The measurement station (39.204576∘ N, 119.004028∘ E) for the “Shipping Emission and Impacts by Switching Oil in Bohai Bay” (SEISO-Bohai) campaign is located on the corner of a main navigational channel to the third pool in Jingtang Harbor (hereafter referred to as JT), on the property (and with the support of) the Hebei Tangshan Harbor Economic Development Zone Office (Fig. 1). Located in Bohai Bay, JT belongs to the Port of Tangshan, one of the core ports in the designated DECAs. China Port Yearbook (2017) reported a total throughput of 520 million tons in the Port of Tangshan, from which JT handled over 270 million tons. The population density around JT and the surrounding “Port Economic Development Area” is high, with over 650 people per square kilometer. JT is described in more detail elsewhere (Xiao et al., 2018). The station consists of a measurement container, which has a small meteorological monitoring station placed on the roof, and aerosol sampling instruments.

A small meteorological monitoring station was placed on the roof of the container and obtained temperature (∘C), relative humidity (%), wind speed (m s-1), wind direction and radiation intensity every 1 min, from 28 December 2016 to 15 January 2017. Abrupt high-temperature values were subtracted from the results as they were obviously invalid data, for example, when an instrument indicated 40 ∘C for ambient temperature in winter.

Continuous concentrations of six gases (NO, NO2, NOx, SO2, and O3 in ppb, and CO in ppm) were measured every 1 min, and PM2.5 and PM10 (in µg m-3) were measured every 1 h, using a Sailhero XHAQMS3000 air quality continuous monitoring system, from 28 December 2016 to 13 January 2017. Monitoring modules consist of NO, NO2 and NOx measurement using an analyzer, SO2 detection using a UV fluorometer, CO using IR absorption, O3 using UV spectrophotometry and particles using β-ray absorption (ISO 10473:2000). The instrument displayed a short, erroneous measurement at the beginning of the campaign, possibly due to unskilled operation, which resulted in some negative values for gas pollutants and overexaggerated values for SO2, CO and O3. The instrument was immediately repaired, and the abovementioned values were removed to ensure the accuracy of the data. Invalid values of O3 occurred sporadically during the campaign, appearing as a sinusoid fluctuation below 10 ppb, and were subtracted from the results. It should be mentioned that the air quality of Xinli Primary School (hereafter referred to as XL, an official air quality monitor site as shown in Fig. 1) was provided by an official air quality monitor and was used as a control (see in http://www.aqistudy.cn, last access: 15 April 2017).

The campaign resulted in the collection of 14 valid particle samples: 2 parallel samples were collected per day before 31 December 2016 and 1 sample was collected per day after that. The filters were exposed for 23 h (normally from 16:30 to 15:30 LST, local standard time, the next day and were labeled according to the end date) on 80 mm (diameter) preferred quartz microfiber filters (CHM QF1 grade) using a Laoying Model 2030 TSP sampler. All samples were immediately put into their original polyethylene plastic boxes, wrapped in two layers of prebacked tinfoil, and then subsequently housed in a refrigerator. In order to avoid any possible contamination of the samples, all of the abovementioned procedures were strictly quality-controlled.

A 0.55 cm2 section of each exposed filter and blank filters were measured for organic carbon (OC) and elemental carbon (EC) concentrations using the thermal optical transmission method in a DRI 2001 organic carbon/elemental carbon (OC/EC) analyzer. OC and EC values were determined via the Interagency Monitoring of Protected Visual Environments protocol (referred to as the IMPROVE A method). Samples were heated in a completely oxygen-free helium atmosphere, using four increasing temperature steps to remove all of the OC on the filter, during which part of the OC was pyrolyzed. Then the pure helium eluent was switched into a 10 % oxygen/helium mixture in the oven and stepped to 800 ∘C for EC determination. OC and EC were detected using a flame ionization detector after oxidation to carbon dioxide and then reduced to methane. The detection limit of this analysis was 0.82 µg cm-2 for OC and 0.2 µg cm-2 for EC.

A 50 cm2 section of each exposed filter and blank filters were extracted using 10 mL of ultra-pure water in an ultrasonic bath at 4 ∘C for 30 min. Inorganic ions of Na+, NH4+, K+, Mg2+, Ca2+, Cl-, NO3- and SO42- were analyzed using a DIONEX ICS-2100 ion chromatograph. The ion chromatograph system was calibrated using a standard solution before the samples were run. Data obtained from a sample were compared to data from the known standard, achieving the identification and quantification of sample ions. The detection limit was 0.1 µg L-1.

A 20 cm2 section of each exposed filter and blank filters were digested using 25 mL of an 8 % HCl/3 %–HNO3 solution in an ultrasonic bath at 69 ∘C for 3 h. The solutions were cooled, vortex mixed and then placed in a centrifuge at 2800 rpm for 15 min to settle any insoluble particle, from which a 5 mL aliquot was taken for the analysis of the following 30 elements using an X Series 2 ICP-MS mass spectrometer: Be, Na, Mg, Al, K, Ca, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, As, Se, Sr, Mo, Ag, Sn, Ba, La, Ce, Hg, TI, Pb, Th and U. Measured Be concentrations were generally 0 µg m-3 throughout the sampling period, and this species was consequently removed from the results. Cr was also removed, as the blank value exceeded most of the sample results. Several concentrations of Cd and Mo were below detection and were also removed.

The climate of JT is strongly influenced by the sea breeze. The mean relative humidity during campaign was 69.4 % (ranging from 21.8 % to 99.9 %), while the mean temperature was -0.6 ∘C. Temperature exhibited a clear diurnal cycle: it was lowest before dawn (-2.3 ∘C), rose after sunrise (07:00 LST) and reached a maximum (14 ∘C) at 14:00 LST (Fig. 1). The prevailing wind directions were west (23.4 %) and north-northwest (13.0 %), and the wind speed mainly ranged between 1 and 4 m s-1 (2.7 m s-1 on average). Coastal meteorological patterns, such as those mentioned above, play an important role in the dispersion, transformation, accumulation or removal of air pollutants (Gariazzo et al., 2007).

During the campaign, the day-to-day variation in emission was large due to variation in both the complicated sources and removal in JT; however, the data generally exhibited a heavily polluted environment in JT. As the primary pollutant at site, PM2.5 concentrations were used to classify the local pollution level. On over 50 % of days, the PM2.5 concentration was above 115 µg m-3, which is the Grade IV criterion of China's daily air quality standard (HJ 633-2012) (Fig. 3), and the mean concentration during the campaign (147.85 µg m-3) was much higher than that of city area of Tangshan (117.9 µg m-3) during the same season (Zhang et al., 2017). The PM2.5 concentration was even three times the wintertime PM2.5 concentration observed in Hong Kong (Gao et al., 2016), and twice that reported at the Yangshan Port, Shanghai (Zhao et al., 2013). This suggests severe air pollution in JT, which is understandable as winter (December–February) is the most polluted period in Tangshan due to the higher emissions and unfavorable atmospheric conditions (e.g., lower mixing heights and more frequent temperature inversions). Since December 2013, when an official air quality monitor station began to operate in the area (http://www.aqistudy.cn, last access: 15 April 2017), the reported PM2.5 concentration during the cold season has always exceeded 100 µg m-3. This situation indicates the necessity for implementing appropriate particle emission reduction measures. Gas pollutants have also been found to be abundant in JT due to the heavy traffic. Average concentrations of NOx, SO2, O3 and CO were 146.93, 21.91, 29.68 ppb and 2.21 ppm, respectively. NOx seemed to be an issue in the port, showing a high maximum hourly concentration of 692.6 ppb during the campaign, whereas SO2 reached a maximum of 165.5 ppb. Peak levels of NOx and SO2 were mainly linked with ship activities, as the measurement site was very close to the channel and the berths. A lower level of O3 was observed in JT compared with Yangshan Port in Shanghai (Zhao et al., 2013), and a clear diurnal cycle of O3 was noted: the concentration rise at daytime (29.18 ppb) and fell at night (16.38 ppb). The combined influence of coastal meteorology was responsible for this cycle to some degree. During daytime, photochemical reactions and transportation of ozone-rich air increases O3, whereas the reaction with NO and dry deposition destroy O3 at night. Research has shown that O3 can be totally destroyed if the NO source is large enough (Finlayson-Pitts and Pitts, 2000), and as our study site was located in a busy port, our data verify this finding (e.g., the O3 concentration was approximately 0 ppb at 21:00 LST on 4 January 2017). Peaks of CO and NOx coexisted to some degree, but overall there was no evident pattern for CO due to complex local combustion emissions.

The SO2 level reduced notably with the maximum hourly concentration dropping from 165.5 ppb before 1 January 2017 to 67.4 ppb after this date, with similar vessel activity. The variation in Fig. 4 confirms this distinct reduction in JT. SO2 exhibited a prompt drop from 77 to 20 ppb on 30 December 2016 compared with the high and steady concentration at XL (the control group from upwind of JT). This SO2 reduction was mainly attributed to local sources, as JT was under the influence of the prevailing atmospheric conditions from XL via diffusion and transmission (NOx and PM2.5 covaried at both sites), where SO2 displayed little change. More precisely, the reduction was most likely a direct response to fuel switching, compared with all of the other variables at port. The wind map shows that the reduction of SO2 was even more notable in almost every wind direction that blew from the navigational channel of JT to our observational site (Fig. 5). As shown in Fig. 1c, a westerly wind blew through the third pool, and a northeasterly wind blew through the second pool, and the SO2 concentration dropped significantly in both directions. On the contrary, in the southwest direction, where wind blew from the city and road, SO2 barely changed, indicating steady non-marine anthropogenic emissions. From this perspective, fuel switching in JT indeed resulted in a reduction in the ambient SO2 concentration of around 70 %.

Variations in carbonaceous and ionic species are depicted in Fig. 6. The mean (range of) concentrations of carbonaceous species determined in PM2.5 were 6.52 (5.46–7.69) µg m-3 for EC, and 23.10 (9.88–41.60) µg m-3 for OC (OC levels are uncorrected for artifacts from absorption/volatilization of gaseous organic species). The levels of EC and OC were fairly consistent with that of PM2.5 collected during the same period in Beijing (X. Li et al., 2018). EC is considered to be a tracer for primary emissions (incomplete combustion), from sources such as ships, vehicles and power plants in our study, which are affected by fuel quality and combustion. However, little variation was observed after fuel switching in this study due to the complicated contributors in the study are (JT). On the contrary, OC concentrations were much higher with large variation, showing the clear prevalence of organic carbonaceous species over EC. However, no discernible effect of fuel switching was found on OC concentrations.

As the major long-range transported aerosol components, SO42-, NO3- and NH4+ dominated the ionic species, with an average concentration of 22.04, 25.95 and 13.55 µg m-3, respectively, and strong correlations with one another. Ca2+, Mg2+, Na+, K+ and Cl- are major constituents of sea salt and mineral dust, with an average concentration of 1.10, 0.21, 0.84, 2.10 and 3.90 µg m-3, respectively (Fig. 6). Port-related emissions have been proven to be one of the major sources of local emission in JT. Cl- was relatively abundant, as the Cl-/Na+ ratio in the aerosol was 4.79, which is much larger than the ratio in the sea salt (1.8), indicating other strong anthropogenic sources such as coal combustion (Yao et al., 2002) and biomass burning (Li et al., 2007, 2009). Ca2+, as an indicator of mineral dust, was higher than that observed in the city area of Tangshan (0.7 µg m-3), which was considered to be related to port activities (load and unload bulk materials). Moreover, the mass ratio of Mg2+/Na+ (0.27) was higher than the value of 0.12 reported for sea water, suggesting additional magnesium sources such as soil dust containing dolomite, which may also be related to port activities. To evaluate the contribution of stationary emissions and mobile emissions to air pollution (Gao et al., 2011), the mean mass ratio of NO3-/SO42- in JT was calculated; the result (1.19) was higher than that of that of the city area of Tangshan (0.7), indicating that JT was more affected by mobile sources than city area.

The high humidity in JT promotes secondary aerosol formation from local emissions (Yu et al., 2018). As the chemical composition of atmospheric particulate matter is largely affected by prevailing weather conditions, the samples were categorized as “polluted days” or “clean days” based on the corresponding PM2.5 mass concentration for comparison (Röösli et al., 2001). The proportion of sulfates and nitrates rose from 55 % on clean days to 70 % on polluted days, showing a large concentration of secondary aerosols either due to transportation or formation. Generally, the SORs and NORs in JT were higher than those of city areas of Tangshan and Beijing (Fig. 7), and increased significantly from clean days (0.14 and 0.10, respectively) to polluted days (0.48 and 0.37, respectively), suggesting a strong localized photochemical redox reaction. The OC/EC ratio can be used as an indicator of the extent of the formation of secondary organic aerosols (Cabada et al., 2004). Despite the fact that the OC/EC emission ratio is dependent on both fuel type and engine type, tests show that it is still able to be utilized to distinguish between marine combustion sources (ratios typically over 10) (Celo et al., 2015; Moldanová et al., 2009; Sippula et al., 2014) and on-road diesel engines (ratios typically ranging from 0.25 to 1) (Oanh et al., 2010). In this study, the mean OC/EC ratio was 3.58, which is much higher than that of the Port of Thessaloniki in Greece (Tolis et al., 2015) and a port in Hong Kong (Gao et al., 2016), which indicates a higher influence from ship emissions in JT. Therefore, the localized photochemical reactions and aerosol formation driven by ship emissions contribute remarkably to air pollution in JT. This sheds light on the fact that secondary pollution should be treated via the reduction of local SO2 emissions as part of pollution control in this harbor, which may be achieved by reducing ship emissions by fuel switching.

The ranges and mean concentrations of all measured elements are shown in Fig. 8. Overall, the mass concentrations of Al, Ti, Mg, Fe, Na, K, Mn, V, Ni, Zn and Pb were abundant and they varied largely with the sample time. Samples were categorized into three batches based on the PM2.5 limit of Chinese IAQI level standard (HJ 633-2012) during sampling, considering the influences of ambient pollution on particulate chemical composition. One specific sample 2017/01/04 was set as a background/control group due to no ship activity during its sampling time (according to the ship traffic information provided), and intra-batch comparison was then performed to estimate the variations in elements after fuel switching.

The enrichment factor (EF) was used to normalize the observed concentrations of elements and to evaluate the influences of crustal and anthropogenic sources. Generally, elements from Ca to K in Fig. 8 are mainly from geological sources and are thus classified as “crustal elements”. Ca was mostly from stable crustal sources which had the lowest EF values. With EF values around 10, and no evident temporal variation, the elements from Ti to U in Fig. 8 may have a major local crustal origin such as dust. Regression analysis comparing the EF values of Na and K revealed a strong correlation (R2 value of 0.994), implying a primarily marine source. Conversely, Co, Mn and V were moderately enriched (EF < 100) and the elements from Ni to Se in Fig. 8 were highly enriched (EF > 100) due to the contribution of anthropogenic sources, and were all classified as “pollution elements”. Co, Mn, Cu and Zn may have originated from the various bulk materials carried within the harbor area (Almeida et al., 2012; Moreno et al., 2007). With EF values that are strongly correlated with one another, Mn, TI, Cu, As, Sn, Zn, Pb, Hg and Ag would likely have a same major anthropogenic source, which is thought to be traffic pollution. According to tunnel studies (Lawrence et al., 2013; F. H. Li et al., 2018) and in situ measurements (Terzi et al., 2010; Thorpe and Harrison, 2008), Mn, Cu, Sn, Zn and Pb in PM2.5 are related to vehicle emissions as well as tire and brake wear. Weckwerth (2001) reported As enrichment in PM2.5 from the shaking of rusting rails due to passing trains; this would explain our observations, as our measurement site is in the vicinity of the train rails around the first pool. With respect to Mo and Cd, there were not enough values to present patterns, although the literature shows that Mo may be a contribution from diesel exhaust and brake wear (Weckwerth, 2001), whereas Cd may stem from motor vehicle emissions (F. H. Li et al., 2018). Again, this proved that JT was more affected by mobile sources, from which heavy metals generally originate.

The establishment of a marker to deduce variations in ship emissions is crucial. There has been a particular focus on Ni and V in PM2.5, as several recent studies have clearly revealed that V and Ni are representative of ship exhaust particles using an aerosol time-of-flight mass spectrometer (Ault et al., 2010; Healy et al., 2009); furthermore, higher V levels in ship emissions have been found to be associated with residual fuel combustion instead of distillate fuel (Agrawal et al., 2008; Celo et al., 2015). In JT, V and Ni were considered to mainly originate from anthropogenic sources, whereas they were considered to be crustal elements in the city area of Tangshan. This indicated a unique contributor in JT, which was clearly ships consuming residual fuel. However in this study, Ni was proven to be less credible as a residual fuel marker, as the concentration of Ni was even inconsistent between parallel samples collected on the same day. On the contrary, the concentration of V, which showed significant intra-batch decreases in all three pollution levels, proved to be highly related to fuel switching; hence, V was identified as the perfect marker for residual fuel emissions in JT.

The chemical composition of MDOs/MGOs indicates that V is below the detection limit, but PM2.5 samples presented the existence of V after fuel switching. This suggests that regionally transmitted V should not be overlooked. According to back trajectories, during the sampling period of sample 2017/01/04, when no ship activity existed, the air mass in control group moved into the Bohai Bay, before turning back to JT. This air mass was able to bring in particles containing a large amount of V from ships cruising in the sea that could still legally use IFO, verifying the influence of air transportation on particle content (Fig. 9). As this pathway would effectively influence the content of air mass, other trajectories were clustered into three typical types based on the transport pathways of air masses, each representing continental dominant, marine dominant and mixed airflows (which are plotted in Fig. 9). The V concentration after fuel switching was 17.4 ng m-3 under coastal airflows (marine and mixing airflow together), which is much higher than 9.0 ng m-3 under continental airflows, indicating the significant effect of ship emissions on coastal areas. As shown in Fig. 9, samples that shared similar transport patterns from Mongolia–Inner Mongolia were compared to rule out the portion of transmitted V. The results showed that ships had switched fuel in advance and, most importantly, that the implementation of low-SF fuel reduced V from ships by 97.1 %, from 309.9 ng m-3 before fuel switching to 9.1 ng m-3 after its implementation.