Simulated photochemical response to observational constraints on aerosol vertical distribution over North China

The significance of aerosol–photolysis interaction (API) in atmospheric photochemistry has been emphasized by studies utilizing box models and chemical transport models. However, few studies have considered the actual aerosol vertical distribution when evaluating API effects due to the lack of observations and the uncertainties in model simulation. Herein, we integrated lidar and radiosonde observations with the chemical transport model (GEOS-Chem) to quantify the response of photochemistry to observational constraints on aerosol vertical distribution across different seasons in North China. The underestimation of aerosol optical depth (AOD) in lower layers and the overestimation in upper layers in GEOS-Chem model were revised. In response, photolysis rates changed following AOD, showing 33.4 %–73.8 % increases at the surface. Surface ozone increased by an average of 0.9 and 0.5 ppb in winter and summer, and the default API impact on ozone reduced by 36 %–56 %. The weaker response in summer can be related to the compensatory effects of stronger turbulence mixing in the boundary layer. The long-lasting underestimation of ozone levels in winter was also greatly improved. Due to the enhanced photochemistry, PM_2.5 increased by 0.8 µg m⁻³ in winter and 0.2 µg m⁻³ in summer and increased strongly during pollution events, with a maximum daily change of 16.5 µg m⁻³ at Beijing station in winter. The weakened API effect in turn enhanced nitric acid formation by increasing atmospheric oxidizing capacity (13.5 % increase for OH radical) in high NO_x emission areas, and this helps explain the strong response of PM_2.5 in winter.

1 Introduction

China has witnessed a significant reduction in fine particulate matter (PM_2.5) concentration over the past decade (Zhang et al., 2019b; Xiao et al., 2022), while ozone levels increased (Lu et al., 2018; Liu et al., 2023). Ozone pollution has emerged as a critical air quality issue in China (Wang et al., 2022), especially in North China (Li et al., 2020), and is now spreading into the colder seasons (Li et al., 2021a). Many studies concluded that the rapid decline in aerosol concentrations is a key factor contributing to the increasing ozone trend (Liu and Wang, 2020; Wang et al., 2020b; Li et al., 2021b; Ma et al., 2021; Shao et al., 2021; Liu et al., 2023), highlighting the urgent need for further investigation on the interactions between aerosols and ozone.

Surface ozone is produced through the photochemical reactions of volatile organic compounds (VOCs) and nitrogen oxides (NO_x), in which aerosol can also play an important role. Ozone chemistry can be affected via the heterogeneous uptake of reactive gases on aerosol surfaces (Li et al., 2017, 2018a, 2019b; Ivatt et al., 2022). Furthermore, aerosols can absorb or scatter solar radiation, thereby influencing ozone concentrations by both altering meteorological conditions through aerosol–radiation feedback (ARF) (Xing et al., 2017; Qu et al., 2020; Zhao et al., 2023) and directly changing the photolysis rates that is referred to as the aerosol–photolysis interaction (API). Overall, whether during multi-pollutant air pollution episodes (Yang et al., 2022) or under seasonal mean conditions (Li et al., 2024), the impact of API on surface ozone can overwhelm that of ARF. As such, we will focus on the API process in this study.

Early studies have confirmed the significance of API, employing either observation-based methods or chemical transport models. Dickerson et al. (1997) reported that scattering aerosols can enhance ultraviolet radiation and accelerate photochemical reactions, while absorbing aerosols have the opposite effect. Subsequent studies further investigated the impact of API during different types of pollution around the world, such as biomass burning (Mok et al., 2016; Baylon et al., 2018; Li et al., 2018b), wildfire (Jiang et al., 2012), dust events (Kushta et al., 2014; Li et al., 2017), and anthropogenic pollution (Li et al., 2011, 2018a; Gao et al., 2020; Wu et al., 2020; An et al., 2021; Yang et al., 2022). In the highly polluted areas of China, API can lead to a 5 %–12 % decrease in surface ozone during anthropogenic pollution episodes in warm seasons (Li et al., 2011; Yang et al., 2022) and a 5 %–20 % decrease in cold seasons (Li et al., 2018a; Gao et al., 2020; Wu et al., 2020). The effects of long-term aerosol changes on photolysis have also been estimated in China. Using the Troposphere Ultraviolet and Visible (TUV) radiation model, Zhao et al. (2021) found that the photolysis rate of nitrogen dioxide (J[NO₂]) in North China increased by $\mathrm{1.3}\times {\mathrm{10}}^{-\mathrm{4}}$ s⁻¹ yr⁻¹ from 2013 to 2019 due to the aerosol declines. Consequently, the ozone concentration in North China increased by approximately 1 ppb during the period of rapid decline in aerosol concentration (2013–2017) due to API, as indicated by WRF-CMAQ simulations (Liu and Wang, 2020). This impact of API on long-term ozone trends was also confirmed using the box model approach (Wang et al., 2020b; Ma et al., 2021).

What's more, API can alter the atmospheric oxidizing capability and suppress secondary aerosol formation, resulting in changes in PM_2.5 concentrations. For example, API reduced PM_2.5 concentrations by 2.5 % to 10 % in the North China Plain during wintertime pollution events (Wu et al., 2020) and by 2.4 % in Delhi during the crop residue burning period (Chutia et al., 2024). Li et al. (2024) reported that the impact of API could decrease PM_2.5 by 5.4 µg m⁻³ in winter but increase PM_2.5 by 2.5 µg m⁻³ in summer over eastern China. These results indicate that the effect of API on air quality changes should not be overlooked both in short episodes and in the long-term period.

The key challenge in quantifying the impact of API on photochemistry is to reasonably represent the vertical distribution of aerosols since their effects vary across different altitudes (Jacobson, 1998; Liao et al., 1999). Shi et al. (2021) input observed PM_2.5 and black carbon profiles into an optical model and a one-dimensional radiative transfer model, demonstrating that scattering aerosols can suppress photochemical reactions in the lower boundary layer but enhance them in the upper boundary layer. However, other observation-based studies using radiative transfer models and box models often rely on empirical vertical distributions of aerosols, which typically follow an exponential profile. Pietruczuk et al. (2022) compared four different aerosol vertical distribution schemes and concluded that the choice of different schemes and the variations of aerosol column depth (AOD) have a comparable impact on photolysis rates. More importantly, the actual vertical distribution of aerosols is often highly heterogeneous and far more complex than an exponential distribution (Hu et al., 2020; Sun et al., 2023).

Previous modelling-based studies on the impact of API primarily compared the simulated surface PM_2.5 concentrations and AOD to observations without validating the aerosol vertical profiles. As such, using simulated profiles can introduce great uncertainty in the estimation of API effect as chemical transport models (CTMs) still struggle to reproduce the vertical distribution of aerosols accurately (Kim et al., 2015b; Huszar et al., 2020; Zhai et al., 2021; Lu et al., 2023). Our previous work revealed that the Goddard Earth Observing System Chemical Transport Model (GEOS-Chem) tends to underestimate the aerosol extinction coefficient (AEC) by approximately 30 % in the lower atmosphere during both summer and winter, while overestimating it by 30 % to 60 % at higher altitudes in summer (Chen et al., 2024). Chen et al. (2022) verified the simulated black carbon vertical profile from WRF-Chem and found an overestimation of 87.4 % at the surface level, as well as an underestimation of 14.9 % between 300 and 900 m. Similar bias in PM_2.5 profile simulations has also been reported by Liu et al. (2021a). Introducing the urban canopy model (Kim et al., 2015b) or increasing minimum turbulent eddy diffusivity (Liu et al., 2021a) can improve the simulation of surface aerosol concentration, but the overestimation in the upper layer remained. Overall, it is urgent that API effects are better understood by considering the actual aerosol vertical distributions.

In this study, we used continuous aerosol vertical distributions observed by ground-based lidar and radiosonde, along with the GEOS-Chem model, to investigate the response of photochemistry to observed constraints on aerosol vertical distributions. Based on these observations, we modified the vertical distribution of aerosols in the photochemical module of GEOS-Chem over North China. The responses of photolysis rates, ozone, and PM_2.5 to observational constraints on aerosol vertical distribution through API across different seasons were quantitatively assessed by comparing control and sensitivity simulations. A description of the observations and model settings is presented in Sect. 2. Section 3 outlines the existing biases in the model and provides a detailed account of how the aerosol vertical distribution was modified. Section 4 discusses the responses of photolysis, ozone, and PM_2.5 to this modification and their underlying mechanisms, followed by conclusions in Sect. 5.

2 Observations and model

2.1 Observations

To constrain the aerosol vertical distribution in the model, we utilized consecutive observations of aerosol extinction coefficient (AEC) from a ground-based Mie-scattering lidar located at the Institute of Atmospheric Physics, Chinese Academy of Sciences (39.982° N, 116.385° E) in Beijing (Fig. 1a). The lidar scans every 15 min, and we employed quality-controlled AEC data at the wavelength of 532 nm with a vertical resolution of 30 m. Further details about the lidar system and data processing can be found in Wang et al. (2020a). Data from January and July 2017 were used to represent winter and summer seasons, respectively.

Figure 1 (a) Spatial distribution of lidar and radiosonde stations. BJ: Beijing; BD: Baoding. The black rectangle shows the region where vertical distribution modification was performed. Vertical distribution of aerosol extinction coefficient (532 nm) from CALIPSO on (b) 25 January 2017 and (c) 11 July 2017. The corresponding CALIPSO overpassing tracks are marked with red and blue lines in (a).

To verify the results constrained by lidar, we also utilized vertical PM_2.5 concentration profiles from radiosonde observations conducted in Baoding (Fig. 1a). The radiosonde data were collected every 3 h from 17 June to 7 July 2019, at Baoding station (38.783° N, 115.5° E) with an altitude of 19 m. This radiosonde measurement provides high-resolution boundary layer profiles containing PM_2.5 and PM₁₀ concentrations, temperature, relative humidity (RH), and wind. A detailed description of the design of the radiosonde observations and the instruments used can be found in Li et al. (2023). Surface observations of hourly NO₂, ozone, and PM_2.5 concentrations from the Ministry of Ecology and Environment (MEE) network were accessed at http://quotsoft.net/air (last access: 31 August 2023; Wang, 2023) for the model evaluation.

2.2 GEOS-Chem simulation

The GEOS-Chem model version 13.3.3 was used in this study, employing a nested-grid simulation with a horizontal resolution of 0.5° × 0.625° over the region of 15–55° N, 70–140° E. Vertically, we selected the scheme with 47 layers, where the first 8 layers are situated below 1 km, and the first 14 layers are below 2 km. MERRA-2 meteorological data from NASA and chemical boundary conditions from a global simulation with a resolution of 2° × 2.5° were used to drive the nested simulations. The boundary layer mixing, dry deposition and wet deposition schemes, and emissions were consistent with those used in our previous work (Chen et al., 2024). The standard chemistry simulation with fully coupled ozone–NO_x–hydrocarbon–aerosol chemistry mechanisms was employed in this study. In GEOS-Chem, the photolysis rates are calculated by the updated Fast-JX v7.0 scheme (Eastham et al., 2014), which accounts for the presence of aerosols, clouds, and absorbing gases. Developed from the Fast-J radiative transfer algorithm (Wild et al., 2000), FAST-JX provides a full scattering calculation for 18 wavelength bins covering 177–850 nm. The photolysis rates are determined by the resolved flux and cross sections of different species in each wavelength bin. The impacts of aerosols, including sulfate–nitrate–ammonium (SNA), organic aerosol (OA), black carbon (BC), sea salt, and dust, on photolysis rates are closely related to aerosol optical properties, which are calculated using the Mie code based on the simulated three-dimensional aerosol concentrations. Here, we modified the aerosol vertical distribution with observational constraints after the calculation of AOD, and more details are presented in Sect. 3.

The simulation periods included January and July 2017 for the lidar observation time frame and June–July 2019 for the radiosonde observation time frame. For each period, we performed three sensitivity experiments to examine the impact of revised aerosol vertical distribution on photochemistry, as detailed below:

1. BASE: the baseline simulation, which includes the API with the original aerosol vertical distribution in the model;

2. REGION: same as the BASE simulation but incorporating observational constraints on aerosol vertical distribution over North China (37–42° N, 114.375–118.125° E), wherein only the aerosol vertical distribution below 3 km was modified in the calculation of photochemical rates based on lidar and radiosonde observations;

3. API_OFF: same as the BASE simulation but with the effect of API turned off in the photochemical module.

By comparing the BASE and REGION simulations, we can quantify the photochemical response to observational constraints on aerosol vertical distribution over North China. The role of all aerosols in photolysis can be estimated by examining the differences between BASE and API_OFF simulations.

3 Revisiting model aerosol vertical distribution based on observational constraints

3.1 Existing biases in model aerosol vertical distribution

The model performance in simulating surface air pollutant concentrations and aerosol vertical distributions was evaluated. Table S1 in the Supplement presents the correlation coefficient and normalized mean bias (NMB) of NO₂, ozone, and PM_2.5 concentrations in North China. These statistics were calculated based on the average hourly concentrations at 12 observation stations in the region (37–42° N, 114.375–118.125° E) where modification of aerosol vertical distribution was applied. The GEOS-Chem generally reproduced the variation in surface concentrations in winter, achieving correlation coefficients of 0.75, 0.69, and 0.69 for NO₂, ozone, and PM_2.5, with underestimations of 22.1 %, 24.8 %, and 4.3 % for NO₂, ozone, and PM_2.5, respectively. The model exhibits higher correlation coefficients (0.80–0.84) with the ozone observations during summer, except for PM_2.5 in summer 2017. This seasonality of model performance aligns with validations reported in many previous studies (Kim et al., 2015a; Akimoto et al., 2019; Li et al., 2019a; Huszar et al., 2020; Zhai et al., 2021). The model simulated NO₂ and ozone best in the summer of 2019, with an NMB of −2.8 % and −0.9 %, but there was an overestimation of PM_2.5. Since we focused on the differences between sensitivity experiments, the model shows an acceptable performance in simulating surface air pollutants.

Figure 2 Vertical profiles of average (a–c) AOD and (d–f) relative changes of J[NO₂], and J[O¹D] in model layers at observation stations. The red, light, and dark blue lines represent observed, BASE-simulated, and REGION-simulated AOD, respectively. (g–i) Time series of daily average relative changes of surface J[NO₂] and J[O¹D] at observation stations. From top to bottom: winter 2017 (a, d, g), summer 2017 (b, e, h), and summer 2019 (c, f, i), respectively. The averages are calculated using data during the daytime when photolysis occurs.

Figure 2 compares the average AOD profiles derived from the lidar observation with the BASE simulation. In winter, the AOD was underestimated by 20.8 % and 30.1 % in model layers 1 and 2, respectively, but it was subsequently overestimated until layer 10 (∼ 1.29 km), with a maximum NMB of 18.9 %. In summer, similarly, the AOD was underestimated by 3.4 %–43.9 % below layer 6 (∼ 0.73 km) and was then overestimated by more than 50 % above layer 9 (∼ 1 km). As reported in our previous work, emission sources, optical properties, size distributions of aerosols, and RH are the key factors contributing to the model bias in AOD (Chen et al., 2024). As photolysis rates can also be affected by the single scattering albedo (SSA), we validated the simulation of SSA at 440 nm with the observation from AERONET in January 2017 (Fig. S1 in the Supplement). It shows that SSA was well reproduced, with an NMB of 3.8 %, so only AOD profiles were revised in this work. We also compared the daily average PM_2.5 profiles from radiosonde data and GEOS-Chem (Fig. S2). Consistent with previous studies (Kim et al., 2015b; Liu et al., 2021a; Chen et al., 2022), the model has difficulty in accurately capturing the complex characteristics of aerosol vertical distribution. The comparisons of observed and simulated PM_2.5 profiles for three typical cases are presented in Fig. S1. Overestimation of PM_2.5 in the upper layer was common, while the model performance varied in the lower layer. The above simulation biases in aerosol vertical distribution underline the importance of observational constraints that will be elaborated in the following.

3.2 Modification of model aerosol vertical distribution

The ground-based lidar provided high-resolution AEC profiles, and we used the hourly average AEC profiles to conduct the modification after interpolating to model layers. The first model layer (approximately 58 m) is in the fade area of the lidar; therefore, we assumed the AEC in layer 1 to be the same as that in layer 2. We tested the impact of this assumption on ozone with the radiosonde data, which has observed data in layer 1 (Fig. S2). The results indicate that when this assumption was adopted, the effect of revised aerosol vertical distribution on ozone barely changed in both spatial distribution and regional averages. Then, AOD for each layer can be obtained by multiplying AEC by the layer height. Subsequently, the observational AOD vertical distributions were incorporated into the model with the following equations:

$$\begin{array}{}\text{(1)}& {\displaystyle }{r}_i={\displaystyle \frac{{\mathrm{AOD}}_{i,\mathrm{obs}}}{{\mathrm{AOD}}_{\mathrm{1}\text{–}\mathrm{17},\mathrm{obs}}}},\text{(2)}& {\displaystyle }{\mathrm{AOD}}_{i,\mathrm{new}}=r_i\times {\mathrm{AOD}}_{\mathrm{1}\text{–}\mathrm{17},\mathrm{sim}},\end{array}$$

where r_i represents the ratio of AOD in layer i to the total AOD within 3 km (layers 1 to 17 in the model), AOD_i, obs is the AOD in layer i from observations, AOD_{1–17, sim} refers to the column AOD below 3 km from the BASE simulation, and AOD_i, new is the AOD in layer i after modification. The AOD in GEOS-Chem model was summed up by sulfate, nitrate, ammonium, OA, BC, sea salt, and dust AOD, and the contribution of each aerosol component remained unchanged in the REGION simulation. This modification of aerosol vertical distribution was applied only within the photochemistry module when calculating photolysis rates.

Unlike lidar, radiosonde observations have no fade area near the surface. Therefore, we used PM_2.5 concentration profiles from radiosonde data collected in summer 2019 to validate the reliability of results derived from observational constraints based on lidar. Before constraining the model, we converted the PM_2.5 concentration profiles into AEC profiles using the Interagency Monitoring of Protected Visual Environments (IMPROVE) algorithm (Pitchford et al., 2007). This algorithm estimates AEC based on measured species concentrations and has been widely validated in China (Cao et al., 2012; Tao et al., 2014; Xiao et al., 2014; Bai et al., 2020). The PM_2.5 and PM₁₀ concentrations and RH profiles from radiosonde data and the simulated proportions of aerosol components were utilized in the IMPROVE algorithm. Afterward, the model's aerosol vertical distribution was revised according to Eqs. (1)–(2), and it was updated every 3 h.

The modification of aerosol vertical distribution was carried out over North China (37–42° N, 114.375–118.125° E) as shown in Fig. 1a. Based on the observed surface pollutant concentrations, we found a strong temporal consistency between hourly concentrations at vertical observation stations and the average hourly concentrations at 12 city stations in the modified region (Table S2). The correlation coefficients of ozone (0.88–0.95) and PM_2.5 (0.53–0.86) indicate that air pollution in North China usually occurs regionally. Moreover, we also used observations of AEC profiles from CALIPSO (Level 2, V4.20) to demonstrate that aerosol vertical distributions in North China exhibit regional characteristics. Vertically, aerosols can occasionally be transported in the upper layer, as evidenced by an aerosol layer at approximately 1 km on 25 January 2017, extending for about 200 km across North China (Fig. 1b). Near the surface, as observed on 11 July 2017 (Fig. 1c), high values in the lower layers were also observed over a regional extent of several hundred kilometres. Besides, we analysed the correlation of simulated daily average aerosol vertical distributions at the observed grid point and the other 76 grid points in the modified region, finding that over half of the grid points exhibited significant correlation (Fig. S4). Li et al. (2022) also highlighted the regionality of the upper aerosol layer through joint observations from multiple radiosonde stations. Figure S5 shows the vertical distribution of PM_2.5 concentrations in model layers at six radiosonde stations for winter 2019. According to Li et al. (2022), these stations can be spaced up to 150 km apart, and simultaneous observations for a consecutive month show that the timing of occurrence and vertical profiles of PM_2.5 concentrations at different stations are very similar. The differences in real aerosol vertical distribution between the observation station and the modified region are inevitable. However, the consistency of pollutant concentrations at surface observation stations (Table S2), the regional characteristic of vertical distributions from satellite observations in Fig. 1, and the similarity of simulated aerosol vertical distributions in the modified region, combined with the PM_2.5 profiles at six radiosonde stations, could demonstrate the spatial representativeness of the measurements. As such, we believe that the aerosol vertical distribution in the modified region closely aligns with that observed at the Beijing and Baoding stations. The impact of aerosol vertical distribution modification on photochemistry in North China in this work may represent an upper estimate due to the regionally assumed profiles.

4 Impact of revised aerosol vertical distribution

Modifying aerosol vertical distribution can alter photolysis rates and thereby affect photochemical reactions. We evaluated the impact of revised aerosol vertical distribution on photolysis rates (Sect. 4.1), ozone (Sect. 4.2), and PM_2.5 (Sect. 4.3) across different seasons and discussed the influencing factors. It is noted that all impacts in this section were only evaluated during the daytime when photochemical reactions occur.

4.1 Impact of revised aerosol vertical distribution on photolysis rates

The photolysis of nitrogen dioxide (NO₂) and the photolysis of ozone are two major photolysis reactions in ozone production: ${\mathrm{NO}}_{\mathrm{2}}+hv\to \mathrm{NO}+{\mathrm{O}}^{\mathrm{3}}\mathrm{P}$ (λ<420 nm) and ${\mathrm{O}}_{\mathrm{3}}+hv\to {\mathrm{O}}_{\mathrm{2}}+{\mathrm{O}}^{\mathrm{1}}\mathrm{D}$ (λ<340 nm). Figure S6 shows the responses of their photolysis rates (J[NO₂] and J[O¹D]) at the surface layer to the observational constraints over North China. The photolysis rates only increased in the modified region, with a more pronounced increase from north to south. In winter, the regional average ΔJ[NO₂] was 63.4 %, and ΔJ[O¹D] was higher at 73.8 %. The relative changes of J[NO₂] and J[O¹D] in North China were 33.4 % and 41.8 % in summer 2017, respectively, and spatial distributions were similar to those in winter. Compared with the results from lidar constraints, similar changes in photolysis rates were also found in radiosonde constraints. The J[NO₂] and J[O¹D] increased by 34.8 % and 44.7 % in summer 2019, respectively, with the difference in change rates compared to winter primarily attributed to the southern part of North China. In comparison, the impact of aerosol vertical distribution on photolysis rates in both winter and summer was much greater than the impact of absorbing OA (13.5 %–18.1 %) and BC (11.4 %–12.5 %) over North China, as reported by Li and Li (2023). This further confirmed the importance of accurate representation of aerosol vertical distribution in evaluating API impact.

Figure 2 presents the vertical distributions of ΔJ[NO₂] and ΔJ[O¹D] at observation stations. With the observational constraints employed, the AOD at the lidar station increased below layer 4 in winter and decreased above this layer. Consequently, the photolysis rates increased in the lower layers and decreased in the higher layers, with the transition height approximately 0.5 km higher than AOD. Similarly, the modified AOD increased below layer 6 and layer 10 for the summer of 2017 and 2019 (Fig. 2b–c), which was more aligned with the observed exponential attenuation, and the changes in photolysis rates were consistent with the variations in AOD profiles. Differently, there was roughly a 5 % greater increase in J[O¹D] compared to J[NO₂] in all layers during summer. In addition, the accuracy of column AOD simulations plays an important role in the impact of revised aerosol vertical distribution. Figure S7 shows the vertical distribution of AOD, ΔJ[NO₂], and ΔJ[O¹D] under different scenarios of AOD simulation bias. When the column AOD was underestimated, the observationally constrained layer AOD was still lower than the observations. The revised vertical distribution was more consistent with observations when the column AOD was overestimated, but the revised AOD in lower layers was much higher than the observations. Consequently, the relative change of photolysis rates was approximately 3 times higher with overestimated AOD compared to underestimated AOD in winter and 1.5 times higher in summer. In this work, we focused on aerosol vertical distribution, and the improvement of AOD simulation should be promoted in future study.

We also compared the temporal changes in photolysis rates (Fig. 2g). The photolysis rates increased primarily during several pollution periods in winter, with peak values exceeding 100 %. However, during clean days, aerosol vertical distribution had a limited impact on photolysis. The peak of the daily average photolysis rate changes in summer was notably lower than that in winter (Fig. 2h–i), yet the photolysis enhancement during summer was more consistent throughout the period.

To further verify the impacts of aerosol vertical distribution modification, we also compared the simulated surface J[NO₂] to data at Xianghe station (39.76° N, 116.96° E) located in North China from Zhao et al. (2021) (Fig. 3). This dataset was reconstructed by observational J[NO₂], total ultraviolet radiation, and the Troposphere Ultraviolet and Visible (TUV) radiation model, with an improved R² value of 0.94 between the observed and reconstructed data. The seasonal average J[NO₂] in summer was higher than in winter, and it was greater in summer 2019 than in summer 2017 due to the decline in aerosol concentrations. More importantly, the BASE simulation underestimated observation-based J[NO₂] during all simulation periods. In winter, the underestimation was 20.1 %, but after the revision of aerosol vertical distribution, it turned into a 4.8 % overestimation. A similar improvement in the underestimation of photolysis rates was also demonstrated in the summers of 2017 and 2019. These results indicate that revising the aerosol vertical distribution can better simulate photochemical processes.

Figure 3 Simulated and observed average surface J[NO₂] at Xianghe station for winter 2017, summer 2017, and summer 2019. The seasonal average J[NO₂] from Zhao et al. (2021) is presented by grey bars.

4.2 Impact of revised aerosol vertical distribution on ozone

4.2.1 Wintertime

Figure 4 shows the spatial distributions of changes in surface ozone and NO₂ in response to observational constraints on aerosol vertical distribution. Generally, as the aerosol layer moved from the upper layers to the lower layers, more solar radiation reached the surface and promoted photolysis. Increased photolysis of NO₂ accelerated ozone production, leading to a rise in surface ozone levels and a corresponding decrease in NO₂. The variations in ozone displayed patterns that closely resembled the spatial distribution of changes in photolysis rates. In winter, ozone increased by an average of 0.9 ppb, with a maximum increase of 2.7 ppb in the modified region (Table 1). Figure 4c illustrates the default effect of API on ozone in winter, with a regional average of −1.6 ppb. The impact of revised aerosol vertical distribution on ozone can reduce the API-induced ozone by 56 %, implying that the vertical distribution is as important as the column AOD intensity in determining API. Moreover, the underestimation of ozone in North China was reduced from −25 % in the BASE simulation to −17 % in the REGION simulation (Table S1). This suggests that improving the accuracy of aerosol vertical distribution representativeness could mitigate the underestimation of ozone levels in winter.

Figure 4 Simulated responses of daytime ozone and nitrogen dioxide (NO₂) to observational constraints on aerosol vertical distribution over North China (left and middle columns), along with the effect of aerosol–photolysis interaction (API) on daytime ozone (right column). From top to bottom: winter 2017 (a–c), summer 2017 (d–f), and summer 2019 (g–i), respectively. The black rectangle shows the region where vertical distribution modification was applied.

Table 1 The regional maximum and average changes in ozone and PM_2.5 over North China for seasonal and daily mean. O₃_API is the default effect of aerosol–photolysis interaction (API) on ozone. Relative changes are also given in brackets.

Figure 5a and d further present the changes in daily average NO₂ and ozone along the longitude profile of the lidar station in winter. In line with the change in photolysis rates, the changes in NO₂ and ozone were also periodic. It is evident that the increase in NO₂ shown in Fig. 4b was not contributed by a single event. The surplus of NO_x in the winter atmosphere, combined with the increase in ozone concentrations due to enhanced photolysis, led to a greater oxidation of NO. Therefore, NO₂ increased in the region with high NO_x emission when ozone production was accelerated. For daily average, API led to a reduction of ozone by 0.1–6.2 ppb in North China. For reference, Xing et al. (2017) reported that API reduced daily maxima 1 h ozone by 2–8 ppb in January 2013 over North China Plain. With revised aerosol vertical distribution, ozone was primarily increased during the periods of 3–8, 15–18, and 23–28 January when photolysis was enhanced. The regional average daily ozone was increased by 0.1–3.4 ppb, and the regional maximum daily ozone was increased by 0.2–10.5 ppb (Table 1). Our results indicate that the impact of revised aerosol vertical distribution can greatly weaken the default API effect.

Figure 5 Simulated responses of daytime average ozone and nitrogen dioxide (NO₂) to observational constraints on aerosol vertical distribution along the longitude profiles of observation stations. From left to right: winter 2017 (a, d), summer 2017 (b, e), and summer 2019 (c, f), respectively. The dotted lines indicate the latitude range of the modified region, and the longitudes of the stations are marked at the top.

4.2.2 Summertime

In summer, areas with increased photolysis rates experienced a decrease in NO₂ concentrations and an increase in ozone (Fig. 4d–e). However, the impact of revised aerosol vertical distribution on ozone concentrations was only half that in winter, with an average of 0.5 ppb and a maximum of 1.6 ppb in North China. The default effect of API on regional ozone was −1.4 ppb, which was reduced by 36 % after revising aerosol vertical distribution. Furthermore, ozone statistics in the REGION simulation show little difference from those in the BASE simulation (Table S1). These results indicate that the significance of aerosol vertical distribution on ozone was less pronounced in summer than in winter. The response of ozone to observational constraints based on radiosonde data was similar to that based on lidar data in terms of both the spatial distribution and the regional average. The regional average change in ozone during summer 2019 was 0.7 ppb, with a regional maximum of 2.3 ppb, both smaller than the changes in winter. The similar results constrained by these two sets of different observational data in summer confirm the reliability of this finding.

From the perspective of temporal variation, the increase in ozone and the decrease in NO₂ during summer corresponded to the periods of increased photolysis rates shown in Fig. 2. Ozone increased most significantly from 3 to 6 July 2017, when photolysis rates increased most. The regional average daily ozone and the regional maximum daily ozone increased by 0–1.9 and 0.4–7.1 ppb in summer 2017 and 0.1–1.3 and 0.5–6.4 ppb in summer 2019, respectively (Table 1). These results further confirmed that the impact of revised aerosol vertical distribution on ozone in summer is weaker compared to winter.

4.2.3 Role of PBL mixing

As shown in Fig. 2, the modification of aerosol vertical distribution mainly increased the AOD in the lower layers, while AOD was decreased in the upper layers across different seasons. With similar modifications of aerosol vertical distribution, the various responses of surface ozone concentration across different seasons are closely linked to planetary boundary layer (PBL) mixing (Yin et al., 2024). Figures 6 and 7 present the average diurnal variation in the vertical distribution responses for typical modification cases characterized by high planetary boundary layer height (PBLH) and low PBLH. The cases with high PBLH were in summer, while those with low PBLH were in winter (see detailed information in Table S3).

In Fig. 6a, the AOD decreased in the upper PBL and increased in the lower PBL after modification, with the PBLH reaching layer 15 (∼ 2.2 km) at noon. Consequently, the photolysis rates changed following the changes in AOD, as shown in Fig. 6c and d. The changes in photolysis rates resulted in an increase in ozone in the lower PBL, a decrease in ozone production in the upper PBL, and a larger vertical gradient of ozone concentration. Then, the strong PBL mixing led to the high ozone concentrations at the surface being entrained into the upper PBL, partially offsetting the increase in surface ozone and, in some cases, even leading to a decrease in surface ozone. This kind of compensatory effect of PBL mixing on surface ozone in response to API was recently mentioned in previous studies (Gao et al., 2020; Shi et al., 2021; Yan et al., 2023). When the PBLH started to decrease at 10:00 UTC, the photolysis rates increased in the PBL, and the response of surface ozone recovered. In Fig. 7, the reduction of AOD was primarily above the PBL, and the photolysis was promoted throughout the entire PBL, with the maximum PBLH reaching layer 6 (∼ 0.73 km) at noon. As such, ozone increased in all layers in the PBL. After sunset (18:00 UTC), the PBLH suddenly dropped to the lowest layer, and the increased ozone in the residual layer reacted with the surplus NO in the atmosphere, increasing nighttime NO₂. Overall, the impact of revised aerosol vertical distribution on surface ozone was larger under conditions with low PBLH. Due to stronger solar radiation and warming, the PBL usually develops higher in summer than in winter. Hence, the weaker impact of revised aerosol vertical distribution on surface ozone in summer could be attributed to the compensatory effect of PBL mixing.

Figure 6 Diurnal variation in the vertical distribution responses of (a) AOD, (c) J[NO₂], (e) J[O¹D], (b) nitric oxide (NO), (d) nitrogen dioxide (NO₂), and (f) ozone at the observation stations under conditions with high planetary boundary layer height (PBLH).

Figure 7 Same as Fig. 6 but under conditions with lower planetary boundary layer heights (PBLHs).

4.3 Impact of revised aerosol vertical distribution on PM_2.5

Modifying the vertical distribution of aerosols in the model resulted in significant changes in photolysis rates, which subsequently affected the atmospheric oxidizing capacity and, in turn, the chemical rate of secondary aerosol formation. In this section, we will discuss the impact of revised aerosol vertical distribution on atmospheric oxidizing capacity and PM_2.5 formation.

4.3.1 Wintertime

Figures 8 and S8 display the spatial distribution of changes in surface PM_2.5 concentration and the relative changes, respectively. In winter, the increase in PM_2.5 concentration was primarily concentrated in the modified region, with high-value areas corresponding to the regions of NO₂ increase (Fig. 8a). PM_2.5 concentration was increased by an average of 0.8 µg m⁻³ (0.2 %), with a maximum increase of 3.1 µg m⁻³ (1.2 %) in the modified region (Table 1). The underestimation of PM_2.5 in winter was slightly improved with observational constraints on vertical distribution (Table S1). The acceleration of photolysis in the surface layer caused by modifications of aerosol vertical distribution enhanced the atmospheric oxidizing capacity and promoted secondary aerosol formation.

Figure 8 Simulated responses of PM_2.5 concentration to observational constraints on aerosol vertical distribution over North China. From left to right: winter 2017 (a), summer 2017 (b), and summer 2019 (c), respectively. The black rectangle shows the region where vertical distribution modification was performed.

Figure 9 presents the changes in daily average PM_2.5 concentration and the relative changes along the longitude profile of the observation station. The increase in PM_2.5 in North China coincided with periods of enhanced photolysis and increased ozone in winter. As illustrated in Table 1, the regional average daily PM_2.5 changed by −0.1 to 2.9 µg m⁻³ (−0.6 % to 1.5 %), and the regional maximum daily PM_2.5 changed by 0.1–21.5 µg m⁻³ (0.4 % to 5.3 %). The regional average API impact on PM_2.5 was −0.1 % to −4.4 % in North China, which was reduced by approximately 30 %. Although the impact of revised aerosol vertical distribution on seasonal average PM_2.5 was relatively small, its contribution during pollution events was notable. For example, on 25 January, the daily average PM_2.5 increased by 16.5, 8.0, 18.8, and 10.3 µg m⁻³ at Beijing, Tianjin, Baoding, and Langfang stations, respectively.

Figure 9 Simulated responses of daily average (daytime) PM_2.5 to observational constraints on aerosol vertical distribution along the longitude profiles of observation stations. (a–c) Mass concentration (µg m⁻³) and (d–f) proportion (%). From left to right: for winter 2017 (a, d), summer 2017 (b, e), and summer 2019 (c, f), respectively. The dotted lines indicate the latitude range of the modified region, and the longitudes of the stations are marked at the top.

4.3.2 Summertime

In summer, the impact of aerosol vertical distribution on PM_2.5 concentration was a quarter of that in winter, with an average of 0.2 µg m⁻³ and a maximum of 0.7 µg m⁻³ in the modified region. However, the relative change in PM_2.5 was higher than in winter, with an average of 0.3 %. The results from observational constraints based on radiosonde data also show an average of 0.2 µg m⁻³ (0.5 %) increase in PM_2.5 in summer 2019. Zhao et al. (2023) reported that the emission reduction in the summer of 2019 can cause an increase in PM_2.5 by 0.54 µg m⁻³ in the Beijing–Tianjin–Hebei region through API. This suggests that changes in aerosol vertical distribution are comparable to emission reduction when evaluating the effects of API on PM_2.5 in summer. Different from the wintertime, the differences in daily average changes of PM_2.5 between each day were relatively small (Fig. 9e–f). In summer 2017, the regional average changes in daily PM_2.5 and the regional maximum changes were 0–0.7 µg m⁻³ (0 %–1.0 %) and 0.2–2.9 µg m⁻³ (0.5 %–2.8 %). The increase in PM_2.5 over North China in summer 2019 closely mirrored the values in 2017.

4.3.3 Role of atmospheric oxidizing capacity

To explore the impact of revised aerosol vertical distribution on PM_2.5, we further analysed the major components contributing to the PM_2.5 increase. Secondary aerosols can account for nearly all of the PM_2.5 increase, and their formation is strongly influenced by the atmospheric oxidizing capacity. We compared the relative changes of hydroxyl radicals (OH) in Fig. S9. It is found that OH increased by 13.5 % in winter and 14.2 % in summer over North China, indicating no significant seasonal differences. As we have mentioned in Sect. 4.1, J[O¹D] increased 73.8 % in winter and 41.8 % in summer, which promoted the OH production. In winter, OH formation from the photolysis of HONO is also important (Tan et al., 2018). The revised aerosol vertical distribution accelerated this process, resulting in a regional average reduction of 4.4 % in HONO concentration. Therefore, the atmospheric oxidizing capacity was enhanced by the increase in these OH source reactions. Herein, we added the emission of HONO from transportation by applying a $\mathrm{HONO}/{\mathrm{NO}}_x$ emission ratio (Zhang et al., 2019a), and the heterogeneous reaction of NO₂ (Shah et al., 2020) was also included in the GEOS-Chem model, but the other HONO sources like soil emissions (Tan et al., 2023), livestock farming (Zhang et al., 2023), and particulate nitrate photolysis (Andersen et al., 2023) were not included. It should be noted that the enhancement of atmospheric oxidizing capacity may be weakened due to the underestimation of HONO concentrations in the model. Moreover, the acceleration of J[O¹D] and the photolysis of HONO were more significant in winter than in summer, with a similar increase in OH concentration, suggesting that the oxidation of NO₂ by OH was promoted more in winter.

Figure 10 Simulated responses of daily average (a) HNO₃ and (b) OH to observational constraints on aerosol vertical distribution along the longitude profiles of the lidar station in winter. The dotted lines indicate the latitude range of the modified region, and the longitude of the station is marked at the top.

Among secondary aerosols, nitrate was the major contributor to the increase in PM_2.5 in both winter and summer, accounting for approximately 75 %. Next are ammonium and sulfate, contributing approximately 20 % and 8 %, respectively, while secondary organic aerosols (SOAs) can be ignored. The formation of nitrate during the daytime is primarily contributed by the aerosol uptake of nitric acid (HNO₃) (Lu et al., 2019), which is mainly from the oxidation of NO₂ with OH. The enhancement of photolysis led to an increase in OH during both summer and winter; however, with abundant NO_x in the atmosphere during winter, more NO_x reacted with OH to form HNO₃. Figures 10 and S10 show the changes in HNO₃ and OH in winter and summer, respectively. Corresponding to the periods with OH increase, HNO₃ concentration increased significantly in winter, reaching up to 60 %. The areas with remarkable relative changes in HNO₃ concentration aligned with the increase in NO₂ in Fig. 5a and the strong response of PM_2.5 in Fig. 9a. In summer, the intensity of atmospheric oxidizing capacity was stronger than in winter (Liu et al., 2021b), and the NO_x concentration was lower; as a result, HNO₃ increased less than 10 %, even though the increase in OH was similar to winter. In general, the enhancement of atmospheric oxidizing capacity, combined with high NO_x emissions, contributed to the strong response of PM_2.5 to observational constraints on aerosol vertical distribution during winter pollution events.

5 Conclusions

In this study, we quantified the response of atmospheric photochemistry to observational constraints on aerosol vertical distribution in different seasons using continuous observations from ground-based lidar and radiosonde. Sensitivity experiments were conducted by modifying the aerosol vertical distribution over North China in the photochemical module of the GEOS-Chem model; the simulations in winter 2017 and summer 2017 were based on lidar data, and the simulations in summer 2019 were based on radiosonde data.

Compared to the observations, the GEOS-Chem model underestimated the AOD by 20.8 %–30.1 % below layer 2 (∼ 0.19 km) in winter and by 3.4 %–43.9 % below layer 6 (∼ 0.73 km) in summer, while overestimating AOD above these layers. After observational constraints, the AOD increased in the lower layers and decreased in the upper layers. The changes of photolysis rates of NO₂ and ozone (J[NO₂] and J[O¹D]) were consistent with the vertical changes in AOD in both seasons. The regional average ΔJ[NO₂] and ΔJ[O¹D] were 63.4 % and 73.8 % in winter and 33.4 % and 41.8 % in summer. In terms of time series, the photolysis rates primarily increased during pollution events in winter, reaching a maximum increase of over 100 %. In contrast, the daily average photolysis rate changes in summer were more steady, with lower peaks than in winter.

Surface ozone increased due to the enhancement of photolysis rates in the modified region, with an average of 0.9 and 0.5 ppb in winter and summer, respectively, which is about half of the default impact of API on ozone. During pollution episodes, the regional average daily ozone increased by 0.1–3.4 ppb in winter and 0–1.9 ppb in summer following the enhancement of photolysis; the regional maximum daily changes of ozone reached 10.5 and 7.1 ppb in winter and summer, respectively. The compensatory effect of PBL mixing under conditions with high PBLH could weaken the impact of revised aerosol vertical distribution on surface ozone. The results from observational constraints based on radiosonde data in summer 2019 were consistent with those in summer 2017, consolidating the reliability of our work. Additionally, the model underestimation of ozone in North China was also greatly reduced during winter after the modification of aerosol vertical distribution.

Due to the promotion of secondary aerosol formation under more active photochemistry, regional PM_2.5 concentration increased by 0.8 µg m⁻³ (0.2 %) in winter and 0.2 µg m⁻³ (0.3 %) in summer. The response of PM_2.5 was weaker in summer, with an increase in the daily average by 0–0.7 µg m⁻³, which was close to the results based on radiosonde constraints. However, in winter, the regional average daily PM_2.5 changed by −0.1 to 2.9 µg m⁻³, with a maximum of 8.0 to 18.8 µg m⁻³ at some city stations, indicating the importance of aerosol vertical distribution during the pollution period. The increase in nitrate concentration resulting from enhanced HNO₃ formation was the primary reason for the increase in PM_2.5 during winter. The increase in atmospheric oxidizing capacity can account for the strong response of PM_2.5 during heavy pollution.

Early studies based on box models have emphasized that API can lead to great reductions in photolysis rates and ozone concentrations (Hollaway et al., 2019; Wang et al., 2020b). They found a greater reduction in the net photochemical production of vertical ozone than surface ozone. Recent three-dimensional chemical transport model studies have highlighted the importance of the overlooked process in box model studies – boundary layer mixing, which can account for this inconsistency (Gao et al., 2020; Yan et al., 2023). Our study, based on observational constraints of aerosol vertical distribution, resolves an underappreciated issue and provides a deeper understanding of the impact of API. When considering observational constraints, the impact of API on surface ozone decreased significantly by 36 %–56 %, suggesting that the impact of API in previous studies would be overstated. Despite the weakening of API, revising aerosol vertical distribution can inspire more accurate predictions of ozone levels and the atmospheric oxidizing capacity in winter, particularly in regions with complex aerosol vertical distributions. The comparison of cases with different boundary layer heights also offered valuable insights into the discrepancies in changes in photolysis rate and surface ozone.

In future studies, more efforts are needed to improve the model performance in aerosol vertical distribution. Although the vertical distribution was constrained, discrepancies still exist between the simulated AOD and the observations. More accurate aerosol optical properties, such as single-scattering albedo (SSA), which can help distinguish between the scattering and absorption of light, are strongly needed in the model simulation. Moreover, the model validation with observed photolysis rates should be involved, as the photolysis rate is one of the most direct variables for verifying the simulation of photochemical reactions. In summary, our results confirmed the importance of reasonable representativeness of aerosol vertical distribution in API effects and provided valuable implications for future studies.