for the case of stratospheric NO₂, also for HCHO a fixed reference spectrum is used. The reference spectrum is from 23 October 2016 (∼ 29.2^∘ S) around noon. This date was chosen because of low HCHO content in the overhead column (expected from satellite observations), whereas closer to the Equator, higher HCHO values would be expected. For the conversion to VCDs, dAMF were calculated at the wavelength 338 nm as shown in Eq. (5).

Our measurements show HCHO dSCDs consistently above the detection limit. The measurements have a mean rms of 2.9 × 10⁻⁴, which leads to a detection limit of ∼ 2.2 × 10¹⁵ molec cm⁻² for daily mean dSCDs, depending on the number of measurements per day. Applying a dAMF of 1.1 (representative of 30^∘ elevation angle; see also Sect. 3.7), this translates to a detection limit of ∼ 1.9 × 10¹⁵ molec cm⁻² in the VCDs.

For the analysis, only measurements at SZA < 70^∘ are used, because for larger SZA the measurement uncertainty increases. HCHO has a global background due to the oxidation of CH₄ (Sect. 1). Therefore, the overhead column from the reference measurement needs to be considered for the conversion into total VCDs, which is done by adding the simulated column (VCD_Model = 3.7 × 10¹⁵ molec cm⁻²) from the MOZART-4 data at the location of the reference measurement to the VCD:

$$\begin{array}{}\text{(9)}& {\mathrm{VCD}}_{\mathrm{total}}=\mathrm{VCD}+{\mathrm{VCD}}_{\mathrm{Model}}.\end{array}$$

3.8.4 Glyoxal

The CHOCHO fitting window is from 433 to 460 nm with a polynomial degree of 4. Absorption cross sections are CHOCHO (Volkamer et al., 2005) and O₃ (Bogumil et al., 2003) at 223 K with an I₀ correction of 10²⁰ molec cm⁻². Furthermore, NO₂ cross sections (Vandaele et al., 1998) at 298 K with I₀ correction of 10¹⁷ molec cm⁻² and at 220 K orthogonalised to 298 K, an O₄ cross section (Thalman and Volkamer, 2013) at 293 K, a H₂O cross section (Rothman et al., 2010) at 296 K with an I₀ correction of 10²⁴ molec cm⁻², as well as a cross section to correct for the Ring effect (Vountas et al., 1998) are included in the fit. The fit residual shows structures which are related to strong water vapour absorption over the ocean and possible missing peaks in the water cross section (not shown).

To improve the fit residual, two additional cross sections were tested: an alternative H₂O cross section (Polyansky et al., 2018) and a mean residual as an additional cross section. The alternative H₂O cross section did not reduce the fit residual, and therefore it is not used here. The mean residual was calculated for all measurements during the cruise and the fit was repeated including this mean residual as additional pseudo-absorber. This additional cross section clearly improved the fit residual, and therefore this pseudo-absorber is used in the following. Figure 6 shows an example fit for CHOCHO.

Figure 6Example of a CHOCHO fit for 13 October 2016 at 11:33 UTC (latitude: ∼ 10.8^∘ N; longitude: ∼ 20.0^∘ W). The SCD is 6.17 × 10¹⁴ molec cm⁻² with a rms of 1.3 × 10⁻⁴ at 29.75^∘ SZA and a viewing angle of 7^∘.

Download

As for HCHO, a reference spectrum from 23 October 2016 around noon was chosen. Also for CHOCHO, low overhead columns are expected in this area. Since background concentrations of CHOCHO are not expected, no global background value has to be added. For the conversion into VCDs, AMFs were calculated at the wavelength 433 nm (see Eq. 5), and only measurements at SZAs < 70^∘ are used in order to limit noise occurring at low intensities.

Our analysis of the MAX-DOAS measurements yields CHOCHO dSCDs which are often below the detection limit. The rms is on the order of 1.6 × 10⁻⁴, resulting in a detection limit of ∼ 1.7 × 10¹⁴ molec cm⁻² for daily averages of the dSCDs, depending on the number of measurements (Fig. 11). Applying a dAMF of 1.2 (representative of 30^∘ elevation angle; see also Sect. 3.7), this translates to a detection limit of ∼ 1.4 × 10¹⁴ molec cm⁻² in the VCDs.

4.1 Evaluation of stratospheric NO₂

To demonstrate the performance of the instrument setup, we analysed the diurnal and latitudinal variation of stratospheric NO₂ and compared our results with satellite data and previous studies. The diurnal cycle of stratospheric NO₂ is closely related to the photolysis of the reservoir species N₂O₅ (Solomon et al., 1986). During daylight, NO₂ increases as result of N₂O₅ photolysis, whereas at night, N₂O₅ increases and NO₂ is removed. Thus, it has a minimum in the morning, followed by a linear increase. As shown by Peters et al. (2012), this diurnal cycle can be observed over remote oceanic areas, where the stratospheric signal is usually not impaired by the presence of tropospheric NO₂. On cruise MSM58/2, such a diurnal cycle could also be observed, which is exemplarily shown for 15 October 2016 in Fig. 7. On that day, the linear increase during daytime amounted to 7.31 × 10¹³ molec cm⁻² h⁻¹ (∼ 2.5^∘ N, ∼ 13.9^∘ W). Previous studies found similar results, e.g. Peters et al. (2012), with an increase of 8.7 × 10¹³ molec cm⁻² h⁻¹ for the tropics and Gil et al. (2008) with an increase of 6 × 10¹³ molec cm⁻² h⁻¹ for the subtropics.

Figure 7The diurnal cycle of stratospheric NO₂ (15 October 2016; latitude: ∼ 2.5^∘ N; longitude: ∼ 13.9^∘ W). In red, the data are binned to 0.5 h values to improve the signal-to-noise ratio. Also shown is the regression line which represents a linear increase during the day. The data are shown for SZAs smaller than 92^∘. Furthermore, the errors of the slope (m_err) and of the intercept (b_err) are presented.

Download

The diurnal cycle of stratospheric NO₂ depends on the latitude, with a smaller increase in the tropics and a stronger increase towards the midlatitudes, as shown in Fig. 8. The presented morning and evening MAX-DOAS values (averaged over all measurements with SZAs between 88 and 92^∘) approximately correspond to the first and last red dots in Fig. 7, respectively. They both show a local minimum near the Equator and increase towards the midlatitudes.

The morning MAX-DOAS values range from 2.0 × 10¹⁵ molec cm⁻² at 32.8^∘ N to 1.2 × 10¹⁵ molec cm⁻² close to the Equator and increase again in the Southern Hemisphere to 2.3 × 10¹⁵ molec cm⁻² at 33.7^∘ S. On the other hand, the evening MAX-DOAS values are higher (3.4 × 10¹⁵ molec cm⁻² at 35.4^∘ N to 2.0 × 10¹⁵ molec cm⁻² close to the Equator and 3.5 × 10¹⁵ molec cm⁻² at 32.8^∘ S). This “U” shape can also be observed in the satellite measurements. Given the 09:30 local time (which corresponds to ∼ 11:00 UTC) overpass time of both GOME-2 instruments, one would expect that the satellite values are close to the morning MAX-DOAS values. Taking measurement uncertainties into account, this behaviour can indeed be found for the GOME-2B data. However, the GOME-2A measurements are slightly higher than the observed morning MAX-DOAS values, which is probably related to the degradation of the instrument (Dikty and Richter, 2011). The uncertainties for the retrieved MAX-DOAS measurements are calculated by error propagation using the fitting error and the assumed AMF uncertainty of ±1. The uncertainty in VCD_ref has been neglected (an assumed uncertainty of 30 % in VCD_ref amounts to a value of 0.59 × 10¹⁵ molec cm⁻², which corresponds to an uncertainty of 0.65 × 10¹⁵ molec cm⁻² for the SCD_ref, resulting in a negligibly small uncertainty of 0.05 × 10¹⁵ molec cm⁻² for twilight measurements).

Figure 8The latitudinal dependency (from north to south) of stratospheric NO₂ over the Atlantic Ocean shows a local minimum near the Equator and an increase towards the midlatitudes. MAX-DOAS morning and evening values have been averaged over all zenith sky measurements for SZAs between 88 and 92^∘. The colour-shaded areas indicate the errors of the measurements.

Download

Our results for stratospheric NO₂ are in good agreement with previous studies from Kreher et al. (1995) and Senne et al. (1996), especially in the Northern Hemisphere and across the Equator. In the Southern Hemisphere (at 30^∘ S), however, the values reported by Kreher et al. (1995) and Senne et al. (1996) are significantly higher than our measurements presented here, with values of approximately 3 × 10¹⁵ molec cm⁻² (morning) and 5 × 10¹⁵ molec cm⁻² (evening). This could be related to either intra-annual (our measurements were taken a few weeks earlier in the year compared to the two previous studies) or interannual variability. Kreher et al. (1995) and Senne et al. (1996) analysed measurements from 1990 and 1993, respectively. These measurements differ by approximately 1 × 10¹⁵ molec cm⁻² at 30^∘ N. Furthermore, Kreher et al. (1995) showed that the values differ for higher latitudes between the individual years. In the Southern Hemisphere, we observe a stronger increase towards the midlatitudes compared to the Northern Hemisphere, consistent with the findings by Kreher et al. (1995) and, for the Pacific Ocean, by Peters et al. (2012).

Overall, the instrument performed well as the diurnal cycle and the latitudinal dependency of stratospheric NO₂ are clearly visible in our measurements and our results agree with previous studies as well as satellite data. Therefore, we are confident in the suitability of our measurements to analyse the weak absorbers (HCHO and CHOCHO).

4.2 Formaldehyde

The latitudinal variation of daily mean MAX-DOAS HCHO dSCDs observed during the cruise shows enhanced values in an elevated layer at both ∼ 10^∘ N and ∼ 5^∘ S. This coincides with the area of expected outflow from the African continent as seen in satellite measurements (Fig. 1a). Typically, the HCHO concentration in polluted areas is expected to be highest close to the surface (De Smedt et al., 2008; Heckel et al., 2005) due to the primary emission sources of HCHO and rapid photochemical production from VOC precursors. Thus, also the highest dSCDs should be observed at low elevation angles near the sources. Figure 9 illustrates the daily mean HCHO dSCDs during the cruise at different elevation angles. During the cruise, whenever low HCHO columns (∼ 0.5 × 10¹⁶ molec cm⁻² for zenith sky measurements) are observed, the dSCDs are indeed highest at the lowest elevation angles.

Around 10^∘ N and 5^∘ S (on 13, 14, and 17 October, respectively), we observe enhanced HCHO values (> 1 × 10¹⁶ molec cm⁻² for the off-axis measurements). Here, the dSCDs are largest at higher elevation angles, which indicates that HCHO is predominantly located in an elevated layer. Still, the zenith sky measurements show the lowest values, but the other elevation angles are ordered differently than usual, with highest HCHO dSCD at elevation angles between 8 and 15^∘. This pattern is not only visible in the daily mean as depicted in Fig. 9 but can also be observed for the individual vertical scans (not shown here).

Figure 9Daily mean HCHO dSCDs over the Atlantic Ocean over the course of the cruise (from north to south). Enhanced values in the areas of expected outflow show a different scan angle dependency.

Download

When converting the measured HCHO dSCD to VCDs (using the 30^∘ elevation angle and HCHO profiles from the MOZART-4 model), the values on 13, 14, and 17 October remain clearly enhanced (see Fig. 10a). At least the first region sampled on 13/14 October (∼ 10^∘ N) indeed coincides with the area for which satellite observations regularly show enhanced HCHO values (compare Figs. 1a and 2). On these days, also MOZART-4 data are enhanced in the region. However, differences between the datasets are visible. On 13 October, MAX-DOAS and MOZART-4 data are higher than the satellite data. While on 14 October, MAX-DOAS and satellite data are slightly smaller than on 13 October, MOZART-4 data are further increased. On 17 October (5^∘ S), there is no clear sign of a HCHO enhancement in the satellite and model data in contrast to MAX-DOAS measurements. These differences could be related to several reasons. The MAX-DOAS data are measured on single days, whereas the satellite datasets are monthly means and MOZART-4 time series are interpolated on the cruise track. Thus, the differences between MAX-DOAS and satellite data could be explained by single, isolated outflow events on specific days, which are not distinguishable from background values in the monthly averaged satellite data. The interpolation of the model data could lead to differences between model and MAX-DOAS data as, for example, isolated events might not be represented in the model data. Furthermore, the satellite and model data are averages over larger areas which would dilute the magnitude of the measured or simulated values. Thus, comparing HCHO columns retrieved from OMI and GOME-2B radiances (Sect. 3.2) and integrated columns from simulated MOZART-4 profiles (Sect. 3.4; interpolated on the cruise track) to our MAX-DOAS VCDs, the results generally confirm the finding of enhanced HCHO satellite and model columns. The datasets show good agreements with correlation coefficients larger than 0.72 (Table 2).

Table 2Correlation, slope, and intercept between HCHO MAX-DOAS and satellite measurements as well as model simulations. The values in brackets are the standard errors for slope and intercept.

Download Print Version | Download XLSX

The MAX-DOAS measurements are mostly higher than the satellite observations (Fig. 10a), which results in a slope larger than 1 (1.91, OMI and 1.24, GOME-2B) and a large offset for GOME-2B (3.05 × 10¹⁵ molec cm⁻², GOME-2B and 0.01 × 10¹⁵ molec cm⁻², OMI) of the regression line (Table 2). These differences are clearly visible between 20 and 32^∘ N (9–11 October) and between 10 and 22^∘ S (18–21 October; Fig. 10a), which have been measured in clean remote ocean areas with low pollution.

Compared to the MOZART-4 model values, the MAX-DOAS observations are often higher, which can also be observed in a low slope of 0.68. Close to the Equator, the values show good agreement in the area of expected pollution outflow which leads to a high correlation coefficient of 0.72 between the two datasets. However, the offset is also high between the two datasets (4.82 × 10¹⁵ molec cm⁻²).

Figure 10Daily mean HCHO VCDs over the Atlantic Ocean over the course of the cruise (from north to south). (a) Enhanced HCHO values can be observed in the area of expected outflow. Satellite observations and model values are also enhanced in this area. (b) The AOD and Ångström exponent from AERONET measurements are partly enhanced in these areas, depending on the source region.

Download

Several reasons can contribute to an enhancement in MAX-DOAS data. Large differences between the MAX-DOAS and the satellite or model data are found over regions with low air pollution north and south of the Equator. In these areas, high measurement uncertainties can be found in the satellite data due to the low columns which might influence the retrieved satellite values. It is also possible that the model and satellite results underestimate the VCDs, because of the potentially localised nature of the enhancements (see also Sect. 5). Increased uncertainties in the MAX-DOAS data in this region can be ruled out, as the DOAS fit rms is nearly constant during the whole cruise (see Sect. 3.8.3). Furthermore, H₂O interferences might contribute to the differences. In that case, the differences should be reduced/increased for higher/lower elevation angles, respectively. However, this can also be ruled out as a similar behaviour is visible for all viewing directions. Additionally, bad weather conditions can be ruled out, because the affected days of the MAX-DOAS measurements had different weather conditions (see Table 1), and an intensity filter was used to exclude poor viewing conditions (see Sect. 3.1). Similar conditions were also used for satellite values. Here, only measurements with geometric cloud fraction smaller 0.3 were included in the analysis.

The AMF could introduce the differences between the datasets. However, this seems unlikely as the MOZART-4 model is used for the AMF calculations for both MAX-DOAS and satellite measurements, and this model does not show a similar behaviour to the MAX-DOAS measurements. For example, if the model underestimates/overestimates the total amount of HCHO in the atmosphere, this would in good approximation not change AMFs, and therefore the VCDs of both datasets are not influenced. Thus, no difference would be introduced between the satellite and MAX-DOAS data. Possibly, the differences between the datasets can be related to the HCHO model profile if it would differ from the real atmospheric profile. On the one hand, the model could put the HCHO enhancement at the wrong altitude. This could lead to both an underestimation or an overestimation of the MAX-DOAS VCDs, depending on the profile and the SZA or relative azimuth angle, similar for satellite measurements. On the other hand, the model could miss an additional, elevated layer of transported HCHO or an additional layer close to the surface. During those days, the HCHO in the model is located at ∼ 3 km altitude, where the MAX-DOAS sensitivity is highest. Therefore, an additional layer of HCHO would reduce the MAX-DOAS AMF, leading to a higher MAX-DOAS VCD; if located close to the surface, also the satellite VCD would increase. In this case, the difference between MAX-DOAS and satellite VCD would be reduced due to the stronger altitude dependence of the satellite BAMF compared to the MAX-DOAS BAMF. If the additional layer were located above the model elevated layer, the satellite AMF would increase due to the better visibility, leading to decreased satellite VCDs, further increasing the differences between satellite and MAX-DOAS VCDs. Additionally, aerosols could influence the AMFs leading to differences in satellite and MAX-DOAS VCD observations.

Table 3Correlation, slope, and intercept between CHOCHO MAX-DOAS and satellite measurements as well as model simulations. The values in brackets are the standard errors for slope and intercept.

Download Print Version | Download XLSX

4.3 Glyoxal

The observed CHOCHO dSCDs are mostly higher at low elevation angles, again indicating that the trace gas is located close to the ground (Fig. 11). The measured dSCDs are mostly smaller than 0.2 × 10¹⁵ molec cm⁻², fluctuating around zero. Thus, the measurements are below the calculated detection limit (see Sect. 3.7). Nevertheless, enhanced CHOCHO dSCDs were observed on 13 and 14 October (around 10^∘ N) as for HCHO, with a maximum value of about ∼ 0.30 × 10¹⁵ molec cm⁻² for 15 and 6^∘ elevation angle measurements (see Fig. 11). On these two days, the measurements are above the detection limit and the scan angle dependency is slightly different, which indicates that the observed CHOCHO is located in an elevated layer, albeit not as clearly as for HCHO. On some days, the measurements for lower elevation angles are above the detection limit (17  and 21 October). Here, CHOCHO might be located closer to the ground.

Figure 11The latitudinal dependency (from north to south) of daily mean CHOCHO dSCDs over the Atlantic Ocean. On 13 and 14 October, a different scan angle dependency is shown.

Download

When converting the measured CHOCHO dSCDs to VCDs (again using the 30^∘ elevation angle and CHOCHO profiles from the MOZART-4 model), the values on 13 and 14 October remain enhanced (see Fig. 12). This region indeed coincides with the area for which satellite observations regularly show enhanced CHOCHO columns (see Fig. 1b). Comparing columns retrieved from OMI radiances and integrated from simulated MOZART-4 profiles (interpolated on the cruise track) to our MAX-DOAS VCDs, this generally confirms the finding of enhanced CHOCHO satellite and model columns. CHOCHO VCDs from all three datasets show enhanced values around 10^∘ N, while MAX-DOAS and MOZART-4 values further north and further south are close to zero. In comparison, OMI observations show enhanced CHOCHO columns throughout the tropics in Fig. 12. This behaviour could be explained by an elevated CHOCHO layer which is not represented in the model and cannot be detected by MAX-DOAS measurements due to their low sensitivity in the free troposphere. The enhanced values throughout the tropics are also represented in the slope of the regression line with 0.32; nevertheless, only a negligible offset of −0.05 × 10¹⁵ molec cm⁻² was found (Table 3). The correlation between MAX-DOAS and OMI CHOCHO VCDs is 0.56. The modelled MOZART-4 data are mostly close to or slightly higher than the observed MAX-DOAS values; in particular, the observed CHOCHO enhancements are less pronounced in our measurements than in the model data, resulting in a slope of the regression line of 1.06 with a high standard error of 0.29, whereas the offset is small (−0.06 × 10¹⁵ molec cm⁻²). Overall, the MOZART-4 and MAX-DOAS CHOCHO data have a correlation coefficient of 0.55.

Figure 12The latitudinal dependency (from north to south) of daily mean CHOCHO VCDs over the Atlantic Ocean. (a) Enhanced CHOCHO values can be observed in the area of expected outflow. As for HCHO, satellite observations and model values are enhanced in this area. (b) AOD from AERONET dataset is enhanced in this area. For this plot, a similar axis as for Fig. 10 is used.

Download

4.4 Aerosol

Around 10^∘ N (13 October) the AERONET AOD (0.40) and the MAX-DOAS HCHO and CHOCHO are clearly enhanced in an elevated layer for the traces gases (Figs. 10b and 12b). Similar results for MAX-DOAS measurements were found at 6^∘ N (14 October), where no AOD observations are available. At approximately 20^∘ N (11 October), the AERONET AOD shows also slightly enhanced values (0.33). For the MAX-DOAS measurements, the values are slightly increased for HCHO with the same scan angle dependency as expected for pollution close to the ground. Here, it is important to remember that for CHOCHO the values are below/close to the detection limit. On both days (11 and 13 October), the Ångström exponent is on the order of 0.3. The classification of Toledano et al. (2007) shows that in this area the main aerosol type is desert dust, which agrees with the findings of Takemura et al. (2000), Ridley et al. (2012), Ichoku et al. (2004), and Schepanski et al. (2017).

At approximately 5^∘ S (17 October), the relation between MAX-DOAS measurements, AOD, and Ångström exponent is slightly different. The AOD is 0.20, with a high Ångström exponent of 1.41 indicating a different source than in the Northern Hemisphere. Here, the main aerosol type is on the edge between marine and continental aerosol, after the classification of Toledano et al. (2007). Because of the enhanced HCHO in an elevated layer, it might originate from continental sources.

Figure 13HCHO MOZART-4 profiles as used for VCDs calculations, interpolated on the cruise track (from north to south). The blue triangles on the bottom indicate the position of RV Maria S. Merian on the days with unusual scan angle dependency. For the calculation of the altitude, a mean temperature profile is used.

Download

At approximately 18^∘ S (20 October), the AOD is increased with a value of 0.26, and the Ångström exponent has a value of 0.84. However, HCHO and CHOCHO show no enhancement. On this day, a container vessel was sailing in front of RV Maria S. Merian and increased tropospheric NO₂ columns were observed with a clear scan angle dependency expected for pollution close to the surface (not shown). The main aerosol type classified after Toledano et al. (2007) is desert dust. The reason for the AOD enhancement might be that the air mass is strongly mixed with marine aerosols which were visible on the first days and the last days of the cruise. Thus, a mixture of exhaust and marine aerosols is observed, potentially leading to the desert dust classification.

4.5 Model simulations

MOZART-4 model simulations (Emmons et al., 2010) show elevated layers of enhanced HCHO and CHOCHO concentrations between ∼ 3000 and ∼ 6000 m (Figs. 13 and 14, respectively), on 13, 14, and 17 October. On these days, the scan angle dependency of the measured dSCDs suggests the presence of an elevated layer of HCHO and CHOCHO and of only HCHO, respectively. However, the model shows generally between 20^∘ N and approximately 30^∘ S increased values in an elevated layer, which cannot be confirmed by our measurements. On 13 and 14 October, the model additionally shows enhanced VOC concentrations between the surface and approximately 2000 m altitude. This lower layer of high VOC concentrations could explain the overall higher dSCDs compared to the other days, probably indicating that the enhanced dSCDs measured on that day are only partly located in an elevated layer.

Figure 14CHOCHO MOZART-4 profiles as used for VCD calculations, interpolated on the cruise track (from north to south). The blue triangles on the bottom indicate the position of RV Maria S. Merian on the days with unusual scan angle dependency. For the calculation of the altitude, a mean temperature profile is used.

Download

Figure 15The emission sensitivity response function to the lowest 500 m layer for air parcels arriving at the receptor position in different altitudes above the RV Maria S. Merian on 13 October 2016 at midday, indicated by red dots (compare Figs. 9 and 11). The emission sensitivities were integrated over 2 d backwards. The black circles are fires which were detected between 10 and 12 October taken from the FINN database (Sect. 3.6). The circles are scaled with the calculated CO₂ emissions.

To further complement the interpretation of our results, we conducted backward simulations with the FLEXPART dispersion model (Stohl et al., 1998, 2005; Stohl and Thomson, 1999), following air masses from the position of the ship towards potential sources which are assumed to be in the lowest 500 or 1000 m. For the VOC enhancements observed on 13 October, the simulated emission sensitivities indicate air masses originating from the Sahel and the more southerly forest, shrub-, grass-, and croplands (Mayaux et al., 2004; Fig. 15) were measured at the vessel. These areas are potential source regions for emissions of biogenic precursors of HCHO and CHOCHO. Furthermore, fires from the FINN dataset are shown in Fig. 15. Some of these fires were detected in the possible source regions which can also be a source for VOCs. The simulated particles reach the measurement location over the open ocean at altitudes above 1500 up to 3000 m after 2 d (Fig. 15) or longer (not shown), depending on the source region. Thus, the measured VOC enhancements are probably caused by export of precursor molecules from the African continent due to the short lifetime of HCHO and CHOCHO.

The results are different for 14 October when the measurements also show enhanced HCHO and CHOCHO values. The emission sensitivity response function is generally low for the lowest 1000 m (Fig. 16; similar results were also found for the lowest 500 m, not shown). Only air older than 2 d (not shown) might originate from the continent, but also 4 d old air has only a small sensitivity to the continent (Fig. 16). The highest emission sensitivity to the lowest 1000 m is found above the open ocean south of the vessel's position. Furthermore, fires are unlikely to be important for the observed enhanced HCHO and CHOCHO values, as only a few small fires were detected in the potential continental source regions. Additionally, the source altitude is less clear due to similar emission sensitivities for different altitudes. The detected HCHO and CHOCHO might be located at an altitude between 1000 and 5000 m.

Figure 16The emission sensitivity response function to the lowest 1000 m layer for air parcels arriving at the receptor position in different altitudes above the RV Maria S. Merian on 14 October 2016 at midday, indicated by red dots (compare Figs. 9 and 11). The emission sensitivities were integrated over 4 d backwards. The black circles are fires which were detected between 8 and 13 October taken from the FINN database (Sect. 3.6). The circles are scaled with the calculated CO₂ emissions. Only certain altitudes are shown as altitudes of 2000, 2500, 3500, and 4000 m have a similar emission sensitivity pattern to altitudes of 1500 and 3000 m.

Download

For the HCHO enhancements observed on 17 October, the simulated emission sensitivities indicate air masses originating from southerly central Africa (Angola, Democratic Republic of the Congo, Republic of the Congo, Gabon, and Republic of Equatorial Guinea; Fig. 17). The simulated air masses from the African continent reach the positions above the ship after 4 d at an altitude between 2000 and 4500 m, consistent with MOZART-4 model simulations (Fig. 13). In the possible source regions, mostly different types of forests and grasslands were found (Mayaux et al., 2004), as well as fires with large emissions (Fig. 17). Both can emit precursors of HCHO. However, most fires are detected south of the possible source regions.

Figure 17The emission sensitivity response function to the lowest 500 m layer for air parcels arriving at the receptor position in different altitudes above the RV Maria S. Merian on 17 October 2016 at midday, indicated by red dots (compare Fig. 9). The emission sensitivities were integrated over 4 d backwards. The black circles are fires which were detected between 12 and 14 October taken from the FINN database (Sect. 3.6). The circles are scaled with the calculated CO₂ emissions.