Ships going slow in reducing their NOx emissions: changes in 2005–2012 ship exhaust inferred from satellite measurements over Europe

We address the lack of temporal information on ship emissions, and report on rapid short-term variations of satellite-derived ship NOx emissions between 2005 and 2012 over European seas. Our inversion is based on OMI observed tropospheric NO2 columns and GEOS-Chem simulations. Average European ship NOx emissions increased by ∼15% from 2005 to 2008. This increase was followed by a reduction of ∼12% in 2009, a direct result of the global economic downturn in 2008–2009, and steady emissions from 2009 to 2012. Observations of ship passages through the Suez Canal and satellite altimeter derived ship densities suggests that ships in the Mediterranean Sea have reduced their speed by more than 30% since 2008. This reduction in ship speed is accompanied by a persistent 45% reduction of average, per ship NOx emission factors. Our results indicate that the practice of ‘slow steaming’, i.e. the lowering of vessel speed to reduce fuel consumption, has indeed been implemented since 2008, and can be detected from space. In spite of the implementation of slow steaming, one in seven of all NOx molecules emitted in Europe in 2012 originated from the shipping sector, up from one in nine in 2005. The growing share of the shipping contributions to the overall European NOx emissions suggests a need for the shipping sector to implement additional measures to reduce pollutant emissions at rates that are achieved by the road transport and energy producing sectors in Europe.


Introduction
Current emission inventories suggest ship NO x (NO x = NO + NO 2 ) emissions account for approximately 15% (3.0-10.4 Tg N yr −1 ) of total anthropogenic NO x emissions (e.g. Corbett et al 2007, Paxian et al 2010, IMO 2014. NO x emissions lead to tropospheric ozone (O 3 ) production and aerosol formation, both of which deteriorate air quality and influence climate change. Over the past decade, numerous studies have used satellite NO 2 observations to constrain various NO x emission sources, like those from anthropogenic (e.g. Martin et al 2006, Stavrakou et al 2013 and soil activity (e.g. Jaeglé et al 2005, Vinken et al 2014a, or from biomass burning (e.g. Mebust et al 2011, Castellanos et al 2014 and lightning (e.g. Boersma et al 2005, Bucsela et al 2010. However, few researchers have yet attempted to reduce the substantial uncertainties in the shipping sector NO x emission inventories by space-based observations of NO 2 . Bottom-up uncertainties arise from the extrapolation of only a few measurements and assumptions about important emission drivers, such as fuel consumption, and the number and velocity of ships. More recent AIS data-based ship emission estimates use highly accurate information on ship position and speed, but contend with uncertainties on ship engine speed and power, and depend on the availability of AIS data and up-to-date information on the vessels (e.g. Jalkanen et al 2009). Furthermore, existing spatially-resolved emission inventories often cover only one year per decade, and miss the considerable year-toyear increases in ship emissions due to increased world trade over the past decade (estimated at 5% per year (Eyring et al 2005)). Such temporal changes, however, are covered in the IMO reports (IMO 2009(IMO , 2014. According to Faber et al (2012) and IMO (2014), and references therein, fuel consumption and emissions decreased sharply in response to the economic downturn in 2008-2009, but as of yet, this decrease has not been confirmed by independent observations on ship emissions or on ship speed. Here we document rapid changes in ship NO x emissions inferred from space in a period of economic turmoil. This will improve our understanding of what drives these changes over areas where air quality is currently not monitored by other methods.
Busy ship lanes have been identified using satellite observations of tropospheric NO 2 columns. Over the Indian Ocean, Beirle et al (2004) demonstrated that enhanced pollution levels could be detected along well-known shipping lanes in maps of tropospheric NO 2 columns from the Global Ozone Monitoring Experiment (GOME) instrument. Several more ship tracks have been identified over the Red Sea and China Sea using the SCanning Imaging Absorption spectro-Meter for Atmospheric CHartographY (SCIA-MACHY) (Richter et al 2004). Recently, Vinken et al (2014b) used NO 2 observations from the Ozone Monitoring Instrument (OMI) to constrain European ship emissions for 2005 and 2006 in ship lanes in the Baltic Sea, the North Sea, the Bay of Biscay, and in the Mediterranean Sea. The temporal evolution of OMI tropospheric NO 2 columns over the Baltic Sea was also shown by Ialongo et al (2014). These studies demonstrate that satellite instruments are able to observe localised NO 2 pollution over ship lanes, and can provide worldwide, robust observations that are useful to provide constraints on these emissions over long time periods.
Several papers reported on trends in tropospheric NO 2 columns from satellite data sets (e.g. Richter et al 2005Richter et al , van der A et al 2006. In Europe, Castellanos and Boersma (2012)  OMI provides daily worldwide measurements, and has a spatial resolution as small as 13 × 24 km 2 for nadir pixels. We use retrieved NO 2 column densities from the Dutch OMI tropospheric NO 2 (DOMINO) v2.0 product . The error in individual OMI observations is estimated to be 1.0 × 10 15 molecules cm −2 + 25% . The DOMINO v2.0 NO 2 retrieval has been validated with in-situ Multi-Axis Differential Optical Absorption Spectroscopy (MAX-DOAS) measurements ( Figure 1 shows an average map of OMI tropospheric NO 2 columns for June-August, 2005-2010. Pollution from a mixture of several sources (e.g. traffic and industry) can be observed over land, but over remote seas ship emissions can directly be identified. The map shows clear enhancements over ship tracks in the Bay of Biscay and the Mediterranean Sea. NO 2 pollution from ships in the North Sea and the Baltic Sea is not immediately distinguishable in this all-measurement average map of tropospheric NO 2 columns. To optimize the detection of ship tracks in the North Sea and the Baltic Sea, we use a filter technique to exclude observations which: (1) are influenced by strong outflow of pollution from land; (2) cover the area of the ship track only to limited extend (less than 75%, caused by cloud cover); or (3) have strong negative values (lower than −0.2 × 10 15 molecules cm −2 ).
In this study retrieval errors are minimized by excluding clouded, snow, or ice covered pixels, and we only use observations with a cloud radiance fraction below 0.5, and a surface albedo below 0.2. Furthermore, we remove the outer 2 (large) pixels on each side of the swath to reduce spatial smearing. We re-gridded the OMI NO 2 observations onto the GEOS-Chem nested horizontal grid (1/2°× 2/3°), and require that over 75% of a grid cell is covered by OMI observations, with more than 3 valid observations per seasonal or yearly average in each grid cell.
Reports in the peer-reviewed literature (e.g. Corbett et al (2009)) and from the shipping sector indicate that shipping companies started adopting 'slow steaming' from the year 2008 onwards, when companies lowered the speed of ships to reduce fuel usage and overcapacity (Bonney 2010, Rodrigue et al 2013. This is the first self-imposed emission reduction regime for an international industry sector, and is highlighted in the last IPCC report as a 'model for future international cooperation for other sectors' (Sims et al 2014). Combining the OMI-inferred ship NO x emissions with information on the number of ships in a sea provides information on the average emission factor per ship, which should have decreased as a result of adopting slow steaming. In this study, we derive the ship density (number of ships in a pre-defined maritime area) from observations of satellite-born altimeters as recently reported by Tournadre (2014). The detection method is based on the detectable signature in the portion of the echo waveform above the sea of high-resolution satellite altimeter waveforms (Tournadre 2007). These waveforms are constructed by measuring the backscattered power of a nadir looking radar by the sea surface (or ship) as a function of time. Tournadre et al (2012) demonstrated that this method could be used to detect the distribution of small icebergs in the Southern Ocean, and Tournadre (2014) recently combined observations from seven altimeter satellite instruments to present a 2-decade database of ship densities for the global oceans. The altimeter data are representative for the density of all ships (commercial vessels, ferries, cruise and military ships) in a particular stretch of the world's seas, but over narrow shipping lanes the altimeter data are probably more representative for those ships that travel along these lanes (e.g. tankers, containerships). Information on the speed of ships in the Mediterranean Sea (v ship ) can be derived (see appendix A) by: where F in represents the inflow of ships in the Mediterranean Sea (i.e. ships passing through the Suez Canal (Suez Canal Authority 2014), and C ship is the ship density (number of ships per unit area) detected by the altimeter instruments over the Mediterranean Sea (Tournadre 2014).

Estimating ship NO x emissions from OMI satellite observations
We use a mass-balance approach to estimate ship NO x emissions using the local ratio of OMI NO 2 observations to NO 2 columns simulated with the GEOS-Chem chemistry transport model (appendix B). We do this for 4 areas characterized by a distinct enhancement in tropospheric NO 2 as indicated by the dashed rectangles in figure   prior NO x emission inventory from the ratio of observed-to-simulated NO 2 columns. In the second step, we simulate NO 2 columns with GEOS-Chem with the scaled-up prior inventory. From the difference between the initial and scaled runs we calculate the sensitivity of the simulated NO 2 column to local NO x emissions changes. This sensitivity, the β-factor (Lamsal et al 2011), accounts for the non-linear response of NO 2 columns to changing NO x emissions, and details of its calculation can be found in Vinken et al (2014b). In the final step, we scale the prior inventory (E p ) by the ratio calculated in step one, multiplied with the β-factor, to obtain top down OMI ship NO x emissions (E t ) taking into account nonlinear chemistry: where N O and N G represent the local OMI and GEOS-Chem tropospheric NO 2 columns that have been integrated along the shipping lanes, and corrected for background, non-shipping contributions (Vinken et al 2014b).
We eliminate the effect of a priori NO 2 profile shape on the comparison of model and observations, by applying the averaging kernel (Eskes and Boersma 2003) on the GEOS-Chem simulated NO 2 columns. The averaging kernel, provided along with the OMI retrieval, improves consistency between model and observations by accounting for the vertical sensitivity of the satellite instrument. Our top down NO x emission estimates for 2005 and 2006 agree to within 10% with earlier results presented in Vinken et al (2014b), where we did not apply the kernel but replaced the TM4 with GEOS-Chem a priori profiles. The top-down ship NO x emissions are lower because of the use of the averaging kernel, but patterns are consistent. We note, however, that ship lanes are less resolved in the maps of NO 2 columns due to the use of the averaging kernel (as a result of the rather coarse TM4 a priori NO 2 profiles at 3°× 2°resolution).

Results
We proceed and calculate constraints on NO x emissions over the four European seas, which represent 26% of all European ship NO x emissions, based on equation (2)     (1)ever more or ever larger ships are sailing through the Mediterranean Sea; or (2)ships are sailing at lower speeds, which would increase their residence time in the Mediterranean Sea.
In both cases, the probability of ships being detected by the altimeter instruments has increased throughout the 2005-2012 period. Figure 4(a) also shows the temporal variation of the numbers of ships passing through the Suez Canal (Suez Canal Authority 2014) (orange asterisks). In the period 2005-2008, the counted number of ships passing through the Suez Canal increases in line with the altimeter-detected ship density until 2008, indicating that the number of ships sailing through the Mediterranean Sea indeed increased between 2005 and 2008. In 2009, the number of ships through the Suez Canal falls by 18%, whereas the altimeter-detected ship density continues to increase. This indicates that from this point in time on ships have likely decreased their speed in the Mediterranean Sea (via equation (1)), as the traffic through the Suez Canal represents a large fraction (but not all) of the Mediterranean traffic.
The reported number of ships through the Suez Canal allows us to analyze the change in ship speed, by calculating the derived measure for the average speed of ships in the Mediterranean Sea using equation (1). The average speed of ships is proportional to the ratio of the number of ships passing through the Suez Canal to the altimeter-detected ship density in the Mediterranean Sea. The resulting space-based indicator for ship speed is shown in figure 4(b) by blue triangles, and indicates that the average speed of ships varied by We make full use of the altimeter ship densities by also inferring information on the change of (average) NO x emissions per ship. The ratio of OMI-inferred ship NO x emissions to the altimeter-detected ship density over the Mediterranean Sea can be interpreted as a measure of the average ship NO x emission factor (green squares in figure 4(b)). Ship average NO

Discussion
Uncertainties in the absolute top-down ship NO x emission estimates originate from possible systematic errors in the OMI NO 2 retrievals, in GEOS-Chem NO 2 simulations, and in the inversion method. In line with earlier work by Vinken et al (2014b) we estimate the overall error in the OMI ship NO x emission estimates at 40-60%, assuming that the uncertainty in emissions for seas without OMI constraints is about 50%. However, there is much less uncertainty in the temporal evolution of the NO x emissions than in their absolute levels. Many of the model and retrieval errors are constant in time (e.g. identical assumptions on vertical transport in the model, assumptions on albedo in the calculation of the AMF), and these will express themselves in NO x emission errors in a similar manner from one year to the other. The temporal variability in ship NO x emissions is thus mostly driven by variability in the retrieved NO 2 slant columns (spectral fit), as has been discussed in many earlier papers (Richter et al 2005(Richter et al , van der A et al 2006. The uncertainty in these changes is therefore mostly related to year-toyear differences in sampling, and instrumental effects such as the row anomaly. This detection 'limit' is very difficult to calculate, but we estimate it to be ±10%, in line with previous studies, which indicated that OMI is capable of capturing trends of this magnitude, and from the good consistency between time series of NO 2 columns over shipping lanes derived from GOME and SCIAMACHY with those of OMI (de Ruyter de Wildt et al 2012). The uncertainty is most likely somewhat higher for emission estimates after the row-anomaly.
The year-to-year variation in NO x emissions is near the detection 'limit' (of ±10%) but substantial. Our results contribute to better estimates of the magnitude and temporal variation of regional ship NO x emissions for which currently no other estimates are available.

Conclusions
We report on short-term variability in European ship NO x emission estimates for the period from 2005 to 2012. We estimated ship NO x emissions over four distinct European shipping lanes in the Mediterranean, Bay of Biscay, North Sea, and Baltic Sea based on a mass-balance inversion, tropospheric NO 2 columns retrieved from OMI, and GEOS-Chem model simulations. Ship NO x emissions in European seas have increased from 2005 to 2008, in clear contrast to decreases in land-based NO x emissions in that period. In 2009, average ship NOx emissions over European seas dropped sharply in response to the global economic downturn, and have remained at the 2009 level until 2012, so there is little net change in overall shipping emissions between 2005 and 2012.
We found that the temporal variation in OMIinferred ship NO x emissions corresponds with independent indicators of economic activity. However, variations in ship NO x emissions are not just the result of changes in the European trade volume, but also originate from changes in operational practice by the shipping sector, driven by increasing fuel prices, falling profits, and overcapacity. We used information from independent sources to determine how ship NO x emissions have changed in response to the new industry practices. Using the number of ships per unit area detected by satellite-borne altimeter instruments, we found that average per ship NO x emission factors fell by ∼46% in 2009 (overall emissions fell by 69%) in the Mediterranean Sea and stayed relatively constant afterwards. By combining altimeter ship density data with statistics on ships passing through the Suez Canal, we inferred the average speed of ships in the Mediterranean Sea. The temporal evolution of average ship speed shows a distinct, 30% reduction from 2008 to 2009, and persistently lower ship speeds in successive years. We interpret this as direct evidence that the practice of slow steaming, i.e. reducing ship speed to save fuel, has indeed been implemented widely, resulting in detectable reductions in ship NO x emissions.
Our results indicate that the implementation of slow steaming in 2009 has contributed to offsetting the 2005-2007 increase in NO x emissions over European shipping lanes, but the relative contribution of the shipping sector to total European NO x emissions increased from 11% in 2005 to 14% in 2012. This suggests that, in spite of the implementation of slow steaming, the shipping sector is responsible for an ever-larger share of the European NO x emissions, and additional measures might be required to reduce pollutant emissions at rates that are achieved by the road transport and energy producing sectors. Demand for waterborne freight transport is anticipated to grow deep into the 21st century, and because ships will keep using oil as fuel (Sims et al 2014), achieving (NO x ) emission reductions to benefit climate and human health will be unlikely in the next decades. The improved capacity to monitor ship NO x emissions of space-borne sensors such as the S5P TROPOMI and geostationary sensors (GEMS over Asia, Sentinel-4 over Europe, and TEMPO over North America), all due for launch in the years to come, will provide timely and detailed information on the air pollution caused by ships, which can be used for policy decisions and to better inform the public. with F in as the inflow of ships (in ships m −2 s −1 ), and F out, the outflow from the Mediterranean Sea. k represents the loss (departure) rate (s −1 ) of ships from the Mediterranean Sea. Assuming steady state within a particular year, and no major changes loss processes in the Mediterranean Sea (such as the sinking of ships or the shifting of routes), the steady-state solution can be written as: With C ship representing the density of ships in the Mediterranean Sea, and F in representing the number of ships counted as passing the Suez Canal in that year. The loss rate k can be expressed as the ratio of the ships flowing out of the box per unit time to the ships within the box: The only assumptions we make here is that the fraction of ships sailing through the Mediterranean Sea (on its way to or coming from the Suez Canal) to those from or towards the Black Sea does not change between 2005 and 2012, and that the route ships take through the Mediterranean Sea has not changed (i.e. ships take the shortest route).