In this study, the WRF-Chem model (version 3.6.1; Grell et al., 2005) with updated gases-phase and heterogeneous-phase mechanisms of new reactive nitrogen species (L. Zhang et al., 2017) was used to simulate the transport, mixing, and chemical transformation of trace gases and aerosols. Briefly, the updated module was based on the default Carbon Bond Mechanism-Z (CBMZ) module (Zaveri and Peters, 1999), in which the O3 production came from the traditional photochemical mechanisms with only two gas-phase sources of HONO (OH + NO → HONO and HO2 + NO2 → HONO) and no chlorine chemistry. In the updated module, CBMZ-Reactive NitrOgen Mechnism (CBMZ-ReNOM), the HONO sources included the additional gas-phase reactions between NOx and HOx (OH + HO2); the heterogeneous reaction of NO2 on the particle, urban, leaf (Kurtenbach et al., 2001), and sea surfaces (Zha et al., 2014); and direct emission from vehicles (Kurtenbach et al., 2001; Gutzwiller et al., 2002; Sun et al., 2020). For this study, an additional HONO source from the photolysis of particulate nitrate (PNO3 → 0.67 HONO + 0.33 NO2) was included into our model. The photolysis rate constant of PNO3 (JPNO3) was calculated following the approach used in Fu et al. (2019): JPNO3 = (8.3×10-5/7×10-7) × JHNO3-WRF-Chem, where JHNO3-WRF-Chem is the photolysis rate constant of gaseous HNO3 calculated online in the WRF-Chem model. For ClNO2 production, the parameterization from Bertram and Thornton (2009) was used to represent the N2O5 uptake coefficient and ClNO2 production yield. This parameterization reproduced the order of magnitude and variation in the observed N2O5 and ClNO2 levels in a background site in Hong Kong (Dai et al., 2020). The other six chlorine species, with relevant photolysis reactions and subsequent reactions between released Cl radical and VOCs, were also added to the default module (L. Zhang et al., 2017). The details of other chemical and physical schemes for the simulation can be found in L. Zhang et al. (2017).

The model simulations were performed from 28 June to 31 July 2018. The first 72 h of the simulations was considered spin-up time. The initial meteorological conditions were provided by reanalysis data from the Final (FNL) Operational Global Analysis dataset provided by the National Centers for Environmental Prediction (NCEP; http://rda.ucar.edu/datasets/ds083.2/, last access: 1 November 2020). The model had 31 vertical layers with a fixed top of 100 hPa. The domain covered a large part of Asia with a horizontal resolution of 36 km × 36 km (Fig. 1a). The surface layer was 30 m above the ground, and the lowest 11 layers were approximately within the height of the planetary boundary layer at noon.

Five sets of emission inventories (EIs) were used for anthropogenic emissions in our study. For mainland China, the Multi-resolution Emission Inventory for China (MEIC; http://www.meicmodel.org/, last access: 1 November 2020) in 2016 was used. For the rest of Asia, we applied the MIX dataset (http://www.meicmodel.org/dataset-mix, last access: 1 November 2020) for 2010 (Li et al., 2017). For international shipping, the emission database in the Community Emission Data System (CEDS) (McDuffie et al., 2020) for 2017 was used. The HONO emissions from land transportation sources were calculated using their NOx emissions and the HONO/NOx ratio of 0.8 % for gasoline and 2.3 % for diesel. These commonly used ratios in model studies (Zhang et al., 2016; Fu et al., 2019) are based on the previous measurements of vehicle exhausts (Kurtenbach et al., 2001; Gutzwiller et al., 2002) and are generally consistent with more recent emission result (Liu et al., 2019; Trinh et al., 2017). For ship-emitted HONO, we set the emission ratio of HONO/NOx as 0.51 % based on the reported ratio in fresh ship plumes in Chinese waters (Sun et al., 2020). For anthropogenic chloride emissions, the high-resolution (0.1∘ × 0.1∘) EIs of HCl and fine particulate Cl- for 2014 were applied for mainland China (Fu et al., 2018). These EIs included four sectors and have been shown to offer a reasonable model simulation of particulate chloride by the WRF-Chem model (Dai et al., 2020). The Reactive Chlorine Emission Inventory (RCEI; Keene et al., 1999) was used for anthropogenic chloride emissions in the other regions. For natural emissions, the biogenic emissions were calculated by the Model of Emission of Gas and Aerosols from Nature (MEGAN) version 2.1 (Guenther et al., 2006).

The spatial distribution of ship NOx emissions is shown in Fig. 1a. The main ship routes with high emission intensity are clearly identified in the ship emission inventory. One major shipping lane is located in the southern Bay of Bengal (BOB) in the Indian Ocean, passes through the Strait of Malacca, and extends to the South China Sea (SCS) and other Asian countries. Distinct shipping lanes are also shown along the coast of the East China Sea (ECS) and the Sea of Japan (SOJ). Over the western Pacific Ocean (WPO), congested ship routes are distributed among Japan and other countries (Southeast Asian countries, Australia, and North America). Based on the distribution of ship NOx emissions, six water zones were selected, including three waters around China (SCS; ECS; Bohai Rim, BR), two waters in other regions (SOJ and BOB), and one open ocean (WPO; Fig. 1b). In addition, three densely populated city clusters were chosen (the North China Plain, NCP; the Yangtze River Delta, YRD; and the Pearl River Delta, PRD).

Eight simulations were conducted with different emissions and chemistry, as listed in Table 1. In the Def and Def_noship cases, the WRF-Chem model was conducted with default chemistry (i.e. the default CBMZ mechanism with only the two HONO sources and no chlorine chemistry). The differences between Def and Def_noship (i.e. Def - Def_noship) represent the effects of ship emissions with the default nitrogen chemistry. In the Cl and Cl_noship cases, an updated chlorine chemistry in the revised WRF-Chem model was used. The differences between the two cases (i.e. Cl - Cl_noship) represent the effects of ship emissions with the default and additional chlorine chemistry. Similarly, in the HONO and HONO_noship cases, the additional HONO chemistry was used, and the difference between the two cases (i.e. HONO - HONO_noship) represents the default impact of ship emissions with additional HONO chemistry. In the BASE and BASE_noship cases, the integrated HONO and chlorine chemistry are considered. The differences between BASE and BASE_noship represent the shipping impact with the integrated effects of HONO and chlorine species. The results from the BASE experiment will be used to validate the model performance.

Meteorological data from surface stations from the NOAA National Climatic Data Center (NCDC; Fig. S1a in the Supplement) comprising wind direction, wind speed, surface temperature, and specific humidity were used to validate model performance for the meteorological parameters. Conventional air pollutant data (NO2, PM2.5, and O3) from surface stations (obtained from China's Ministry of Ecology and Environment; Fig. S1b) were used to evaluate the simulated air pollutants over mainland China. Table S2 summarizes the statistical performance of our model results. For meteorological parameters, high R values (>0.85) and low mean bias (MB) values indicate good performance for the meteorological field. For regular air pollutants, the model overpredicted PM2.5 (MB = 10.6 µg m-3) and slightly underpredicted NO2 (MB = 3.3 ppbv) and O3 (MB = 5.5 ppbv). These biases in simulation can be partially explained by uncertainties in the model input, such as the land-use data (Dai et al., 2019) and emission inventory (Li et al., 2017).

The O3 data from two remote sites, Ryori and Yonagunijima (Yona) in Japan (https://www.data.jma.go.jp/ghg/kanshi/ghgp/o3_e.html, last access: 1 November 2020) and one coastal background site in Hong Kong, Hok Tsui (HT), were used to compare the model performance over the maritime areas. These three sites are located along the coasts of the SCS, ECS, and WPO regions (Fig. 1b). As shown in Fig. S2, with the default chemistry, the model underpredicted the O3 mixing ratio at the coastal and marine sites, with an underestimation of 2.8, 4.8, and 2.3 ppbv at the HT, Ryori, and Yona sites respectively (Table S3). With the addition of the HONO and ClNO2 chemistry, the simulated O3 levels at the marine sites were improved, with an MB of -2.8 to -1.5 ppbv at the HT site, -3.0 to -2.3 ppbv at the Ryori site, and -2.3 to -0.7 ppbv at the Yona site.

Figure 2a shows the horizontal distribution of the average HONO at the surface layer in the BASE case. The predicted HONO was widespread over the oceans, with mixing ratios ranging from 0.005 to 0.300 ppbv and distinct higher concentrations along the main shipping lanes. The distribution of HONO was consistent with that of NO2 (Fig. S3) due to the heterogeneous conversion of NO2 to form HONO and direct HONO emission from ships. In the vertical direction, the simulated HONO was concentrated at the surface and reached up to 400–600 m in the coastal and marine areas (see Fig. S4). Figure 3 shows the vertical profile of HONO from ship emissions in the nine selected regions. Consistent with the overall HONO vertical distribution, ship-contributed HONO also peaked at the surface in the oceanic areas, with average HONO levels of 3–120 pptv in the marine boundary layer (MBL). The greatest contribution of ship emissions was simulated in the WPO (96 %), followed by the SOJ (80 %), the BOB (49 %), the three Chinese waters (14 %–16 %), and the coastal cities (3 %–12 %). The varying ship contributions in these regions can be explained by the relative strength of the emissions from ships and from the adjacent land areas.

High values of ClNO2 were simulated along the coasts and peaked in the lower MBL (Fig. 2b), with mixing ratios ranging from 2 to 400 pptv in oceanic areas and the highest value in the BR region. This distribution was in line with that of its precursors, N2O5 (Fig. S5) and particulate chloride (Fig. S6). Vertically, the peak value of ClNO2 was simulated in the residual layer (100–300 m; Fig. 3), with mixing ratios of 8–350 pptv in oceanic areas. Similar to HONO, the greatest ship contribution to ClNO2 was also simulated in the WPO (61 %), followed by other oceanic areas (9 %–24 %) and coastal cities (5 %–11 %).

We compared the modelled HONO and ClNO2 with field observations made at some coastal and marine sites (see Table S4). The simulated HONO mixing ratios were 0.1–0.3 ppbv and 0.01–0.1 ppbv over the respective coastal BR and SCS sites and were comparable to the measurements at the marine sites of the BR (0.2 ppbv) (Wen et al., 2019) and SCS (89 pptv) (Table S4). For the coastal areas of other Asian countries, the simulated HONO compared well with the measurements in South Korea (0.60 ppbv) (Kim et al., 2015) and Japan (0.63 ppbv) (Takeuchi et al., 2013). HONO was simulated at approximately 5 pptv in the open ocean and 10–25 pptv along the main shipping lanes (Fig. 2a), which are values comparable to the measured HONO (3–35 pptv) in the open ocean in Europe and North America (Meusel et al., 2016; Kasibhatla et al., 2018; Ye et al., 2016). For ClNO2, the order of magnitude and variation in the measured N2O5 and ClNO2 levels at the HT site were reasonably reproduced by our model for early autumn of 2018 (Dai et al., 2020). The model performance of HONO and ClNO2 in the land areas of mainland China for the summer of 2014 was also evaluated by L. Zhang et al. (2017). Overall, our model ability with respect to simulating HONO and ClNO2 is acceptable, and the model results are sufficiently reliable for further analysis.

In this section, we evaluate the ship effects on the main atmospheric radicals (ROx, OH + HO2 + RO2), O3, and PM2.5 with the default chemistry (described in Sect. 2.3) and with the additional HONO and chlorine chemistry.