Shipping has always been an important mode of transportation throughout the course of history. In contrast to the past, today ships almost exclusively carry freight, with the exception of a small number of cruise ships and ferries. Globalization of markets has lead to an enormous increase in world trade and shipping traffic in the last decades, with growth rates being typically about twice those of the world gross domestic product (GDP) .

Shipping is generally the most energy-efficient transportation mode, with the lowest greenhouse gas emissions per tonne per kilometer (3–60 gCO2t-1km-1), followed by rail (10–120 gCO2t-1km-1), road (80–180 gCO2t-1km-1) and air transport (435–1800 gCO2t-1km-1), which is by far the least efficient . At the same time, with a volume of 9.84 billion tons in 2014, shipping accounts for four-fifths of the worldwide total merchandise trade volume , as compared to, for example, the total air cargo transport volume of 51.3 million tons in 2014 . As a result, shipping accounts for a significant part of the emissions from the transportation sector .

Despite growth rates now being lower compared to those prior to the 2008 economic crisis, seaborne trade is growing faster than the rest of the transportation sector, with an annual growth rate of 3–4 % in the years 2010 to 2014, compared to 2.0–2.6 % for the global merchandise volume . The number of ships larger than 100 gross tonnes increased from around 31 000 in 1950 to over 52 000 in 1970 to 89 000 in 2001 and is estimated to increase to about 150 000 in 2050 . At the same time, total fuel consumption and emissions increased as well . predicted that future development of shipping emissions will depend more on the usage of new technologies and imposed regulations than on the economic growth rates.

The most important pollutants emitted by ships are carbon dioxide (CO2), carbon monoxide (CO), nitrogen oxides (NOx = NO + NO2), sulfur dioxide (SO2), black carbon (BC), volatile organic compounds (VOCs) and particulate matter (PM) . This study focuses on NO2 and SO2 because both are emitted in considerable amounts and both absorb light in the UV–visible spectral range and therefore can readily be measured by differential optical absorption spectroscopy (DOAS), which is explained in Sect. . In 2001, shipping emissions accounted for 15 % of all anthropogenic NOx and 8 % of all anthropogenic SO2 emissions .

NOx is predominantly formed thermally from atmospheric molecular nitrogen (N2) and oxygen (O2) during high-temperature combustion processes in ship engines in an endothermic chain reaction called the Zeldovich mechanism. The emitted NOx comprises mainly NO, with less than 25 % of NOx being emitted as NO2 . measured emission factors for gaseous and particulate pollutants onboard three Chinese vessels and found that more than 80 % of the NOx was emitted as NO and that emission factors were significantly different during different operation modes.

In the ambient atmosphere, NO is rapidly converted to NO2 by reaction with ozone (O3), leading to a lifetime of only a few minutes. During the daytime NO2 is photolyzed by UV radiation (λ<420nm), releasing NO and ground-state oxygen radicals (O(3P)). In a three-body-collision reaction involving N2 or O2 the oxygen radical reacts with an oxygen molecule to reform ozone . When daylight is available, these reactions form a “null-cycle” and the transformation between NO and NO2 is very fast, leading to a dynamic equilibrium. This is also known as the Leighton photostationary state. Owing to the lack of photolysis, NO reacts rapidly with O3 to form NO2 during the night. In addition, the nitrate radical (NO3) is formed by reaction of NO2 with O3. An equilibrium of NO2 with NO3, forming N2O5, the acid anhydride of nitric acid HNO3, results .

During the day OH reacts with NO2 in a three-body reaction to form HNO3. An important sink for NO2 in the troposphere is wet deposition of the resulting HNO3. The mean tropospheric lifetime of NOx varies between a few hours in summer and a few days in winter , depending on altitude. Inside ship plumes, found a substantially reduced lifetime of NOx of about 1.8 h compared to approximately 6.5 h in the background marine boundary layer (around noon). This is attributed to enhanced levels of OH radicals in the plume.

Unlike for NOx, ship emissions of SO2 are directly linked to the fuel sulfur content. Around 86 % of the fuel sulfur content is emitted as SO2 . found a linear relationship between SO2 and sulfate particle emission and that only around 4.8 % of the total sulfur content is either directly emitted as or immediately transformed into particles after the emission. An important sink for SO2 is wet deposition after oxidation by OH radicals to the extremely hygroscopic sulfur trioxide (SO3) reacting rapidly with liquid water to form sulfuric acid (H2SO4) . Another important sink is dry deposition, leading to a lifetime of approximately 1 day in the boundary layer, which can be even shorter in the presence of clouds .

Sulfate aerosols influence climate directly by scattering and absorbing solar radiation and indirectly by increasing cloud condensation and changing cloud reflectivity and lifetime . In the presence of VOCs, nitrogen oxides are important precursors for the formation of tropospheric ozone and therefore photochemical smog. The release of both NO2 and SO2 leads to an increase in acidification of 3–10 % in coastal regions, contributing significantly to acid rain formation that damages ecosystems . The deposition of reactive nitrogen compounds causes eutrophication of ecosystems and decreases biodiversity .

Around 70 % of shipping emissions occur within 400 km of land , contributing substantially to air pollution in coastal areas . Ship emissions were found to provide a dominant source of air pollution in harbor cities . In addition, transport of tropospheric ozone and aerosol precursors over several hundreds of kilometers also affects air quality, human health and vegetation further inland, far away from their emission point .

NO2 and SO2 can cause a variety of respiratory problems. Tropospheric ozone is harmful to animals and plants, causing various health problems. The European Union (EU) legislation for O3 exposure to humans has set a target limit of 120 µgm-3 (∼ 60 ppbv) for maximum daily 8 h mean but allows exceedances on 25 days averaged over 3 years . As mentioned above, both NO2 and SO2 play a role in the formation of particles. Fine particles are associated with various health impacts like lung cancer, heart attacks, asthma and allergies .

International ship traffic is subject to regulations of the International Maritime Organization (IMO). Shipping emissions are regulated by the International Convention for the Prevention of Pollution from Ships (MARPOL 73/78) Annex VI . This annex was added in 1997 and was enforced in 2005. A revision with more stringent emission limits was adopted in 2008 and enforced 2010. With this, limits on sulfur content in heavy fuel oils (HFOs) were globally set and local Sulfur Emission Control Areas (SECAs), later revised to general Emission Control Areas (ECAs), along the North American coast and in the Baltic and North seas (including the English Channel) were established with more stringent restrictions and controls. MARPOL introduced a global fuel sulfur limit of 4.5 %, which was reduced to 3.5 % in 2012 and will be further reduced in 2020 (or 2025 depending on a review in 2018) to 0.5 %. In the established ECAs, from 2010 on, the limit was set to 1.5 % and was further reduced in 2010 to 1.0 %. Carrying out airborne in situ measurements in several flight campaigns in the English Channel and North and Baltic seas, measured an 85 % compliance in 2011 and 2012 with the 1 % fuel sulfur limit. In the Gulf of Finland and Neva Bay area, found a 90 % compliance in 2011 and 97 % compliance in 2012 with the 1 % fuel sulfur limit from ground-based, ship-based and helicopter-based in situ measurements.

Recently, from 1 January 2015 on, the allowed fuel sulfur content in SECAs was further reduced to 0.1 %. Using in situ measurements in Wedel at the bank of the river Elbe, a few kilometers downstream from Hamburg, Germany, showed that in late 2014 more than 99 % of the measured ships complied with the 1 % sulfur limit, and in early 2015 95.4 % of the measured ships complied with the new 0.1 % sulfur limit. By analyzing 1.5 years of SO2 measurements at the English Channel, found a 3-fold reduction in SO2 from 2014 to 2015. They estimated the lifetime of SO2 in the marine boundary layer to be around half a day. measured a substantial drop in SO2 emissions by 91 % when the investigated container ships entered the Californian ECA and switched from HFO with 3.15 % fuel sulfur content to marine gas oil with 0.07 % fuel sulfur content. These estimates were obtained performing airborne in situ measurements.

MARPOL Annex VI also establishes limits dependent on engine power for the emission of NOx from engines built after 2000 (Tier I), 2011 (Tier II) and 2016 (Tier III), but due to the slow penetration of the full shipping fleet, the impact on NOx emissions is not yet clear. Since 2010, a NOx Emission Control Area (NECA) exists around the North American coast and in the Caribbean, while for the North and Baltic seas the establishment of such a NECA is planned and was recently agreed on, but the future enforcement date is still unclear. The EU also established a sulfur content limit of 0.1 % for inland waterway vessels and ships at berth in community ports, which has been in force since 1 January 2010 .

The impact of shipping emissions on the North Sea for different regulation scenarios was investigated in a model study by the Helmholtz-Zentrum Geesthacht (HZG) within the scope of the Clean North Sea Shipping project. For current emissions, a relative contribution of shipping emissions to air pollution in coastal regions of up to 25 % in summer and 15 % in winter for NO2 and 30 % in summer and 12 % in winter for SO2 was found . For the year 2030, the contribution of the continuously growing shipping sector to the NO2 concentrations is predicted to decrease. The extent of reduction depends on the date on which the stricter Tier III regulations will enter into force and on the fraction of the fleet complying with these regulations (i. e. the age of the fleet), with up to 80 % reduction if all ships comply (in the improbable case of a new-ship-only fleet). For SO2, the established fuel sulfur content limit of 0.1 % (ECA) and 0.5 % (globally) will lead to significant reductions, a further decrease is expected if the fraction of ships powered by liquid natural gas grows .

Optical remote sensing using the DOAS technique to measure shipping emissions has been conducted before. For example, performed airborne (from airplane and helicopter) DOAS measurements of NO2 and SO2 in ship plumes by measuring sea-scattered light. and measured flow-rate emissions (mass per second) of NO2 and SO2 for single ships with ground-based multi-axis DOAS (MAX-DOAS) measurements across the Giudecca Channel in the Venice lagoon. measured nocturnal NO2-to-SO2 ratios in ship plumes in the Strait of Georgia with the active long-path DOAS technique. tested and compared optical remote sensing methods (DOAS, lidar, UV camera) and in situ (sniffer) methods for the measurement of shipping emissions in the framework of the SIRENAS-R campaign in the harbor of Rotterdam in 2009. showed that a UV (SO2) imaging camera can be used to measure SO2 in ship plumes at the Kongsfjord at Ny Ålesund, Svalbard, and the harbor of Rotterdam.

The global pathways of the ships can be seen in long time-averaged NO2 measurements from various satellite instruments: from GOME over the Indian Ocean , from SCIAMACHY onboard ENVISAT over the Indian Ocean and the Red Sea , and in even more detail with many more visible ship tracks from GOME-2 onboard Metop-A . The higher resolution of OMI yielded ship tracks in the Baltic Sea and in all European seas .

The current study is part of the project MESMART (Measurements of shipping emissions in the marine troposphere), which is a cooperation between the University of Bremen (Institute of Environmental Physics, IUP) and the Federal Maritime and Hydrographic Agency (Bundesamt für Seeschifffahrt und Hydrographie, BSH), supported by the Helmholtz Zentrum Geesthacht. It aims to monitor background concentration as well as elevated signals of gases and particles related to ship emissions with various methods to cover a wide range of relevant pollutants and their spatial and seasonal distribution to estimate the influence of ship emissions on the chemistry of the atmospheric boundary layer (for further information visit http://www.mesmart.de/).

The objectives of this study are to assess whether measurements of individual ship plumes are feasible with a ground-based MAX-DOAS instrument, to compare MAX-DOAS with co-located in situ measurements, to estimate the contribution of ships and land-based sources to air pollution in a North Sea coastal region, to survey the effect of fuel sulfur content regulations on SO2 concentrations in the marine boundary layer and to analyze the SO2-to-NO2 ratio in plumes to gain information about plume chemistry and the sulfur content in shipping fuels.

In the following, first the measurement site is described, followed by a presentation of the wind statistics and data availability. After this, the DOAS, the MAX-DOAS instrumentation and measurement geometry, and the DOAS data analysis approach used are briefly described. In the next section, selected results from this study are presented: the measured differential slant column densities (DSCDs), the retrieved path-averaged volume mixing ratios (VMRs), the comparison to in situ measurements, the diurnal and weekly variability, the contribution estimates for ships as well as land-based pollution sources, and the analysis of SO2-to-NO2 ratios in ship plumes. Finally, a summary is given and conclusions are drawn.

The principle of optical absorption spectroscopy is the attenuation of light intensity while passing through an absorbing medium, described by the well-known Lambert–Beer law (also known as the Beer–Lambert–Bouguer law). For the general case of electromagnetic radiation passing through an anisotropic medium with a number density n and a temperature- and pressure-dependent absorption cross section σ of an absorbing species along the light path s, the measured intensity at wavelength λ is given by I(s,λ)=I0(λ)⋅exp⁡-∫0sn(s′)⋅σλ,T(s′),p(s′)⋅ds′, with the intensity of radiation entering the medium I0, temperature T and pressure p. For measurements in the atmosphere, this simple model has to be extended by considering multiple trace gases with different absorption cross sections and light scattering on air molecules (Rayleigh scattering), aerosol particles or water droplets (Mie scattering) as well as inelastic scattering by air and trace-gas molecules (Raman scattering). The latter is responsible for the Ring effect , another important extinction process, which can be described by a pseudo cross section.

The key and original idea of the DOAS is to separate the optical depth and the absorption cross sections σ(λ) into a slowly varying function σ0(λ) that accounts for elastic scattering and broadband absorption structures. It is described by a low-order polynomial and a rapidly varying part σ′(λ), the differential cross section, considering the narrow-band absorption structures . The absorption cross sections are measured in the laboratory. Neglecting the temperature and pressure dependence of the absorption cross section, polynomial and differential cross sections are fitted to the measured optical depth ln⁡I/I0 in the linearized so-called DOAS equation: ln⁡I(λ)I0(λ)=-∑i=1NSi⋅σi′(λ)-∑pcp⋅λp+r(λ).

The retrieved quantities are the coefficients of the polynomial cp and the slant column densities Si of the absorbers, which are the integrated number densities along the light path: Si=∫ni(s)ds. The fit residual r(λ) contains the remaining optical depth.

The MAX-DOAS technique is a passive remote sensing method measuring scattered sunlight. The MAX-DOAS instrument used in this study comprises of a telescope mounted on a pan-tilt head, an optical fiber bundle, and two spectrometers (UV and visible spectral ranges), each equipped with a charge-coupled device (CCD) camera. The telescope that is attached to the outer sheathing of the circular platform of the Neuwerk radar tower is used to collect the light from a specific viewing direction and to focus the light onto the entrance of the optical fiber. The combination of converging lens and light fiber leads to a field of view of approximately 1∘. The pan-tilt head allows the instrument to point in different azimuth angles (panning) as well as different elevation angles (tilting). Dark measurements, which are needed for the determination of the CCD's dark signal, are undertaken on a daily basis. Also on a daily basis, line lamp measurements are taken using an internally mounted HgCd lamp for the wavelength calibration of the spectra and the determination of the slit function of the instrument. The spectral resolution, represented by the full width at half maximum (FWHM) of the slit function of the instrument, is about 0.4 nm for the UV and 0.7 nm for the visible channel.

The Y-shaped optical light fiber cable is a bundle of 2 × 38 cylindrical, thin and flexible quartz fibers, guiding the light from the telescope to the two temperature-stabilized spectrometers with attached CCD detectors inside the weatherproof platform building. Each single fiber has a diameter of 150 µm and is 20 m long.

The UV and visible spectral range instruments consist of identical Andor Shamrock SR-303i imaging spectrographs and a grating spectrometer in the Czerny–Turner design with a focal length of 303 mm. The gratings in use are different: the UV instrument is equipped with a 1200 groovesmm-1, 300 nm blaze-angle grating and the visible instrument with a 600 groovesmm-1, 500 nm blaze-angle grating. The UV instrument covers the wavelength range 304.6–371.7 nm; the visible spectrometer covers 398.8–536.7 nm. For the UV range, a Princeton NTE/CCD 1340/400-EMB detector with a resolution of 1340 × 400 pixels and a pixel size of 20 × 20 microns, cooled to -35 ∘C is used. For the visible spectral range, an Andor iDus DV420-BU back-illuminated CCD detector with a resolution of 1024 × 255 pixels and a pixel size of 26 × 26 microns, cooled as well to -35 ∘C, is used.

The measurement geometry for the ground-based MAX-DOAS measurements on Neuwerk is sketched in Fig. . To measure ship emissions, the telescope is pointed towards the horizon, collecting light that passed directly through the emitted ship plumes. A close-in-time zenith sky measurement is used as a reference so that the retrieved tropospheric DSCD S′ is the difference of the slant column densities (SCDs) along paths 1 and 2 in Fig. : S′=S1-S2=Soff-axis-Sreference. The stratospheric light path and trace-gas absorption is approximately the same for both measurements and therefore cancels out, which is important for NO2, which is also present in the stratosphere. This approach also minimizes possible instrumental artifacts.

The assumption that the vertical part of the light path cancels out when taking the difference between off-axis and zenith sky (reference) measurements off course is only valid if the NO2 in the air above the instrument, which is of no interest to us here, is spatially homogeneously distributed. This is usually the case for stratospheric NO2. If a spatially limited pollution plume from point sources like ships or power plants is blown above the radar tower and no plume is in the horizontal light path, the mentioned assumption is violated, leading to an underestimation of the derived DSCD. Also, clouds or fog can make the interpretation of the measured DSCD more challenging due to multiple scattering.

The recorded spectra are spectrally calibrated using a daily acquired HgCd line lamp spectrum, and the dark signal of the CCD detector is corrected using daily nighttime dark measurements. The logarithm of the ratio of the measured off-axis (viewing towards the horizon) spectrum and reference (zenith sky) spectrum gives the optical thickness (also called optical depth). Multiple (differential) trace-gas absorption cross sections obtained from laboratory measurements, as well as a low-order polynomial, are then fitted simultaneously to the optical depth. The retrieved fit parameters are the SCDs of the various absorbers and the coefficients of the polynomial. The fits were performed with the software NLIN_D .

The settings and fitted absorbers vary according to the spectral range used. For the retrieval of NO2 in the UV range, a fitting window of 338–370 nm was used and for NO2 in the visible range a fitting window of 425–497 nm was used, both adapted from experiences during the CINDI and MAD-CAT (http://joseba.mpch-mainz.mpg.de/mad_cat.htm) intercomparison campaigns. The oxygen-collision complex O2-O2, often denoted as O4, is simultaneously retrieved from both NO2 fits. The fit parameters for the DOAS fit of NO2 and SO2 are summarized in detail in Table .

For the retrieval of SO2, several different fitting windows between 303 and 325 nm have been used in previous ground-based studies . This results from the need to find a compromise between the low light intensity caused by the strong ozone absorption around 300 nm on the one hand and the rapid decrease in the differential absorption of SO2 at higher wavelengths on the other hand, limiting the choice of the fitting window. In this study, a fitting window of 307.5–317.5 nm was found as the optimal range for our instrument, which is similar to recommendations in . The fit parameters for the DOAS fit of SO2 are summarized in detail in Table .

Only SO2 measurements with a RMS lower than 2.5 × 10-3 have been taken into account for the statistics, filtering out bad fits with ozone interferences in low light and bad weather conditions.

Under optimal conditions, the RMS of a typical DOAS fit is around 1×10-4 for NO2 in the visible range, 2×10-4 for NO2 in the UV range and 5×10-4 for SO2. By assuming that an optical density of twice the RMS can be detected , it is possible to estimate the detection limit of our instrument regarding the different trace gases. The differential absorption cross section of NO2 is of the order of 1×10-19 cm2molec-1 and of the order of 2×10-19 cm2molec-1 for SO2. Combining this yields a NO2 detection limit of around 1×1015 moleccm-2, corresponding to 0.05 pbb in the visible range and 2×1015 moleccm-2 corresponding to 0.1 pbb in the UV range. The SO2 detection limit lies around 2.5×1016 moleccm-2, corresponding to 0.2 ppb. The typical absolute fit errors are 2–3×1014 moleccm-2 for NO2 in the visible range, 5–6×1014 moleccm-2 for NO2 in the UV range and 2×1015 moleccm-2 for SO2, a factor of 5 to 10 smaller than the detection limit.

To measure shipping emissions at our measurement site, our MAX-DOAS telescope is pointed towards the horizon, where the ships pass our site at a distance of 6–7 km. Since our instrument has a field of view of approximately 1∘, the lowest usable elevation angle avoiding looking onto the ground is 0.5∘, providing us with the highest sensitivity to near-surface pollutants. This is the elevation at which the highest slant columns are usually measured at our site. To convert a MAX-DOAS trace-gas column, which is the concentration of the absorber integrated along the effective light path, into concentrations or VMRs, the length of this light path has to be known. This effective light path length depends on the atmospheric visibility, which is limited by scattering on air molecules as well as aerosols. As described in Sect. , trace-gas absorptions in the higher atmosphere like stratospheric NO2 nearly cancel out using a close-in-time zenith-sky reference spectrum. Following this, we can assume that the signal for our horizontal line of sight is dominated by the horizontal part of the light path after the last scattering event. As introduced by , the length L of this horizontal part of the light path can then be estimated using the SCD S of the O4 molecule, which has a well-known number density n=N/V in the atmosphere:

LO4=SO4,horiz-SO4,zenithnO4=SO4′nO4, with the DSCD S′. The surface number density of O4 is proportional to the square of the molecular oxygen concentration and can be easily calculated from the temperature T and pressure p measured from the radar tower: nO4=nO22=0.20942⋅nair2withnair=pair⋅NATair⋅R, with the Avogadro constant NA and the universal gas constant R.

Knowing the path length, it is then possible to calculate the average number density of our trace-gas x along this horizontal path and the path-averaged VMR ν: nx=Sx,horiz-Sx,zenithLO4=Sx′LO4and thusνx=nxnair.

This O4 scaling in principle takes into account the actual light path and its variation with aerosol loading and also needs no assumption on the typical mixing layer height, therefore overcoming the disadvantages of a simple geometric approximation.

However, when the atmospheric profile of the investigated trace-gas x has a shape that differs from that of the proxy O4, systematic errors are introduced, as has been shown by and in extensive and comprehensive radiative transfer model (RTM) simulations. Pollutants like NO2 and SO2 have a profile shape very different from O4. They are emitted close to the ground (e.g., from ships), have high concentrations in low-altitude layers and tend to decrease very rapidly with height above the boundary layer. They are often approximated as box profiles, while the O4 concentration simply decreases exponentially with altitude. This difference in profile shapes violates the basic assumption that the O4 DSCD is a good proxy for the light path through the NO2 and SO2 layers. The resulting near-surface VMRs will not be representative for the amount of trace gases directly at the surface, but for some kind of average over a certain height range in the boundary layer.

The studies like and use correction factors from radiative transfer calculations to account for this. These correction factors depend on the amount of aerosol present in the atmosphere, often described by the aerosol optical density (AOD); the solar zenith angle (SZA) as well as the relative solar azimuth angle; the height of the pollutant box profile; and the extent and vertical position of the aerosol layer in relation to this box profile . The strong dependence of the correction factors on the height of the box profile for trace-gas layer heights of less than 1 km makes it necessary for the application of the suggested parameterization method to have additional knowledge about the trace-gas layer height, ideally from measurements (e.g., lidar) or otherwise from estimations. The use of this method for low boundary layer heights below 500 m without knowing the actual height is not recommended by the authors .

At our measurement site, no additional knowledge (measurements) about the height of the NO2 and SO2 layers is available and the trace-gas layer heights are typically around 200–300 m. A comparison of the uncorrected MAX-DOAS VMRs retrieved with the upper equations to our simultaneous in situ measurements (see Sect. ) confirms the need for a correction factor but also shows that the scaling factor needed changes from day to day as well as during the course of the day. This indicates that the NO2 and SO2 layer heights are very variable, depending on wind speed, wind direction, atmospheric conditions and chemistry. The lack of comparability between both measurement techniques and geometries, which is further discussed in Sect. , prevents us from estimating diurnally varying correction factors.

The non-consideration of these scaling factors will lead to a systematic overestimation of the effective horizontal path length and therefore to a systematic underestimation of MAX-DOAS VMRs, by up to a factor of 3 .

In summary, a detailed radiative transfer study for the determination of the right correction factors is out of the scope of this study, which focuses on the statistic evaluation of a 3-year dataset of shipping emission measurements in the German Bight. Therefore, when the following MAX-DOAS VMRs are shown, it has to be kept in mind that these are uncorrected VMRs obtained by the formulas above.

This approach has been applied successfully by and for measurements in urban polluted air masses over Mexico City and the city of Hefei (China) using MAX-DOAS measurements at 1∘ and 3∘ elevation and only at 1∘ elevation , respectively. applied this approach to measurements at a high mountain site at the Izaña Atmospheric Observatory on Tenerife (Canary Islands), at Zugspitze (Germany) and at Pico Espejo (Venezuela). Due to the low aerosol amounts at such heights, the latter two studies applied the approach without using correction factors. The fact that our instrument is located on a radar tower at a height of about 30 m above totally flat surroundings (the German Wadden Sea) allows an unblocked view to the horizon in all feasible azimuthal viewing directions. This led to the idea of trying to apply this approach to our shipping emission measurements on Neuwerk.

Since the O4 DSCD is retrieved simultaneously to NO2 in both the UV and visible DOAS fits for NO2, this approach can be applied to NO2 retrieved in both fitting ranges. The approach can also be applied to SO2, although the difference of light paths due to the different fitting windows in the UV for O4 (NO2) and SO2 introduces an uncertainty that has to be accounted for. derived an empirical formula from RTM calculations for a variety of aerosol scenarios to convert the path length at 310 nm from the path length at the O4 absorption at 360 nm: L310=0.136+0.897×L360-0.023×L3602, in which L310 and L360 are given in kilometers. This formula was also applied to our measurements to correct the light path length for the SO2 fitting window. Although this formula has been calculated for polluted sites, the authors state that the deviations for other sites with different conditions are expected to be small .

Using Eqs. –, several problems can arise from the division by the DSCD of O4. For example, if the O4 DSCD is negative, which can happen at low signal-to-noise-ratio DOAS fits (e.g., under bad weather conditions), the resulting path length will be negative. If at the same time the trace-gas DSCD is positive, then the trace-gas VMR will be negative as well, a nonphysical result. However, even when there is no NO2 or SO2, there is still some noise and therefore the retrieved VMRs are not exactly zero but scatter around zero. Thus, slightly negative values have to be included when averaging over time to avoid creating a systematic bias. If, however, the O4 DSCD is close to zero, the path length will be very small, leading to extremely high (positive or negative) mixing ratios, which are also unrealistic. To address both problems, measurements with negative or small retrieved horizontal path lengths are discarded. For the measurements on Neuwerk, with respect to the characteristics of the measurement site, a minimum path length of 5 km seems to be a reasonable limit. This value provides the best compromise between the number of rejected bad measurements and the total number of remaining measurements for NO2 in the UV and visible ranges as well as for SO2. For statistics on DSCDs, however, no such filtering is applied since negative values are not unphysical in this case and just mean that there is more trace-gas absorption in the reference measurement than in the off-axis measurement.

In addition to the MAX-DOAS instrument, in situ observations are also taken, using the Airpointer, a commercially available system that combines four different instruments in a compact, air-conditioned housing. The manufacturer is Recordum (Austria), distributed by MLU (http://mlu.eu/recordum-airpointer/). The Airpointer device measures carbon dioxide (CO2), nitrogen oxides (NOx=NO+NO2), sulfur dioxide (SO2) and ozone (O3) using standard procedures. Table shows more detailed information about the different included instruments, their measurement methods, precision, and time resolution.

In this study the in situ 1 min means of all compounds were used. NO2 itself is not directly measured but calculated internally by subtracting the measured NO from the measured NOx concentration.

In this study, 3 years of continuous MAX-DOAS measurements on Neuwerk have been evaluated. Figure  shows for one example day in summer 2014 the measured DSCDs of NO2 in the UV and visible spectral ranges as well as of SO2 for the 0.5∘ elevation angle (viewing to the horizon) and the -25 ∘ azimuth angle (approximately NNW direction; see Fig. ). Sharp peaks in the curves originate from ship emission plumes passing the line of sight of the instrument. On this day, elevated levels of NO2 were measured in the morning, corresponding to a polluted air mass coming from land, which appears as an enhanced, slowly varying NO2 background signal below the peaks. The systematic difference between the NO2 in the UV range (red curve) and the NO2 in the visible range (blue curve) emerges from the longer light path in the visible range due to stronger Rayleigh scattering in the UV range (wavelength dependence ∝λ-4). This is further investigated in Sect.  below.

By comparing SO2 (green curve) with NO2 (red and blue curves), it can be seen that for many of the NO2 peaks there is a corresponding and simultaneous SO2 peak, but not for all of them. This indicates a varying sulfur content in the fuel of the measured ships. Fuel with a higher sulfur content leads to higher SO2 emissions (see also Sect. ).

By comparing measurements in different azimuthal viewing directions, the movement direction of the ship (and its plume) can be easily distinguished. The zoom in on the right of Fig.  shows the visible range NO2 measurements in different azimuth directions for one example peak from the time series shown on the left. The color-coded viewing directions (see also Fig. ) are sketched schematically below. From the measurements it can be seen that the emitted plume was consecutively measured in all directions at different times. It was first measured in the easternmost viewing directions and at last in the westernmost direction, indicating that the ship and its plume moved from east to west.

For the identification of sources for air pollution on Neuwerk, the wind direction distribution for the DSCDs of NO2 and SO2 measured in 2013 and 2014 is plotted for four different elevation angles (0.5, 2.5, 4.5 and 30.5∘) in Fig. . When the wind comes from the open North Sea (blue shaded sector) the measured NO2 and SO2 DSCDs are clearly lower than for other directions, for which the wind comes from the coast (green and yellow shaded sectors) and blows land-based air pollution to the island. The wind direction dependence is more or less similar for both trace gases but with a higher fraction of ship-related signals in the overall SO2 columns. The values are especially high when the wind comes from the cities of Cuxhaven (ESE direction) and Bremerhaven (SSE) for both NO2 and SO2.

Elevation angle sequences of slant columns (i.e., vertical scanning) contain information on the vertical distribution of trace gases. For lower-elevation angles, the measured trace-gas slant columns for tropospheric absorbers are usually higher because of the longer light path in the boundary layer.

As expected, higher elevations show on average lower DSCDs due to the shorter light path in the boundary layer. The highest NO2 and SO2 DSCDs in the lowest elevation angle (0.5∘, blue bars) in relation to DSCDs in higher elevations are measured especially for wind from all northern directions, in a sector ranging from WSW to ESE. These directions coincide with the course of the main shipping lane coming from the WSW direction (the English Channel, the Netherlands, East Frisian Islands), passing the island in the north and running close to the city of Cuxhaven (ESE direction) into the river Elbe. This indicates that the enhanced columns in the 0.5∘ elevation angle are pollution emitted from ships in a near-surface layer.

For southerly wind directions no major shipping lane is in the direct surroundings and land-based pollution sources dominate. The average DSCDs at 0.5 and 2.5∘ elevation are nearly the same for both NO2 and SO2, indicating that the pollution is located higher up in the troposphere.

For the example day presented in Fig. , the path-averaged VMRs retrieved with the approach presented in Sect.  are shown in Fig. .

From the mathematics of the approach, one would expect a good agreement between the NO2 VMRs retrieved in the UV and visible ranges if NO2 is well mixed in the boundary layer since averaging constant values over different paths should give equal mean values. In the figure, in fact, one can see a very good agreement between both NO2 VMRs, in particular for situations characterized by background pollution.

Although the light path in the visible spectral range is clearly longer than in the UV range, for all the peaks shown here the UV instrument measured a higher path-averaged VMR. The reason for that is spatial inhomogeneities along the line of sight.

If NO2 is not distributed homogeneously along the light path, which is the case in the presence of individual ship exhaust plumes, one can expect different values for the means over the two light paths as they probe different parts of the NO2 field. Such differences can be identified in the figure by looking at the peaks.

The light path in the visible spectral range is longer than in the UV range because of more intensive Rayleigh scattering in the UV range. The difference between UV and visible peak values depends on the exact location of the plume within the light paths.

A short distance of the plume to the instrument and its complete coverage by the shorter UV path leads to higher values in the UV since the part of the light path probing the higher NO2 values has a larger relative contribution to the signal than for the longer visible path.

If the plume is further away from the instrument and only in the visible path or close to the UV scattering point, one will retrieve a higher VMR in the visible range. This relationship contains information on the horizontal distribution of the absorber and will be further investigated in a second paper.

To quantitatively investigate the relationship between the NO2 SCDs measured simultaneously in the UV and visible spectral ranges, all single pairs of DSCD measurements with an RMS better than 1×10-3 are plotted onto a scatter plot, shown in Fig. a.

As can be seen from the figure, NO2 DSCDs in the UV and visible ranges are strongly positively correlated with a Pearson correlation coefficient of 0.983. Because of the difference in the horizontal light path lengths in both spectral regions (due to more intense Rayleigh scattering in the UV range), the slope of the regression line is 1.30 corresponding to a 30 % longer light path in the visible range. The intercept of the regression line is small. Figure b shows a histogram of the ratios between both SCDs. The distribution peaks for ratios of 1.3, in good agreement with the retrieved slope from the scatter plot.

When converting the SCDs to mixing ratios using the O4 scaling, the dependence on light path should be removed and quantitative agreement is expected between the UV and visible VMRs. A scatter plot for the horizontal path-averaged VMRs is shown in Fig. c. It is clearly visible that the points scatter symmetrically along the 1,  1 identity line. Comparing this plot with the plot in panel (a) shows that the difference in light path lengths is in fact corrected for by the O4 scaling approach. The slope of the regression line is close to unity and the intercept is very small. The Pearson correlation coefficient has further increased to 0.984. The histogram (Fig. d) peaks at 1.0.

As discussed above, differences are still expected not only as a result of measurement uncertainties but also due to different averaging volumes in case of inhomogeneous NO2 distributions (which is especially the case for ship plumes under certain wind directions). For the horizontal light path lengths, a mean value of 9.3 km with a standard deviation of 2.3 km was retrieved in the UV range, and a mean value of 12.9 km with a standard deviation of 4.5 km was retrieved in the visible range. On days with optimal measurement conditions (clear sky days), typical horizontal light paths are around 10 km in the UV range and 15 km in the visible spectral range.

The detailed information on passing ships transmitted via the AIS and the acquired weather and wind data can be used to allocate the measured pollutant peaks to individual ships.

Measurements from 9 July 2014 are shown in Fig. . Panel (a) shows the MAX-DOAS DSCD of NO2. Panel (b) includes various information about passing ships: the vertical bars indicate when a ship was in the line of sight of the MAX-DOAS instrument. Solid bars represent ships coming from the left and going to the right (from west to east, i.e., sailing into the river Elbe), dashed bars vice versa. The colors of the bars indicate the ship length, with small ships shown in blue and very large ships (> 350 m) in red. Panel (c) displays the wind speed and direction.

On this day, the wind came from northern directions, directly from the shipping lane, with moderate wind speeds of 10 to 35 kmh-1, resulting in low background pollution values (1–2×1016 moleccm-2) as well as sharp and distinct ship emission peaks (up to 1.2×1017 moleccm-2) of NO2. By comparing the ship emission peak positions to the vertical bars (representing times when ships crossed the MAX-DOAS line of sight) in the schematic representation below, it can be seen that most of the peaks can be allocated to individual ships. In some cases, when two or more ships simultaneously cross the line of sight, the single contributions cannot be separated. Large ships (orange and red bars) tend to exhaust more NO2, while the contribution of small ships (length < 30 m) represented by the dark blue bars is usually not measurable.

The fact that our measurement site is also equipped with an in situ device (see Sect.  for a description), makes it possible to compare the MAX-DOAS VMRs of NO2 and SO2 to our simultaneous in situ measurements. The differences of both measurement techniques need to be considered for such a comparison: MAX-DOAS averages over a long horizontal light path, while the in situ device measures at a single location inside the plume. Since ship plumes usually never cover the whole light path but rather a small fraction of it, very high concentration peaks are usually underestimated in the MAX-DOAS VMR.

Figure  shows the horizontal path-averaged NO2 VMR retrieved from the DSCDs shown in Fig.  as well as the in situ NO2 VMR (panel a) in combination with ship data (panel b) and wind data (panel c).

Ship emission peaks measured by the in situ instrument are both higher and broader than the corresponding MAX-DOAS peaks, leading to a considerably larger integrated peak area, showing the systematic underestimation of the NO2 concentrations inside ship plumes by the MAX-DOAS instrument due to the averaging along the horizontal light path.

Normally, a time shift between MAX-DOAS and in situ peaks exists, which is due to the long distance of about 6–7 km to the shipping lane that the plumes have to travel until they reach the radar tower. This time shift depends on the wind velocity and gets smaller for higher wind speeds. In the figure, this dependency can be seen when comparing the magnitude of the time delay for measurements in the morning (low wind speeds) and evening (higher wind speeds) This travel time also explains the broader peaks in the in situ measurements since the emitted plume spreads and dilutes on its way to the radar tower.

However, if the pollution is horizontally well-mixed in the measured air mass, which is approximately the case for background pollution coming from the coast but not for ship plumes, MAX-DOAS and the in situ instrument should in principle measure the same values. However, as discussed in Sect. , correction factors need to be applied to the MAX-DOAS VMRs to account for the different profile shapes of O4 and the investigated pollutants NO2 and SO2. However, in our case, correction factors cannot be determined because no measurements of the height of the NO2 and SO2 layers exist. The uncorrected VMRs shown here can be strongly underestimated (up to a factor of 3) because they have been calculated with an overestimated path length. This is the case for background pollution as well as shipping emission measurements.

Because of the lack of comparability between both instruments for individual measurements, for a meaningful comparison and the computation of a correlation coefficient at this measurement site, an averaging over longer time spans was applied to reduce the impact of the differences between both measurement methods. The averaging over large horizontal distances in the MAX-DOAS measurements should cancel out on temporal average when comparing to in situ measurements.

Figure  in panel (a) shows 3 months of daily mean NO2 VMRs from the in situ and MAX-DOAS UV instruments in summer 2014. Due to instrumental problems with the in situ SO2 device in 2014 (see Fig. ), panel (b) shows 6 weeks of SO2 daily mean VMRs from summer 2013. To have comparable conditions, for the in situ instrument all measurements between the start of the MAX-DOAS measurements in the morning (with sunrise) and the end of measurements in the evening (with sunset) were averaged. The shaded areas show the corresponding standard deviation and indicate the variability during the single days.

The long gap in the SO2 time series was caused by a power outage.

It is clearly visible that the in situ NO2 VMRs are systematically higher than the uncorrected MAX-DOAS VMRs. The scaling factors that would be needed to bring both time series into agreement differ from day to day. A closer look into the individual days shows that these scaling factors also vary over the course of the day, even when wind direction and speed do not change. The scatter plot for this time series of NO2 measurements in Fig. a shows a good correlation between MAX-DOAS and in situ daily means, but it also shows a slope strongly deviating from 1 and also some scatter.

The most important reason for the systematic differences is certainly the non-consideration of the correction factors arising from the different profile shapes of O4 and NO2, leading to a systematic underestimation of the VMRs from the MAX-DOAS instrument (see Sect.  for a more detailed discussion). However, also light dilution, i.e., light scattered into the line of sight between the instrument and the trace-gas plume , might also play a role in reducing the measured off-axis SCDs.

For SO2, the daily mean VMRs from MAX-DOAS and the in situ instrument in Fig. b show a much better agreement. The scatter plot in Fig. b confirms this with a slope much closer to unity, but more scatter around the fitted line.

The difference in scaling factors for NO2 and SO2 can be attributed to plume chemistry. During combustion, mainly nitric oxide (NO) is produced. This has to be converted to NO2 (through reaction with tropospheric ozone) before it can be measured by the MAX-DOAS instrument. Since the MAX-DOAS instrument sees the ship plumes in an earlier state, the fraction of NO2 should be lower than in the in situ measurements, explaining at least a part of the difference.

Although MAX-DOAS and in situ VMRs show systematic deviations in the absolute values, a very good agreement of the shape (the course) of the curves is found for NO2 as well as SO2. This illustrates that MAX-DOAS can determine day-to-day trends as in situ measurements, even though no correction factors have been applied.

Although our measurement station is located on a small island in the German Bight close to the mouths of the Elbe and Weser rivers, our measurements are strongly influenced by air pollution from traffic and industry on land, depending on the prevailing wind direction. As can be seen from Figs. a and , wind coming from northeasterly, easterly, southerly and southwesterly directions will blow polluted air masses from the German North Sea coast and hinterland to our site. In Fig.  the average diurnal variation in the measured NO2 VMRs is shown as hourly mean values. Solid curves show the respective curve for all measurements (with all wind directions); dashed lines show the subset of measurements with wind coming only from the open North Sea with no coastal background pollution. Looking at the diurnal variation in all measurements, the typical daily cycle for road-traffic-influenced air masses with enhanced values in the morning and in the late afternoon during rush hour can be seen. If we restrain the data to periods with wind from the open North Sea (dashed curves), this diurnal cycle vanishes and values are more or less constant over the day and are also considerably lower. This result is in accordance with the expectations that the amount of ship traffic should be almost independent from the time of day.

The mean NO2 volume mixing ratios for each weekday shown in Fig.  again illustrate the influence of land-based road traffic. If we consider the whole time series (solid lines), the lowest values are measured on Sundays, when road traffic is less intense. There is only a small weekly cycle for air masses coming from the open North Sea (dashed lines). Measurements are more or less constant and again considerably lower. Such a weekly cycle for NO2 in polluted regions has been observed and discussed several times before, for example in , , and .

It is also remarkable that except for a scaling factor of approximately 0.4, the shape of the diurnal and weekly cycles retrieved from MAX-DOAS and in situ measurements agrees very well for both instruments.

As already mentioned in Sect. , on 1 January 2015 the sulfur content of marine fuels allowed inside the North and Baltic seas ECAs was substantially decreased from 1.0 to 0.1 %. Therefore, one would expect lower sulfur dioxide (SO2) values in 2015 compared to the years before, especially when the wind is blowing from the open North Sea, where shipping emissions are the only source of SO2. This expectation is confirmed by the measurements. In the data since 2015, no distinct ship emission peaks have been visible anymore (for an example day see Sect.  below). For a more detailed analysis, mean values over the whole time series before and after 1 January 2015 have been investigated, separated according to the prevailing wind direction.

Excluded from the time series were 2 days of SO2 measurements (20 and 30 October 2014) that showed very high values over several hours. Comparisons with our simultaneous in situ measurements and measurements from the German Umweltbundesamt (German Federal Environmental Agency) at the coast of the North Sea on Westerland/Sylt and at the coast of the Baltic Sea on the island Zingst that show a similar behavior, as well as HYSPLIT backward trajectories, suggest that on both days SO2 plumes of the Icelandic volcano Bárðarbunga influenced the measurements in northern Germany.

Figure  shows the wind direction distribution of the mean path-averaged VMRs of NO2 and SO2 for all measurements before and after the change in fuel sulfur limit regulations.

For SO2, a significant decrease is found, particularly for wind directions from west to north with wind from the open North Sea. For this sector, values in 2015 are close to zero. This shows that the new and more restrictive fuel sulfur content limits lead to a clear improvement in coastal air quality. For wind directions with mainly land-based sources, no or only a small decrease is observed.

The typical average SO2 concentrations measured by the German Federal Environmental Agency in 2016 for rural stations in northern Germany are around 0.5–1 µgm-3, corresponding to 0.2–0.4 ppb (conversion factor: 1 ppb =^ 2.62 µgm-3 for SO2). Measurements in cities and especially close to industrial areas show higher values. Bremerhaven, which is the station closest to our instrument, has a mean concentration of 1.77 µgm-3, corresponding to 0.67 ppb. The reported values for rural stations are in good agreement with our measurements of 0.3–0.4 ppb for wind directions with mainly land-based pollution sources (green sector in Fig. b) since January 2015.

For NO2, however, both the directional distribution and the absolute values are nearly identical for both time periods, implying no considerable changes in NOx emissions. This result meets the expectations since no NOx emission limits have been enforced up to now for the North and Baltic seas ECA.

The distribution of measured NO2 and SO2 VMRs depending on the wind direction shown in Fig.  can be used to estimate the contributions of ships and land-based sources to coastal air pollution levels. To trade ship emissions off against land-based emissions (e.g., industry, road transport), two representative sectors of wind directions have been chosen, both 90∘ wide: a northwesterly sector (258.75∘ to 348.75∘) with wind from the open North Sea and ships as the only local source of air pollution and a southeasterly sector (123.75∘ to 213.75∘) with wind mainly coming from land and almost no ship traffic. Air masses brought by wind from the other directions, for example from the mouth of the river Elbe to the east of Neuwerk, can contain emissions from land-based pollution sources as well as ship emissions. These remaining directions will be referred to as “mixed”. It is now assumed that trace-gas concentrations measured during periods with wind from one of these sectors have their source in the according sector. For getting a good statistic, measurements in all azimuth angles have been included. Figure  shows the results in several pie charts.

For both NO2 and SO2, more than half (around 50–60 %) of all measurements have been taken while wind was coming from either the assigned sea or land sector. This implies that not only a small sample but also the majority of measurements can be used for the estimation of source contributions, making the assumption of using these sectors as representative samples for ships and land-based source regions a reasonable approximation. There are differences in the time series of NO2 and SO2 coming from the fact that the SO2 fit delivers realistic values only up to 75∘ SZA and the NO2 was fitted until 85∘ SZA, leading to less measurements for SO2 than for NO2, which is especially pronounced in winter. Despite this, the general distribution pattern of wind direction frequency for NO2 and SO2 is quite similar, with wind coming from the sea 32–42 % of the time and from the land sector 18–24 % of the time.

For NO2 (upper row in Fig. ), more than half of the total NO2 measured on Neuwerk can be attributed to wind from either of both sectors, with 21 % coming from ships and 31 % coming from land.

If we consider only the two sectors, for which we can identify the primary sources and take theses as representative, we can say that 40 % of the NO2 on Neuwerk comes from shipping emissions, but with 60 %, the majority, coming from land. One reason for this is that the island Neuwerk is relatively close to the coastline (around 10 km) and is obviously still impacted by polluted air masses from land, which has also been observed in the diurnal and weekly cycle analysis shown in Figs.  and . This might also show that in coastal regions in Germany land-based sources like road traffic and industry are, despite the heavy ship traffic, the strongest source of air pollution and ship emissions come in second.

For SO2 the whole time series of measurements from 2013 to 2016 was divided into two periods of nearly the same length: The first period is 2013 and 2014, which was before the introduction of stricter sulfur limits for maritime fuels in the North Sea on 1 January 2015. The according statistics for this period are shown in the middle row in Fig. . The second time period, after the change in fuel sulfur limits, includes all measurements from 2015 and 2016, with the corresponding pie plots in the bottom row of Fig. .

Before the change, 32 % of the measurements were taken when the wind was blowing from the sea sector and about 24 % when it was blowing from the dedicated land sector. After the change, the wind came a bit more often from the sea (42 %) and less often from land (18 %), but in general the situation was quite similar.

The contributions of the three sectors (land, sea and mixed) to the total integrated SO2 with 21 % coming from ships, 30 % from land and 49 % from the mixed sector for the time before the change in sulfur limits are very similar to those of NO2. After the change, the contribution from the sea sector shrinks significantly from 21 to 7 %, while the relative contribution from the land sector increased from 29 to 44 %, with the contribution from the mixed sector staying the same at around 49 %. This increase for the land source sector is only a relative increase while the absolute contributions slightly decreased, as can be seen from Fig. . The relative contribution from the sea sector (shipping-only source) decreased by a factor of 3, while the absolute contribution from this sector decreased by a factor of 8, even though the wind was coincidentally blowing more often from the open sea in this time period.

The overall mean SO2 VMR before 2015 is 0.39 ± 0.45 ppb (mean ± standard deviation). For 2015 and 2016, the total mean value declined by two-thirds to 0.15 ± 0.34 ppb (mean ± standard deviation).

These results clearly show that the stricter limitations on the fuel sulfur content are working and significantly improved air quality in the North Sea coastal regions with respect to SO2. This is in good agreement with other studies such as , who found that around 95 % of the ships are complying with the new limits. This implies that the cheaper high-sulfur heavy oil fuel is no longer in use in the region of measurement.

If again the two selected sectors are considered as representative for both land and sea sources, the shares of the contributions in the sea / land ratio changed from 42 % : 58 % (which is very similar to those of NO2) to 14 % : 86 %. This again shows that since 2015, the vast majority of SO2 emissions can be attributed to land sources and ships play only a negligible role. Prior to 2015, shipping emissions were a significant source for SO2 in coastal regions.

One aspect that is neglected in the source allocation to wind sectors is that in situations with good visibility and low wind speeds even for wind coming from southern directions, the MAX-DOAS instrument can measure ship emission peaks in the north of the island, but they are typically very small. Compared to the often strongly enhanced background pollution in cases with southerly winds, the contribution from these peaks is negligible (around 1–3 %) but certainly leads to a small overestimation of land sources.

Monitoring of emissions from single ships requires the analysis of individual plume peaks in the NO2 and SO2 datasets. It is difficult to derive the absolute amounts (e.g., in mass units) of the emitted gaseous pollutants using our MAX-DOAS remote sensing technique. The height and width of the measured peaks does not only depend on the amount of emitted pollutants but also strongly on the geometry, while getting the highest values when measuring alongside the plumes and much smaller values when the plume moves orthogonal to the line of sight of our instrument. In addition, the time span between emission and measurement also plays a role in the height of the NO2 peaks because of NO-to-NO2 titration.

To determine the mixing ratio inside the plumes, additional information on the length of the light path inside the plume would be needed, which cannot be retrieved from our measurements. This means that without further assumptions, we cannot determine emission factors for the emitted gases (e.g for emission inventories, which are used as input for model simulations).

Although emission factors cannot be measured by MAX-DOAS directly, the NO2 and SO2 signals yield the ratio of both. These ratios can then be compared to ratios of emission factors reported in other studies as well as measurements at other sites or with different instruments, bearing in mind possible deviations due to NO-to-NO2 titration.

By comparing SO2-to-NO2 ratios from different ships, it is possible to roughly distinguish whether a ship is using fuel with high or low sulfur content (giving a high or low SO2-to-NO2 ratio). Beecken and Mellqvist from Chalmers University of Technology (Sweden) use this relationship for airborne DOAS measurements of ship exhaust plumes on an operational basis in the CompMon project (compliance monitoring pilot for MARPOL Annex VI) . Following the ships and measuring across the stack gas plume they can discriminate between ships with a low (0.1 %) and high (1 %) fuel sulfur content with a probability of 80–90 % .

From the spectra measured by our MAX-DOAS UV instrument, both SO2 and NO2 columns can be retrieved at once. The two columns are measured at the exact same time along nearly the same light path. To calculate SO2-to-NO2 ratios for the measured pollutant peaks, the ratio of the measured DSCDs simply has to be computed.

In order to separate ship-related signals from smooth background pollution, first a running median filter was applied to the time series of NO2 and SO2 measurements with a large kernel size (e.g., over 21 points). If too many broad peaks are contained in the time series, this is not sufficient and the resulting median might be systematically higher than the actual baseline. In this case, a running median with a smaller kernel size (e.g., 5) was applied to the values in the lower 50 % quantile again, giving a good approximation of the real baseline.

In the next step, this baseline is subtracted from the raw signal. A simple peak detection algorithm was used to identify the peaks in the baseline-corrected NO2 signal. Then the corresponding peaks in the SO2 were assigned, thus accounting for cases when no SO2 enhancement is measured. In a final manual checkup, all the identified peaks were looked through, filtering out for example all the cases in which peaks are too close together to be separated and fine-tuning the baseline detection algorithm parameters if necessary.

To achieve a better signal-to-noise ratio, the integrals over both the NO2 and SO2 peak are calculated and the ratio of both values is computed in the last step.

Figure  shows the approach as well as the results for an example day in summer 2014, before the stricter fuel sulfur content limits were introduced. Both the NO2 and SO2 signals show high and sharp peaks, originating from ship plumes. Most of the peaks are of similar shape in the NO2 signal as well as SO2 signal. The measured SO2-to-NO2 ratios lie in the range from 0.17 to 0.41. The SO2-to-NO2 ratio can vary strongly for different ships. For example, the plume of the ship passing the line of sight around 12:00 UTC has a high NO2 content, but is low in SO2, whereas the opposite is true for the ship passing at 12:30 UTC, indicating that the second ship was using fuel with a considerably higher sulfur content than the first one.

Figure  shows one example day in summer 2015, after the establishment of stricter sulfur limits. For better comparison to Fig. , the y-axis limits are the same. High NO2 peaks also occur on this day. However, the SO2 signal shows no clearly distinguishable peaks anymore, a result of much less sulfur in the fuel. Consequentially, the measured SO2-to-NO2 ratios are much smaller on this day and range from 0 to 0.09. There might be some small peaks in the SO2 signal, but for most of them it cannot be determined if these are real enhancements or just noise fluctuations. The two peaks at 10:40 and 14:00 UTC, slightly above noise level but still very small, might be real SO2 signals from ships with a higher-than-average fuel sulfur content.

For a statistically meaningful comparison of both time periods, two representative samples of ship emission peaks have been selected by hand for days with good measurement conditions, which were identified by using the solar radiation measurement data of our weather station. One sample of more than 1000 peaks, measured in 2013 and 2014 and representing the state before introduction of stricter fuel sulfur content limits, and another equally sized sample of more than 1000 peaks, measured in 2015 and 2016 representing the situation afterwards, were analyzed in a semiautomatic way. It cannot be ruled out that a certain fraction of ships were measured repeatedly on different days. It is also highly probable that the plumes from some individual ships were measured multiple times at different locations in the different azimuth directions while the ship was passing the island.

The distributions of the SO2-to-NO2 ratios derived from the peak integrals for the two samples are shown in a histogram in Fig. . It can be seen that SO2-to-NO2 ratios were considerably higher before 2015, with a mean of 0.30, a standard deviation of 0.13 and a median value of 0.28. After the change in fuel sulfur content limits, the SO2-to-NO2 ratios became much lower, with a mean of 0.007, a standard deviation of 0.089 and a median value of 0.013, a drastic reduction. A Welch's t test (unequal variances t test) shows that the reduction is statistically highly significant. These results can be compared to the overall average SO2-to-NO2 ratios on all days with good measurement conditions from which the peaks have been selected: for the time before 2015, this gives a mean value of 0.10 and a median of 0.17, and for 2015 and 2016, this gives a mean value of 0.024 and a median of 0.058. As expected, these values are significantly lower than the SO2-to-NO2 ratios obtained from the ship plumes that do not include background pollution.

It is also interesting to compare our results with those from other studies, bearing in mind possible systematic differences due to different measurement geometries, techniques and sites and therefore different NO-to-NO2 titration in the plumes.

measured NO2-to-SO2 emission ratios in marine vessel plumes in the Strait of Georgia in summer 2005. In a sample of 17 analyzed plumes, a median molar NO2 / SO2 ratio of 2.86 was found. Translated into a SO2 / NO2 ratio, this yields a value of 0.35, which is, considering the small sample size, in good agreement with our findings for the time before 2015.

Another study was carried out by measuring gaseous and particulate emissions from various marine vessel types and a total of 139 ships on the banks of the river Elbe in 2011. SO2-to-NO2 emission ratios can also be derived from their reported SO2 and NO2 emission factors: for small ships (< 5000 tons) a ratio 0.13 and an average fuel sulfur content of 0.22±0.21 % was found; for medium size ships (5000–30 000 tons) a ratio of 0.24 and a fuel sulfur content of 0.46 ± 0.40 % was found; for large ships (> 30 000 tons) a ratio of 0.28 and a fuel sulfur content of 0.55 ± 0.20 % was found. Especially the values for medium and large ships fit quite well to our results, while plumes from very small vessels (if measurable at all) have often not been taken into account for the statistic because of the low signal-to-noise ratio.

When assuming that the dependency of SO2-to-NO2 ratio on fuel sulfur content is also applicable to our dataset, we can roughly estimate that the ships measured by us before 2015 used an average sulfur content of 0.5–0.7 %, in good agreement with the results of , which since 2015 have decreased drastically, with 0.1 % as an upper limit.