The in-situ measurement was conducted during the period from 1 to 31 May 2024. The sampling location was the Dongzhen Reservoir (DZSK, 118.98° E, 25.48° N), which was categorized as a suburban site in Putian. The DZSK site was 6 km from downtown of Putian and approximately 25 km from the Taiwan Strait. The geographical location of DZSK is shown in Fig. 1. Hourly HONO concentrations were measured using a long-path absorption photometer (LOPAP). The other two gaseous air pollutants, NO₂ and O₃, were measured using a commercial chemiluminescence instrument (Thermo Fisher Scientific 17i and 49i). For mode 17i, we acknowledge that the chemiluminescence instrument with a molybdenum converter used for NO₂ measurements may be subject to positive artifacts from other reactive nitrogen species, which represents a potential source of uncertainty in the NO₂ data used for model evaluation (Dunlea et al., 2007). Air temperature and relative humidity were monitored by a micro automatic weather station (LUFFT). Ultraviolet radiation (UV-A) was simultaneously observed with a solar radiation instrument (Kipp & Zonen). As records of the hourly wind field and precipitation were missing for the study period, observations were obtained from alternative sources. For wind speed and direction, we collected the U10 and V10 variables, representing wind components at 10 m, from the ERA5 reanalysis dataset archived on the European Centre for Medium-Range Weather Forecasts (ECMWF) platform. Additionally, we obtained the precipitation data observed by the regional meteorological station in Putian (119.00° E, 25.44° N) for further analysis.

WRF-Chem is one of the most widely used regional chemical transport models in atmospheric chemistry modeling (Zhang et al., 2024a). It can simulate the evolution of ambient trace gases and aerosols with fully coupled meteorology–chemistry feedback (Grell et al., 2005). This study used WRF-Chem (version 4.1.5) to explore coastal HONO formation and the associated environmental influences. To align with the measurement campaign, the model was run from 26 April to 31 May 2024. The first five days were reserved for spin-up and were excluded from the data analysis. The entire WRF-Chem simulation was manually restarted every seven days to reduce the influence of the accumulation of biases from the numerical computation. The modeling domain focused on the southeastern coastal region of China (Fig. 1), with a grid spacing of 10 km. Fujian Province is highlighted in a blue dashed box in Fig. 1a. The coastal regions of Fujian consist of plains with a terrain elevation of less than 500 m. Central Fujian is occupied by longitudinal mountains. This study focuses on six coastal cities for further discussion. As shown in Fig. 1b, these cities are, from north to south: Ningde (ND), Fuzhou (FZ), Putian (PT), Quanzhou (QZ), Xiamen (XM), and Zhangzhou (ZZ). The observational site DZSK is located in the suburbs of Putian. In addition, 30 vertical layers were set up, extending from the ground surface to a height of 50 hPa.

The initial and lateral meteorological input conditions were derived from the final (FNL) global reanalysis dataset of the National Oceanic and Atmospheric Administration (NOAA). We also utilized four-dimensional data assimilation (FDDA) to constrain gridded surface and upper atmospheric parameters with extra weather observational datasets. Continental anthropogenic emissions were obtained from the Multi-resolution Emission Inventory for China (MEIC version 1.4) (Li et al., 2017; Zheng et al., 2018). Shipping emissions were obtained from the Shipping Emission Inventory Model (SEIM), which was developed by Tsinghua University (Liu et al., 2016). Figure S1 in the Supplement illustrates the spatial distribution pattern of continental and oceanic shipping emissions within the modeling domain. Biogenic emissions were calculated online using the Model of Emissions of Gases and Aerosols from Nature (MEGAN, version 2.0.6) (Guenther et al., 2006). Furthermore, this study adopted various parameterizations to predict sub-grid atmospheric physical and chemical processes. The detailed information was compiled in Table 1. The Statewide Air Pollution Research Center (SAPRC99) mechanism (Carter, 2000) and the Model for Simulating Aerosol Interactions and Chemistry (MOSAIC) (Zaveri et al., 2008) were selected for gas-phase and aerosol chemistry simulations, respectively. The Fast-J module was chosen to calculate the photolytic rates of trace gases (Wild et al., 2000).

In the original WRF-Chem model, the default source of HONO formation is the homogeneous reaction between NO and OH radicals. In a previous study, we incorporated multiple additional HONO sources into the WRF-Chem model (Zhang et al., 2024b), including direct emissions from fuel combustion processes, photo- and dark-oxidation of NO_x, light-enhanced heterogeneous uptake of NO₂ on the ground and aerosol surfaces, and photolysis of particulate nitrate aerosols. Updates to the chemical parameterizations of the HONO sources are summarized in Table 2. The NO₂ heterogeneous uptake on ground and aerosol surfaces was parameterized as a light-dependent process, with uptake coefficients (γ) chosen in accordance with previous studies in China (Zhang et al., 2021, 2024b; Liu et al., 2019). The base nighttime uptake coefficients of NO₂ were set to 8×10-6 for the ground surface and 4×10-6 for aerosol surfaces. During the daytime, these values were dynamically increased with solar radiation using a linear equation, reaching their maximums of 6×10-5 and 1×10-3, respectively (Liu et al., 2019). Sa/V and Sg/V are aerosol and ground surface area densities (m² m⁻³), respectively. Sa/V could be calculated through the MOSAIC aerosol module, which categorized different types of aerosols into four size bins ranging from 3.9 nm to 10 µm, i.e. 0.039–0.156, 0.156–0.625, 0.625–2.500 and 2.500–10.000 µm (Zaveri et al., 2008). Sg/V was derived based on the underlying surface category. In vegetation grid cells, Sg/V was estimated as the ratio of the two-fold of leaf area index (LAI, m² m⁻²) to the model height of the first layer (Zhang et al., 2016). For urban areas, the ground surface area density Sg/V was empirically set from 0.1 to 0.3 depending on the fraction of urban area using a linear formula (Zhang et al., 2024b). It is noted that the model only accounts for heterogeneous uptake of NO₂ on ground surface in the first layer, while the reaction on aerosol surfaces occurs in all model layers. In addition, two key parameters have been updated in this study to more accurately represent the HONO formation rate in typical coastal regions. The first parameter is the ratio of HONO to NO_x (HONO / NO_x), which reflects emission intensity. Here, we utilized the value of 1.45 % proposed by Hu et al. (2022) based on long-term in situ measurements in Xiamen, Fujian. This value more accurately represents direct HONO emissions from local fuel combustion. Simultaneously, we estimated HONO emissions from the shipping sector using the same ratio. Secondly, we set the photolysis frequency of nitrate aerosols (JNO3-) to 120 times that of gaseous HNO₃ (Fu et al., 2019; Ye et al., 2016, 2017).

We used a source-oriented method (SOM) to determine the proportion of HONO production and consumption processes contributed by each source. A detailed description of the SOM diagnostic module can be found in Zhang et al. (2024b). In the SOM analysis, seven variables were configured to trace the formation process of HONO. These variables included the homogeneous reaction between NO and OH (NO + OH), NO_x photo-oxidation, NO_x dark-oxidation, primary emissions, the heterogeneous uptake of NO₂ by aerosols (Hetero-aerosol), the heterogeneous uptake of NO₂ on the ground surface (Hetero-land) and the photolysis of nitrate aerosols (Nitrate-photolysis). The first three sources are categorized as gaseous reaction, while the latter three are categorized as surface reactions. We also quantified two HONO chemical sink pathways: photodissociation of HONO (HONO + hν) as well as OH-oxidation removal (HONO + OH). Additionally, it should be noted that dry deposition, an important sink for HONO especially at night, is calculated within the standard WRF-Chem deposition module but was not explicitly tracked in this chemical budget analysis since our focus here was on chemical pathways. Similarly, we also set up the SOM analysis to investigate the formation of OH radicals. Five formation pathways were identified, including the gas-phase reaction between HO₂ and NO (HO₂ + NO), the reaction between atomic oxygen (O¹D) and H₂O (O¹D + H₂O) initiated by O₃ photolysis (O₃ + hν → O¹D + O₂), photolysis of HONO (HONO + hν), photolysis of hydrogen peroxide (H₂O₂ + hν), and ozonolysis of VOCs (O₃ + VOCs). All of these chemical reactions are primary sources of OH radical formation except HO₂ + NO, which is generally considered as a secondary conversion (Xue et al., 2025).

Our study also conducted sensitivity experiments to quantify the contribution of oceanic shipping emissions to the formation of HONO in coastal areas, and the contribution of HONO chemistry to the enhancement of ambient O radicals and O₃. In total, three simulation cases were run, using the same meteorological initial and lateral conditions and continental anthropogenic and biogenic emissions. As listed in Table S1 in the Supplement, the first case was the BASE case, which considered oceanic shipping emissions but did not include the updated HONO sources. The REV case included both oceanic shipping emissions and the updated HONO formation pathways. The “Noship” case denotes a simulation that used an improved HONO source representation but excluded oceanic shipping emissions. By comparing the BASE case (or Noship) with the REV case, we estimated the respective contribution of the updated chemical sources and shipping emissions to HONO formation.

Figure 2 shows hourly observations of air pollutants and meteorological parameters at the DZSK site from 1 to 31 May 2024. Generally, HONO concentrations ranged from 0.01 to 1.11 ppbv, with an average of 0.23 ppbv. We compared the HONO measurements at the DZSK site with in situ observations collected from previous studies in China. Table S2 presents a dataset of HONO measurements from 43 studies conducted between 2008 and 2025. It can be concluded that observational studies were intensively carried out in urban areas of the NCP region, particularly in Beijing, where recorded HONO concentrations frequently exceeded 1 ppbv. These higher HONO levels were attributed to abundant NO_x emissions. By contrast, HONO concentrations were found significantly lower in remote regions. For instance, HONO concentrations ranging from 0.13 to 0.15 ppbv were measured at the summit of Mt. Tai (Jiang et al., 2020; Xue et al., 2022). Studies of coastal regions witnessed HONO concentrations of around 0.50 ppbv (Hu et al., 2022; Zhong et al., 2023). Overall, the HONO levels measured at the DZSK site were much lower than those reported by previous measurements in China. The mean NO₂ concentration during the study period was 3.2 ppbv, due to low NO_x emissions in this suburban location, which probably contributed to the low HONO concentrations. Meanwhile, the average wind speed was 2.1 m s⁻¹, providing a favourable diffusion condition for air pollutants. The prevailing wind direction at the DZSK site was from the northeast, implying a potential contribution from oceanic shipping emissions.

Figure S2 exhibits the diurnal variations of gaseous air pollutants and meteorological parameters. Based on observations of the regional meteorological station in Putian, we categorized any hour with non-zero recorded precipitation into rainy days. On non-rainy days, air temperature and solar radiation were higher, with an apparent peak occurring around midday. Substantial sunlight accelerated the photochemical production of air pollutants such as O₃. Consequently, concentrations of air pollutants were obviously higher on non-rainy days than on rainy days. At the same time, precipitation also led to the wet deposition of air pollutants. To avoid the influence of precipitation, we focused on the diurnal variation pattern of HONO concentrations excluding rainy days. The maximum hourly HONO concentration at the DZSK site occurred at 14:00 LT on non-rainy days, peaking at 0.38 ppbv. This phenomenon is contrary to the diurnal variations of HONO observed in previous inland measurements. According to Wang et al. (2025), HONO concentrations in inland urban and rural areas typically display the minimum in the afternoon. Despite the decrease in air temperature and radiation after 14:00 LT, the O₃ concentration increased until it reached a daily maximum of 64.8 ppbv at 16:00 LT. The lag of O₃ peak concentration could be attributed to the strengthened AOC conditions resulting from the simultaneous high HONO levels. The chemical production of O₃ is a cumulative process that takes several hours so that O₃ levels could remain high even as solar radiation begins to decline, leading to the O₃ peak occurring later in the afternoon. To explore the mechanisms of HONO formation in coastal regions, especially the diurnal peak during the daytime, our study conducted SOM process analyses in the subsequent sections. The potential contribution of shipping emissions to coastal HONO formation was also quantified using sensitivity experiments.

First, we conducted a model evaluation based on measurements at the DZSK site. Figure 3 demonstrates a comparison between the simulated and observed HONO concentrations. The BASE simulation failed to capture the magnitude and temporal variations of HONO concentrations at the DZSK site. Including multiple HONO formation pathways greatly improved the model performance of HONO (Fig. 3a). The monthly mean HONO concentration increased from 0.03 ppbv (BASE) to 0.25 ppbv (REV). We then applied three statistical metrics to reveal improvements in HONO modeling. The calculation of validation indicators has been elucidated in our previous modeling studies (Zhang et al., 2024b). As summarized in Table 3, while the improvement in the Index of Agreement (IOA, varies from 0 to 1) is modest (from 0.62 to 0.69), the revised model shows a fundamental improvement in capturing the magnitude of HONO concentrations. This is demonstrated by the dramatic enhancements in the Normalized Mean Bias (NMB, varies from -∞ to +∞), which improved from -86 % to +8 %, and the Root Mean Square Error (RMSE, varies from -∞ to +∞), which decreased by 21 %. While the revised model reasonably reproduced the observed temporal variations in HONO concentrations during the study period, an underestimation existed on 16–18 May, suggesting a potential omission of HONO sources. The systematic overestimation during 21–25 May corresponds to a period of continuous rainfall. To provide an evaluation focused on the normal conditions, we also calculated the statistics for non-rainy periods only, where the model performance improved further (IOA = 0.70, NMB = -5 %, RMSE = 0.21 ppbv). Simulations including the updated HONO sources more accurately represented the diurnal variation pattern of HONO concentrations. Figure 3b illustrates that the REV case successfully captured the higher HONO concentrations observed around noon. The Pearson's correlation coefficient (R) between the measurements and simulations increased from 0.657 (BASE) to 0.763 (REV). The REV simulation accurately captured the timing of the observed diurnal peak around 14:00 LT, which the BASE case simulates several hours too early. We further show the comparisons between measurements and simulations in terms of diurnal variations of HONO and NO₂ on rainy and non-rainy days. As shown in Fig. S3, the daytime HONO concentrations simulated by REV is in good agreement with observations on non-rainy day. However, the model fails to reproduce the observed low concentration levels and the temporal variations of HONO on rainy days, possibly attributed to an underestimated wet removal caused by precipitation. Meanwhile, daytime NO₂ concentrations on rainy days also shows an obvious overestimation, further leading to the higher chemical production of HONO.

The model performance for important precursors and products of HONO chemistry, NO₂ and O₃, were improved as well. As illustrated in Fig. 4a and c, the diurnal variations of the simulated NO₂ and O₃ concentrations by REV were more consistent with the observed values. The diurnal correlation coefficients between measurements and simulations for NO₂ and O₃ increased by 5 % and 3 %, respectively. At the same time, underestimation of O₃ concentrations in the original model was effectively reduced. The daily maximum O₃ concentration in the revised version (REV) was much closer to the observed value than in the BASE simulation. The estimated maximum daily average 8 h (MDA-8) concentration of O₃ increased from 33 ppbv (BASE) to 45 ppbv (REV), with an observed value of 53 ppbv (Fig. 4d). Accurately characterizing the HONO chemical budget therefore improves the simulation of O₃ concentrations in 3D chemical transport models. The updated HONO chemistry had a limited net impact on NO₂ concentrations due to several competing chemical pathways. Specifically, the consumption of NO₂ by the newly added HONO formation reactions was largely offset by enhanced NO-to-NO₂ conversion from increased OH radicals and altered O₃ titration, resulting in only a minor change in the overall NO₂ budget.

Several uncertainties exist in the HONO simulations presented in this study, firstly related to the parameterization of key chemical pathways. These are mainly concentrated in the heterogeneous uptake coefficients and the nitrate aerosol photolysis rate. For the nitrate photolysis frequency, Zhang et al. (2022a) summarized that this value is approximately 1-3 orders of magnitude higher than the photolysis frequency of HNO₃. Our study adopted a median value of this range (120 JHNO3), which was inferred based on aircraft measurements in the North Atlantic marine boundary layer and has been widely used in previous studies (Fu et al., 2019; Ye et al., 2016; Zhang et al., 2021). For the heterogeneous uptake of NO₂ on solid surfaces, the dimensionless uptake coefficient typically ranges from 10⁻⁶ to 10⁻³. For the ground surface, we also applied a representative median value, with this coefficient varying from a nighttime baseline in the 10⁻⁶ range up to a maximum of 6×10-5 under peak sunlight (Wang et al., 2025). For the aerosol surfaces, we set the maximum daytime value to 1×10-3, an upper limit reported in the literature, which carries a potential uncertainty. To quantitatively assess the impact of this choice, we conducted an additional sensitivity analysis where the uptake coefficient γa was reduced to 1×10-4 and 1×10-5, respectively. To minimize the computational burden, these simulations were performed for the first seven days of our study period. As shown in Table S4, model results show that while lowering the γa value does lead to a corresponding decrease in the HONO production rate from the heterogeneous uptake of NO₂ by aerosols, the impact on the overall HONO budget is negligible. The average daytime HONO concentration decreased by less than 2 pptv, a relative change of less than 1 %. This finding provides quantitative support for our argument that due to the low aerosol abundance in this coastal region, the heterogeneous aerosol pathway contributes minimally to HONO formation, regardless of the precise γa value. Consequently, the responses in O₃ and OH concentrations were also minimal. This confirms that our use of 1×10-3 as an upper-limit for γa is a reasonable choice and does not compromise the main conclusions of our study.

Secondly, we used an empirical HONO / NO_x value of 1.45 % to estimate the direct HONO emissions from the fuel combustion process. This value was derived from long-term in situ measurements in coastal regions and proposed by Hu et al. (2022). We applied this value to describe fresh HONO emissions from ships. However, Ke et al. (2025) have pointed out that the same ratio of HONO to NO_x may not be applicable to mobile sources, including mobile machinery, ships, and aircraft. The ratios of HONO to NO_x were examined to be 0.8 %–6.0 % and 2.3 % for aircraft and offroad mobile machinery, respectively. Currently, there are no direct measurements of freshly emitted HONO from shipping. Therefore, further observations and experiments are required to constrain HONO emissions from ships. In addition to mobile emissions for HONO from combustion processes, the soil volatilization could be another important contributor to ambient HONO (Tan et al., 2023; Wu et al., 2022), especially when there are large areas of cropland and agricultural activities. The neglect of soil HONO emissions may introduce some uncertainties in the HONO simulation in this study.

Thirdly, the ocean surface may affect HONO formation in other ways. Zha et al. (2014) observed significant evidence suggesting that the ocean was a nonnegligible contributor to coastal HONO production. Consequently, Zhang et al. (2016) parameterized this reaction as heterogeneous uptake by the ocean surface in the WRF-Chem model using a simplified formula. Additionally, an observational study proposed that seawater could increase the solubility of HONO under certain conditions, indicating that the ocean surface could act as a sink for atmospheric HONO (Crilley et al., 2021; Wang et al., 2025). However, the source or sink effect of the ocean on HONO remains a controversial topic. Further studies on this theory and the relevant parameterization are needed in the future.