Waigaoqiao Port (31.337∘ N, 121.665∘ E) is located in northeast of Shanghai city (Fig. 1) and is the largest port in China. The port has about 7 km of docks (a 3 km north section and a 4 km south section). In 2016 the port had yearly traffic of 367 Mt of goods and a container volume of 37.13 million TEU (twenty-foot equivalent unit). Ship categories at the port comprise container vessels (62.4 %), tugs (18.6 %), oil tankers (9.0 %), bulk carriers (1.8 %), Ro-Ro vessels (1.7 %) and other ships (6.5 %) (private data from the port authority). A power plant and a shipbuilding factory reside between the north and south sections of the port, and have their own docks. The air monitoring station at the study site is located on the south bank of Yangtze River, 400 m from the nearest dock.

Monitoring instruments for gaseous species and particulate matter were installed in the station room. The station was equipped with a main sampling tube that extended through the roof. The inlets of the main sampling tube were 1 m above the station roof and 3.5 m above the ground. Ship emission plumes could influence the site when wind directions were from the 300–0–120∘ sector (Fig. 1). During the summer season, the prevailing wind at the site is from the southeast. In the Supplement, a wind rose displaying wind directions during the sampling period is provided (Fig. S1). Approximately 55 % of time, the site was under the influence of port emissions. To the south and west of site there was intense road traffic which comprised of container trucks and the Shanghai outer ring road. In addition to the emissions from ships from the port, the site could also receive important emission influences from traffic when inland winds prevailed.

From 21 June to 21 September 2016, gaseous pollutants – NO–NO2–NOx, SO2 and O3 – were continuously monitored at the port study site using a suit of Thermo Scientific analyzers (NO–NO2–NOx, model 42i; SO2, model 43i; O3, model 49i). Verification and calibration of the instruments were performed regularly using zero checks (by a zero air generator) and span checks (utilizing standard NO2 and SO2 gas of known concentrations; the O3 standard was generated by a calibration photometer system). The PM2.5 concentrations were monitored using the tapered element oscillating microbalance (TEOM) method (Thermo TEOM 1405-F). Calibration of the TEOM did not rely on a standard; the aerosol mass on a filter was monitored by the oscillation frequency change of the tapered element over a specified time. The regular maintenance of the TEOM included the replacement of filters before their mass loadings approached 100 %. The flow rate of the TEOM was checked using a flowmeter. The lower detection limits of the abovementioned instruments are 0.4 µg m-3 for NO and NO2, 0.5 µg m-3 for SO2, 0.5 µg m-3 for O3 and 1 µg m-3 for PM2.5. Weather conditions (temperature, humidity, pressure, wind speed and direction) were monitored by a mini-weather station installed on the rooftop of the station. The weather station sensor was installed about 1 m above the station roof and 3.5 m above the ground. Data from all of the instruments and the weather monitor were managed in a customized database and were set to a 5 min resolution. Atmospheric pollutant concentrations in the Shanghai city area, including gaseous pollutants and PM2.5 concentrations, were monitored concurrently at nine national air quality monitoring stations at a 1 h resolution. The averaged pollutant concentrations at these stations during the same period were included for comparison.

From 21 June to 21 September in 2016, a single-particle aerosol mass spectrometer (SPAMS; Hexin Analytical Instrument Co., Ltd., China) was applied to characterize the single-particle composition and particle size of ambient aerosol at the port study site (Li et al., 2011). The operation principle of SPAMS is briefly described. Ambient aerosol was drawn into the SPAMS vacuum system through a critical orifice with a limited aerosol flow. The particles then entered an aerodynamic focusing lens (AFL) where they were focused into thin beam with transiting velocities in the vacuum as a function of their aerodynamic size. In the SPAMS sizing region, the particles consecutively encountered two continuous laser beams (532 nm wavelength), which reflected light and generated signals in two photomultiplier tubes (PMTs). The time lag between the two PMT signals was used to calculate the particle velocity and to trigger an ionizing laser pulse (266 nm wavelength) at the appropriate time to ionize the same particle. The chemical composition of particles was determined by a dual polar time-of-flight mass spectrometer to record signals for both negative and positive ions. The time lags between the two PMTs of PSL (polystyrene latex) particles of known sizes were used to calibrate the aerodynamic size of ambient particles. Particle size, dual polar mass spectra and particle reflecting signals from the two PMTs were saved for each particle. A PM2.5 cyclone was placed at the inlet of the sampling tube on the roof of the station to remove particles >2.5 µm before they were analyzed by the SPAMS instrument.

Specific components in particles, such as vanadium, are identified by their characteristic mass peaks in the particle spectra. Particles producing vanadium peaks were labeled as vanadium particles. The SPAMS instrument quantified their concentrations in a semi-quantitative manner via the number of detected particles over a specific duration of time. Considering that the aerosol flow was introduced into the SPAMS instrument at fixed flow rate (0.1 L min-1), the detected particle numbers (or the particle detecting velocity) could be utilized as an indication of the ambient particle concentrations. Using ambient sampling, it was shown that particle numbers in the SPAMS instrument were positively correlated with ambient PM2.5 concentrations (R2=0.69 in this study). In the present study, we used the particle detecting velocity of particles containing vanadium as a measure of their concentration. To derive the ambient particle number concentrations from SPAMS particle numbers, we had to consider the efficiency issues of the SPAMS instrument with respect to AFL transmission, laser detection and laser ionization (Wenzel et al., 2003).

The temporal resolution of SPAMS (seconds or minutes) makes it suitable to couple with online gaseous data in order to identify ship emissions. The fluctuations of gaseous concentrations, the shifting of wind directions and the arrival of emission plumes were well captured by SPAMS data. Additionally, this study took advantage of the ability of SPAMS to identify individual ship emission particles by their characteristic composition. Composition patterns of ship emission particles were identified and were then applied to extract the desired particles from all of the analyzed particles. Hence, the temporal trends, the size distribution, the chemical composition and the wind roses of the extracted particles could be examined in further detail.

During the 3-month sampling period, SPAMS generated a large particle dataset (>2.3 million particles were chemically analyzed). To identify ship emission particles from the analyzed particles, we applied a combined method involving peak searching and a clustering algorithm. Specifically, the individual particle mass spectra were visually inspected to get a general mass spectral pattern during ship plumes. It was not feasible to inspect the large number of spectra exhaustively. Instead, we used the concurrent SO2 concentrations to locate ship emission plumes when sharp SO2 peaks occurred, which is typical for RFO combustion (Murphy et al., 2009; Merico et al., 2016). Compared with non-plume periods, the most indicative peaks during plumes occurred at V+(51), VO+(67), Fe+(56), Ni+(58) and serial peaks of elemental carbon at Cn+ (n=1,2,3,…,12) in the positive mass spectrum (Ault et al., 2010, 2009; Healy et al., 2009). In this study the vanadium mass peaks (peak V+(51) and VO+(67)) were determined to be a prerequisite to indicate ship particles during plumes. Further notes regarding this particle identification method for ship emission particles are provided in the Supplement. We applied a rough search with peak criteria of m/z=51 and 67 (i.e., just the existence of mass peaks at 51 and 67 with no peak area limitation) to search all possible candidates from the entire dataset. This particle criteria is not stringent because particles producing organic peaks at the same nominal mass (e.g., C4H3+(51), C4H3O+(67)) could interfere and may enter into searched clusters. The ART-2a algorithm (Song et al., 1999) was then applied to the searched particles to generate subclusters of similar mass spectral patterns (vigilance = 0.85; learning = 0.05; iteration = 20). By inspecting the composition, size and wind rose patterns of subclusters, a small fraction of outlier particles from non-shipping emission sources were subsequently picked out and discarded.

The calculation method for ship emission contributions used in this study, which was originally developed by Contini et al. (2011), is based on the extraction of ship emission plumes from background pollutant concentrations: εA=ΔCAFplmCA,

where εA represents ship emission contributions of pollutant A, ΔCA is the difference between the average concentrations during plumes and non-plumes, Fplm is the fraction of plume cases and CA is the average concentration of pollutant A during the reference period. The uncertainties of εA determined in this method could arise from several factors, such as the definition of the port direction sector, the definition of plumes (the threshold level that determines plumes and background conditions), and pollutant and wind field measurements. This study estimated the uncertainties by subjecting εA to a slight adjustments of the port directions (by ±10∘) and pollutant threshold levels (by 20 %) to inspect its variations. To conform to the original work (Contini et al., 2011), calm wind periods (wind speed <0.5 m s-1) were considered in the evaluation of uncertainties (by either excluding or including calm wind periods).

In the vicinity of the port, the ship emission pollutant concentrations often produced obvious peaks over a relatively short period of time (Fig. 2). These peaks were caused by ship emission plumes relating to shipping activities such as arrival, hoteling (operations while docked) and departure, which typically persist for a few (generally 3–6) hours. The ambient SO2, NO, NO2, O3 and PM2.5 concentrations during a typical period (27–29 August) are shown in Fig. 2 to illustrate several plumes. For comparison, the averaged SO2 concentration in Shanghai city and the vanadium particle number concentrations for the same period are provided. During plume periods, the ambient SO2 concentration peaks correlated well with vanadium particle numbers detected by the SPAMS instrument. The PM2.5 peaks during plumes were not always as unclear as in Fig. 2. In the Supplement we present another period of PM2.5, SO2 and vanadium particle concentrations to demonstrate stronger PM2.5 peaks (Fig. S3). The synchronized gaseous and particulate pollutant peaks in the ship emission plumes were typically observed in port regions (Healy et al., 2009; Ault et al., 2010; Merico et al., 2016). These ship emission plumes were also consistent with the prevailing wind directions of plumes, as shown in Fig. 2.

Considering these facts, the present study defines ship plume periods using SO2 concentrations and vanadium particle number concentrations. For SO2, a minimum threshold of ΔSO2=SO2(Port)-SO2(Shanghai)>5 µg m-3 was applied to indicate the arrival of ship plumes. Additionally, the number concentrations of vanadium particles (PNCv) were considered because in some cases the SO2 peaks were absent or obscure while the number concentrations of typical fresh vanadium particles were high. The probability of the occurrence of this kind of events was low (3 % of cases). These kinds of events were possibly caused by anchored ships burning low sulfur content oil (<0.5 % m/m) to comply with regulations in the port region, which came into force on 1 April 2016; furthermore, the possibility that vanadium particles may have been emitted from industrial sources, such as petroleum refinery companies, in this region cannot be excluded. The wind directions during these events support both of the proposed causes. To identify plumes, we excluded the possible industry influences by limiting the prevailing winds to the port direction only. The present study set the threshold of vanadium particles in ship plumes to PNCv>25 particles h-1. Therefore, ship plumes were identified as either ΔSO2>5 µg m-3 or PNCv>25 particles h-1.

There were about 210 ship emission plumes captured during the entire study period. Table 1 summarizes the statistical analyses of the SO2, NO, NO2, O3 and PM2.5 concentrations at the port site and in an urban area in Shanghai during the study. Vanadium particle number concentrations were represented by the particle detecting velocity measured using the SPAMS instrument. The SPAMS particle detecting velocity values were positively correlated with particle concentrations in the ambient atmosphere, but should not be interpreted as absolute number concentrations without correction for SPAMS efficiency (Wenzel et al., 2003). Statistical analyses were performed on pollution concentrations during both plume and non-plumes periods. To separate the influences from land-based sources, non-plume periods during winds from the direction of the port are calculated in Table 1.

Generally the concentrations of SO2 and NOx at the port site were 40 %–70 % higher than in Shanghai city (Table 1). The SO2 concentrations in non-plume periods were comparable with those in Shanghai city, irrespective of the wind direction; therefore, the non-plume SO2 concentration can be recognized as background SO2 in this area. Conversely, the NOx concentrations showed an obvious dependence on wind direction during non-plume periods, with concentrations being higher when inland winds prevailed; this suggests the importance of land-based emissions at ports in coastal areas. In a similar ambient observation at Yangshan Port, Zhao et al. (2013) obtained average concentrations of 29.4 and 63.7 µg m-3 for SO2 and NO2, respectively, which were higher than the present levels of 15.6 µg m-3 for SO2 and 53.2 µg m-3 for NO2. Noting that SO2 and NO2 were only intermittently measured for about 20 d in the abovementioned study (May and August, 10 d each month), it is not feasible to make a direct comparison. During plume periods, the SO2 maximum hourly concentration in Yangshan (119.0 µg m-3) was close to that found in this study (124 µg m-3); due to land-based emissions, the NO2 maximum hourly concentration at Waigaoqiao Port (260 µg m-3) was higher than that reported at Yangshan Port (199.8 µg m-3).

In general the O3 concentrations at the port site were lower than Shanghai urban region by 13 %–33 %. To inspect whether the O3 depletion was related to the oxidation of primary NO emissions at the port site, we calculated the NO2 ratios to analyze the NOx composition in plumes. The NO2 ratio is defined as the ratio between NO2 and NOx (NO+NO2), and has been used in several relevant characterizations of ship emissions (Alföldy et al., 2013; Kurtenbach et al., 2016). Before the calculation of the NO2 ratio we firstly converted NO and NO2 mass concentrations to molar concentrations. The background NO and NO2 levels were then subtracted to make sure that peaks were due to plumes. The distribution of the NO2 ratio in this study is shown in Fig. 3, where it is compared to the NO2 ratio distribution from ship plumes in another similar study.

The distribution of the NO2 ratios in this study showed several modes. The largest mode occurred at a ratio of about 0.2 (20 %). Obviously this mode was also present in the comparison study (Alföldy et al., 2013), and was recognized as fresh engine emissions from ships. A major difference between the two studies is that a significant fraction of the NO2 ratios occurred in the larger range (>0.4) in the present study, which was not observed in Alföldy et al. (2013). The larger NO2 ratios were initially thought to be emitted from unidentified types of ships. When we correlated the NO2 ratio with ambient O3 concentrations, however, we found that there was an obvious positive correlation between them, as shown in Fig. 3a. This result suggests that the higher NO2 ratios of some plumes were not due to the emission characteristics of ships, but rather due to the transformation of NO to NO2 in the ambient air; hence, if the NO2 ratios in the plumes were high when the plume was discharged and no ambient transformation occurred, there would be no reason to expect the observed dependence of the NO2 ratios on the ambient condition of O3. This is evidence that primary NO emission (from ships or on-road traffic) contributed to the O3 depletion in this area.

With respect to particulate matter, the PM2.5 concentrations in the port area were slightly lower than in Shanghai city, although the PNCv in plumes were approximately 4 times higher than during non-plume periods (Table 1). A longer period of PM2.5 data suggested that the lower PM2.5 concentration is a general trend at this port site. However, this trend is not unique to port regions, and we also observed it in other coastal areas, such as in the PM2.5 spatial distribution for Shanghai (Fig. S4 in the Supplement): there was a general trend of decreasing PM2.5 concentrations from the inner to coastal areas in Shanghai. This fact is assumed to be caused by the dispersion or advection of clean air from the sea. The primary PM from ship emissions at the port site are mostly ultrafine particles with mass emission factors much smaller than NOx and SO2 (Zhang et al., 2017). Therefore, the primary PM from ships or other traffic could not contribute significantly to the ambient PM mass concentrations. The vanadium particle number fractions of the total particles in the SPAMS data were obviously larger (6.7 % on average) at the port site than in urban areas in Shanghai (1 %–2 %) (Liu et al., 2017).

For a coastal port, the evaluation of ship emission impacts on air quality needs to identify the affects of land-based emissions. Obviously these land-based emissions have greater influences on the air quality at the port site than a marine location far from coast (Zhao et al., 2013). To provide an intuitive illustration, the averaged concentrations of SO2, O3, NO, NO2, PM2.5 and vanadium particle numbers under different wind direction conditions are summarized in Fig. 8. The concentrations of pollutants have demonstrated a varied dependency on the local wind conditions. It is evident that, for the coastal port site in this study, the NOx and PM2.5 concentrations were highest when wind from inland directions prevailed. Conversely, the SO2 concentrations and vanadium particle numbers were only dominant when winds originated from port sectors. The hotspots in the wind rose for vanadium particles were most probably produced by individual docks along the riverside. The wind dependence of O3 concentrations was less apparent, except for its depletion in regions with high NOx and SO2 levels in the wind roses, as previously explained. Obviously the port site received very different pollution impacts from land-based emissions and ship-based emissions in port. This study tries to separate land-based emission influences by limiting the wind directions solely to port directions. In the calculation of ship emission contributions, two reference periods were considered: the entire study period (irrespective of wind) and periods when the site was downwind of the port.

Ship emission contributions of air pollutants in the two reference periods are summarized in Table 2. Results show that, if land-based emissions are considered, ship emissions contributed 36.4 % of the SO2 concentration in local air in the port area, which is a much higher value than for NO (0.7 %), NO2 (5.1 %) and PM2.5 (5.9 %). The low contributions of NOx were due to the inclusion of traffic emissions with stronger intensities from inland directions. The main sources of NOx from the inland directions were considered to be not far from the site because the average NOx levels in Shanghai city are lower than those at the port site, as evidenced in Table 1. For the vanadium particle number concentrations (PNCv), ship emissions were the predominant source at the study site (49.5 %). The PNCv contribution is a lower estimation considering that the SPAMS instrument detects smaller particles less efficiently, which is the size range in which vanadium particles tend to concentrate. Contributions of PNCv in different particle size ranges are also shown in Table 2. In both of the reference periods (excluding or including land-based emissions), ship emission contributions to PNCv in the smaller size range (0–0.4 µm) are larger than contributions to PNCv in the larger size ranges (0.4–0.8 and 0.8–2.5 µm).

The relative contribution of PNCv from ship emissions is apparently higher than that of PM2.5 regarding the mass concentration. Previous studies have shown that the direct PM2.5 contribution from ship traffic lies within the range of 1 %–8 % (Contini et al., 2011, 2015). Recent studies carried out in the Mediterranean region found that ship emissions contributed 0.3 %–7.4 % to PM2.5 concentrations in port areas (Merico et al., 2016). Ship emission studies in Europe and other regions were reviewed, and it was concluded that shipping traffic contributions to PM2.5 were in the range of 1 %–14 %, with higher contributions with decreasing particle size (Viana et al., 2014). The calculated value of PM2.5 at the present study site is within the reported ranges. Recently, Merico et al. (2017) compared ship traffic atmospheric impacts using inventories, experimental data and modeling approaches in the Adriatic–Ionian port areas and found that ships contributed 0.5 %–7.4 % of PM2.5 in these areas. The same study also found that the ship traffic contribution to the particle number concentrations (PNC) is 2–4 time larger than the mass concentrations of PM2.5. The PNC is not measured in this study, instead the size distributions and PNC contributions of vanadium particles of different sizes are utilized (as measured by SPAMS), and apparently agree with the abovementioned previous research.

In a study carried out at the Yangshan marine port in Shanghai, the calculated PM2.5 contribution (∼4 %) was found to be smaller than in the present study (5.9 %) (Zhao et al., 2013). In this study a different method was used to evaluate ship emissions, with vanadium concentrations being relied upon to indicate ship emissions. Considering the differences in methodology, it is deemed that the results from the two studies are similar within the uncertainty range (Table 2). A previous estimation in the Shanghai area using an inventory method showed that ship emissions contributed 9 % of NOx and 5.3 % of PM2.5 in the Shanghai area (Zhang et al., 2017), which generally agrees with this study when land-based emissions are included (Table 2). However, for SO2 the estimated contribution of 12 % is significantly smaller than the 36.4 % found in this study. The high SO2 levels in this study are a local character of the port site, which is close to emission sources. After being transported to the urban region, the high SO2 concentrations dissipate and weaken. It is noted that the synchronized SO2 and vanadium particle plumes, as observed at the port site, are observed at a much lower frequency at an urban site in Shanghai city, where another SPAMS instrument is located. An estimation of ship emission impacts on the urban area will be the subject of future studies.

By limiting the analysis to periods when the winds originated from port directions, the influences of land-based emissions could be largely eliminated. As shown in Table 2, the ship emission contributions for all pollutants were magnified considerably in amplitude. The most significant variation occurred for gaseous NOx, as the contributions of this species from ship emission increased to levels larger or comparable with SO2. Contributions obtained here can be compared with a similar study carried out in a European port (Merico et al., 2016). Gaseous emissions of NO, NO2 and SO2 were similar between these two studies, which is impressive considering the larger throughput of goods in Shanghai Port. However, in an absolute sense, this study estimated that ship emissions contributed 5.68 µg m-3 of SO2, 3.00 µg m-3 of NOx and 1.57 µg m-3 of PM2.5 during the sampling period. These values are comparable or higher than the reported results for ports in other regions (Viana et al., 2014). For example, a previous study found that the ship-emitted particles contributed 0.8 µg m-3 of primary particles and 1.7 µg m-3 of secondary particles in Algeciras Bay (Viana et al., 2009). Due to the fact that the study site is adjacent to the port, the calculated PM2.5 contribution could be largely deemed as primary contributions in this study. The relative contributions of pollutants are partly compensated for by the higher background pollution levels in this region.