In this work, the meteorological data at every 1 or 3 h (most at 3 h) from 14 meteorological stations, which were located in or near the core coastal cities mentioned above (Fig. S1), were obtained from the National Climate Data Center (NCDC, ftp://ftp.ncdc.noaa.gov/pub/data/noaa/, last access: 25 October 2018) integrated surface database. The model performance of four major meteorological parameters was evaluated: temperature at 2 m (T2), wind speed at 10 m (WS10), wind direction at 10 m (WD10) and relative humidity (RH). The criteria proposed by Emery et al. (2001) was used to judge the meteorological performance (the mean bias (MB) ≤±0.5 K for T2, MB ≤±0.5 m s-1 for WS10 and MB ≤±10∘ for WD10). High correlation coefficients (R, 0.5–0.9) and low normalized mean errors (NMEs, 6 %–38 %) proved that the model performances were acceptable, although the MB of T2 and WS10 were a litter higher than the suggested goal (Table S4).

We also estimated the model performance of CMAQ v5.2 in predicting the PM2.5 concentrations by comparing the modeled results with observations at 280 monitoring sites of 32 provincial capital cities in China, as described in Table S5. The real-time hourly observations were from the China National Environmental Monitoring Center (CNEMC, http://106.37.208.233:20035/, last access: 25 October 2018), which began to be released since January 2013. The simulated hourly PM2.5 agreed well with observations, with the overall model performance within the performance criteria suggested by Boylan and Russell (2006) (mean fractional bias (MFB) ≤±60 % and mean fractional error (MFE) ≤±75 %), while the model overestimated PM2.5 a little, mainly due to uncertainties in emission inventory and unavoidable deficiencies during meteorological and air quality simulation. In order to verify the spatial accuracy of the simulation, the observed annual mean concentrations of PM2.5 at all sites in the domain were compared with modeling results, as shown in Fig. S2. Model performance was better in coastal areas of eastern China where the economy was more developed and the air quality suffered more serious impacts from shipping emissions, compared to the less developed regions such as western China, due to more accurate emission inventories in the more developed regions. Furthermore, the differences in predicted concentrations of PM2.5 and its components from CMAQ v5.2 and the source-oriented CMAQ were investigated (Fig. S3). In general, the simulated PM2.5 were very similar, but slightly higher concentrations in CMAQ v5.2 were found compared with that in the source-oriented CMAQ. The differences were mostly caused by the difference in the SOA predictions as CMAQ v5.2 included additional SOA formation pathways that were not included in CMAQ 5.0.1 (Woody et al., 2016; Murphy et al., 2017). The predicted secondary inorganic aerosol concentrations showed excellent agreement between the two models, which provided confidence in the predicted source region contributions as described in the results section.

The annual SO2, PM, NOx and HC emissions from ships within 200 Nm of the coastline of China in 2013 were 918.4, 119.3, 1380.9 and 49.3 kt. The emissions were slightly lower than those reported by Li et al. (2018), probably due to the differences in emission factors, AIS data and ship registration databases. The increase in shipping emissions near the coast of China was probably small from 2013 to 2015, because the global CO2 emissions from ships only increased by 2.5 % over that period (Olmer et al., 2017). SO2 and NOx emissions from ships accounted for 20.0 % and 13.5 % of the inland emissions from all sectors in coastal provinces of the MEIC inventory. The cargo ships were the most important contributor to the total shipping emissions, accounting for 43.7 %, 43.4 %, 41.9 % and 40.5 % of SO2, PM, NOx and HC emissions. The container and tanker also contributed 24.7 %–28.4 % and 17.5 %–19.7 % of the total shipping emissions. However, emissions from fishing boats were probably underestimated in this study (approximately 1.0 % of the totals) since most of them had no AIS data, which could affect the air quality significantly (Zhang et al., 2018). The emissions from ships within 12, 12–50, 50–100 and 100–200 Nm were further calculated to identify their contributions (Fig. 1). The emissions within 12 Nm of the shore were the dominant contributors of all pollutants, accounting for 51.2 %–56.5 % of the total shipping emissions. Emissions within 50–100 Nm only accounted for 10.2 %–11.9 % of the totals, which was the least among the four regions. The areal emission rates of SO2, PM, NOx and HC within 12 Nm were 1494.4, 185.7, 2033.5 and 73.3 kg km-2, respectively, approximately 5.1–6.3 times higher than those within 100–200 Nm.

The seasonal variation of shipping emissions was analyzed using the monthly data from January (winter), April (spring), July (summer) and October (fall). The highest shipping emissions of all pollutants were observed in winter, which was on average 1.04, 1.06 and 1.16 times higher than those in spring, summer and fall, respectively. Generally, the changes of total shipping emissions quantities in different seasons were quite small. This pattern was consistent with other studies with similar conclusions (Corbett et al., 1999; Fan et al., 2016; Li et al., 2016). The season variations in the spatial distribution of shipping emissions were also investigated as presented in Fig. 2. Overall, in winter, the shipping emissions were more concentrated along the major lanes between ports than those in other seasons. In addition, the distinct changes were observed in three areas, including the maritime area about 150 km away from the YRD harbors (A1), the southeast of the Taiwan Strait (A2) and the vicinity of Fuzhou port (A3). These seasonal changes were closely related to the activity variations of different ship types (Figs. S4–S6). In spring and summer, mainly due to the increase in long-distance cargo ships, significant emissions occurred in water traffic lanes far from the YRD region (A1). The decrease of cargo ship activities in Fuzhou port during summer and fall also resulted in the obviously reduced shipping emissions in A2. The emissions in A3 were lower in summer and fall because of the decreased activities of all ship types, including cargo ships, containers and tankers. The variations in spatial distribution indicated that monthly shipping emissions used in the air quality model of this work would capture the seasonal impacts more accurately than annual emissions without considering monthly variations.

Based on the results of the CMAQ model with (base case) and without shipping emissions (no ship case) within 200 Nm, the contributions of shipping sector to the inland PM2.5 concentrations were estimated. The annual averaged contribution of the shipping emissions to the increased concentration of PM2.5 (ΔCannual) in 2015 was calculated by averaging the modeling results of January, April, July and October, as presented in Fig. 3. The increased PM2.5 concentration in eastern China caused by shipping emissions was up to 5.2 µg m-3, and as the distance from the coastline increased, ΔCannual decreased dramatically. The most serious impacts were predicted in coastal areas of YRD region (more than 2.5 µg m-3), where large shipping emissions contributed to 20 % of the total shipping emissions in China (Fu et al., 2017). However, due to higher background PM2.5 concentrations mostly induced by land-based anthropogenic sources in China (Fig. S2), the annual mean relative contribution of shipping emissions was less than 12 %, except for Taiwan, which was much smaller than that in Europe (Aksoyoglu et al., 2016). For the same reason, although ΔCannual was smaller in Fuzhou and its surrounding areas than that in YRD, the relative contribution were approximately 2–4 times higher. In addition, in order to quantify how far the shipping emissions can be carried to inland areas, the maximum linear distance (MD) between coastline and areas where the PM2.5 concentration induced by ships was larger than a specific threshold was calculated by GIS (Table 1). The MDs of ΔCannual >0.1, 0.5 and 1 µg m-3 were 960, 510 and 350 km, respectively. The farthest areas affected by shipping emissions were typically located in a similar latitude to YRD (approximately 32∘ N), probably since the low terrain heights in this region were favorable for the long-range transport of air pollutants from sea to the inland areas (Fig. S1). The results illustrated that the shipping emissions influenced the air quality in not only coastal areas but also the inland areas as far as hundreds of kilometers away from the sea.

The shipping emissions caused not only the increase in PPM (element carbon (EC), primary organic aerosol (POA) and primary sulfate (PSO42-)), but also secondary PM (secondary sulfate (SO42-), nitrate (NO3-), ammonium ion (NH4+) and SOA) formed from primary emitted precursors. The contributions of shipping emissions to the annual mean concentrations of the total PM2.5 and individual components in core coastal regions and cities were obtained by averaging the modeled concentrations in all grids of each area (Fig. 4). The averaged increases in PM2.5 concentration in the whole of China and all coastal provinces were 0.2 and 1.3 µg m-3, respectively, and the impacts of shipping emissions in YRD (2.0 µg m-3) and PRD (1.6 µg m-3) were much greater than those in BTH (0.4 µg m-3), where the shipping emissions were more concentrated at port level (Fu et al., 2017). The top four cities most seriously affected by shipping emissions were Qingdao (4.0 µg m-3), Shanghai (3.8 µg m-3), Ningbo (3.8 µg m-3) and Dalian (3.2 µg m-3). In addition to Qingdao, high contributions of the maritime sector were also observed in other coastal cities outside the three core city clusters, such as Lianyungang (2.0 µg m-3) and Fuzhou (2.3 µg m-3), which indirectly indicated that the existing DECA was not long enough in the north–south direction to prevent ship-related air pollution, and it should extend to the entire coastal area of eastern China in the future.

The most important components of PM2.5 contributed by shipping emissions were SNA, accounting for 82.7 %–97.6 % of the total PM2.5 increase, and regional averaged contributions of SO42-, NO3- and NH4+ were 49.2 %, 24.7 % and 15.8 %. The changes of SNA fraction in different areas were both related to shipping emissions and land-based emissions because the NOx and SO2 emissions from ships and ammonia emissions from the land led to the formation of secondary ammonium nitrate and ammonium sulfate. The concentrations of ship-emitted EC and the ratio of PSO42- / EC in primary emissions (8.8 times) from ships were used to calculate the concentrations of ship-induced primary sulfate. The majority of the ship-induced sulfate was formed secondarily from oxidation of SO2, and only about one-third of it was from primary emission. The proportions of EC and organic aerosol (OA, the sum of POA and SOA) were relatively small due to a small amount of SOA formed from HC shipping emissions and the small fractions of EC and POA in PM emissions from ships. Moreover, we determined whether the concentrations of PPM or the secondary PM2.5 was affected more by shipping emissions. In all regions and cities, the PM2.5 pollution caused by shipping emissions was dominated by secondary species, with a mean contribution of 78.8 %. The mandatory fuel oil standard (0.5 % sulfur limit) in DECA would reduce the ship-induced sulfate and PPM concentrations, while our results illustrated that this control policy alone was not enough to reduce the total PM2.5 concentrations, particularly for nitrate and ammonium. After-treatment techniques should also be applied to the marine diesel engines to reduce NOx emissions, and more efforts should be made to reduce NH3 emissions from land-based sources, which is beneficial in reducing the secondary NO3- formation from shipping emissions.

The changes in the mean PM2.5 concentration caused by shipping emissions in each season (ΔCseasonal) were analyzed separately (Fig. 5). Large seasonal variations in peak concentrations and the extent of impacted areas could be observed, and the changing patterns in northern and southern areas were also quite different. The largest impact of shipping emissions in the north of the Yangtze River was predicted in summer, with ΔCseasonal more than 5 µg m-3 in most coastal areas of the Bohai Rim, the YRD and the surrounding areas of Qingdao. The impact during winter was less significant, with ΔCseasonal less than 2.5 µg m-3. The seasonal variations in southern areas presented an opposite trend. Ship emissions were predicted to increase the PM2.5 concentrations by more than 2 µg m-3 in spring and fall but there are fewer effects in summer. The longest MD (ΔCseasonal >0.1 µg m-3) was calculated in summer with the value of 1300 km, showing that the emissions of ships could almost affect the northernmost areas of China (Table 1), while when the threshold of ΔCseasonal increased to 0.5 and 1.0 µg m-3, there was larger MD in fall compared with those in other seasons, which indicated that central China, hundreds of kilometers away from the sea, was critically influenced by ship-related air pollution.

The differences in emission quantities and spatial distributions of vessels in the selected months alone could not explain the predicted seasonal variations. For example, in the YRD region, the emissions were larger and more concentrated in areas near the land in winter, but the contributions were still the lowest in the whole year. Therefore, these seasonal variations were probably related to the temporal changes of the meteorological conditions, particularly the direction of the prevailing winds, which was crucial for the diffusion and long-range transport of shipping emissions. Eastern China lies in the perennial monsoon region, and the summer monsoons could carry air pollutants related with shipping emissions to inland areas (Fig. 5c), while in winter, few contributions were observed in the BTH region and its surrounding areas, mainly due to the dominated wind from the northwest (Fig. 5a).

The contributions of shipping emissions to seasonal PM2.5 concentrations in coastal core regions and cities were further quantified (Fig. 6). The effects were most evident in summer for China and the entire coastal areas, which were 2.5 and 3.0 times higher than those in winter, respectively. In specific regions and cities, the great differences between the seasonal impacts could not be ignored. In summer, shipping emissions increased the PM2.5 concentrations in Qingdao, Dalian and Yantai by 9.4, 6.9 and 5.4 µg m-3, which were 16.4, 26.6 and 11.9 times larger than the smallest seasonal effects in winter, respectively, while for Zhuhai and Guangzhou, the seasonal impacts in spring were around 2 times higher than those in summer. Additionally, in almost all coastal regions and cities, the seasonal relative contributions of the shipping emissions to inland PM2.5 concentrations reached the peak in summer (2.2 %–18.8 %), indicating it played an important role in the inland PM2.5 air pollution during this period. It should be noted that although in the PRD region ΔCseasonal was the lowest in summer, the relative contribution of shipping emissions still remained at a high level. These were mainly because in eastern China, low emissions of land-based sources (e.g., fossil fuel combustion, biomass burning) and favorable meteorological conditions (e.g., large wet depositions, higher atmospheric mixing layer) in summer led to cleaner PM2.5 background level (Zhang and Cao, 2015).

Since the direction of the prevailing wind had an important role in the eventual impacts of emissions from ships on the inland air quality, relationships between their daily means in four cities most affected by shipping emissions were quantified (Fig. 7). The daily mean surface wind directions in the local area were calculated from WRF modeling results by vector-averaging. Based on the geographical location of the selected cities, the wind direction regarded as onshore was identified. To determine the daily impacted levels from shipping emissions, the annual averaged daily contribution of marine transport sectors to the PM2.5 concentrations was defined as a baseline value in each individual city. The results showed that coastal locations frequently experienced onshore winds that transported marine air over land. In Qingdao, Shanghai, Ningbo and Dalian, 48.0 %, 63.4 %, 48.8 % and 61.0 % of the days were considered as onshore flow days, respectively. On these days, higher relative contributions from the shipping sector to the inland air quality were observed. This was because the onshore wind could carry not only the ship-related air pollutants, but also the clean air from sea to land, which could cause a low PM2.5 background level in inland areas. During the onshore flow days, shipping emissions led to a ∼4.7 %–17.1 % increase on average in the inland PM2.5, about 1.8–2.7 times higher than that in rest of the days. The frequency of heavy shipping polluted days (daily relative contributions 1.5 times larger than the baseline) along with the onshore wind was 82.4 %–100.0 %, and simultaneously the inland daily PM2.5 concentrations increased by up to 20.3 %–38.1 % (10.5–26.8 µg m-3) due to shipping emissions. This revealed that in the period of frequent “onshore” winds, such as in summer for YRD and BTH regions, shipping emissions probably became one of the most important contributors to the air pollution. However, not all onshore wind would cause high contributions of ships, and it also depended on the spatial distribution of shipping emissions in surrounding marine areas. For example, during the period of westerly and northwesterly onshore flows in Dalian, the daily contributions were typically small, due to the relatively fewer emissions emitted from ships in the northern part of the Bohai Sea.

The contributions of shipping emissions from maritime areas within 0–12, 12–50, 50–100 and 100–200 Nm to the total ship-related PM2.5 in summer were identified, respectively, using the source-oriented CMAQ model (Fig. 8). Only PPM and SNA formed from shipping emissions were tracked, not including SOA, which was quite a small portion (discussed in Sect. 3.2), and we assumed that the sum of them represented the concentration of the total ship-related PM2.5. Overall, shipping emissions within 12 Nm were the dominant contributor with the contributions of 30 %–90 % over the mainland China and more than 80 % for coastal areas of three major city clusters. As the distance away from the land increased, fewer air pollutants from shipping emissions were transported to the inland areas, and effects of emissions within 100–200 Nm only accounted for less than 20 % in most inland areas.

Due to the differences in distributions of shipping emissions, the surface structure and meteorological conditions, the contributions of emissions from ships within each maritime area may be diverse for individual inland areas. In terms of the areas located in the north of the YRD region and south of Lianyungang, due to the major water traffic lanes far from land around A4 and the increase in shipping emissions around A3 in summer (Fig. 2), significant emissions from ships within 12–100 Nm led to a relative higher contribution (40 %–60 %) in these areas, when the definite seasonal contributions of shipping emissions to the local PM2.5 concentration occurred (approximately 7 %). Moreover, some middle parts of China, including Henan province, Hubei province and Hunan province, were also significantly affected by the long-range transport of shipping emissions away from the coastline with the distance more than 12 Nm, probably due to the benefits of low terrain heights. Our results implied that DECAs within 12 Nm were not large enough in the latitudinal direction to prevent the PM2.5 pollution from shipping emissions, particularly for YRD region and its surrounding areas, and should be expanded to at least 100 Nm to the coastline.