The 21st-century wetting inhibits growing surface ozone in Northwestern China

Previous studies have shown that surface air temperature (SAT) facilitates the formation of surface ozone (O₃), while relative humidity (RH) often inhibits ozone generation. However, the degree to which O₃ may respond simultaneously to rising SATs and RH due to climate change remains less understood. We conducted extensive atmospheric chemistry model scenario simulations to investigate the impacts of long-term trends of humidification and warming on summer O₃ concentrations in Northwestern China (NW) from 1998 to 2017, a period during which this region experienced both warming and the most significant wetting trend in China. We found that the summer mean O₃ level in NW increased by 19.9 % during this time. The changes in meteorology led to a reduction in O₃ levels, primarily due to humidification in NW, which counteracts the warming-induced O₃ increase. We demonstrate that the wetting trend in NW will continue from 2019 to 2030 under a shared socioeconomic pathway (SSP) in conjunction with a representative concentration pathway SSP5-RCP8.5, but will either cease or shift to drier conditions from 2019 to 2060 under the SSP2-RCP4.5. No significant responses of O₃ fluctuations to RH variations were observed, partly due to manual intervention under the SSP2-RCP4.5.

1 Introduction

Atmospheric humidity is recognized as a crucial factor that directly and indirectly influences the formation and evolution of surface ozone (Ghazali et al., 2022; Han et al., 2011; Kerr et al., 2020). The relationship between atmospheric humidity and surface O₃ is complex. From a chemical standpoint, humidity can modify the rates of photochemical reactions, thereby affecting O₃ formation. In addition, humidity plays a role in generating hydroxyl radicals (OH) in the atmosphere, which are vital for breaking down volatile organic compounds (VOCs) and other pollutants (Ding et al., 2023a; Ma et al., 2021). These radicals can interact with O₃ precursors, affecting the rate of O₃ production. High humidity can also enhance aerosol formation, facilitating heterogeneous reactions. Depending on the specific chemical pathways involved, some of these reactions may either generate or deplete O₃ (Eck et al., 2020). From a physical perspective, humidity is critical for cloud and fog formation as well as precipitation. The latter promotes O₃ dissolution in water droplets and its removal from the atmosphere through wet deposition, particularly during precipitation events, which weakens solar radiation and disrupts the energy balance near the surface (Ding et al., 2021; Eck et al., 2020; Kleeman, 2008; Madden and Williams, 1978; Millstein and Harley, 2009).

It has been reported that O₃ levels increased by 30 % over Northern China in summer and 50 % over Eastern China in autumn during a dry year compared to a wet year, indicating that wet climate conditions do not support rising O₃ levels (Ding et al., 2023b). Long-term ambient O₃ measurement data indicate that rising absolute air humidity resulted in lower near-surface O₃ concentrations at temperatures of 0–30 °C (Belan and Savkin, 2019). Other studies examining the relationship between O₃ concentrations and atmospheric humidity have also identified negative correlations between O₃ levels and humidity (Li et al., 2022a). Nevertheless, significant knowledge gaps regarding the relationship between humidity and O₃ persist (Li et al., 2021a). Most previous studies utilized short-term or instantaneous O₃ concentrations to evaluate the associations between O₃ and atmospheric humidity (Gong et al., 2021; Li et al., 2021b). Their long-term relationships still require clarification. Previous investigations of the connections between climate and O₃ primarily focused on the impact of air temperature, which often overshadows the effects of other meteorological variables (Wang et al., 2024). The extent to which atmospheric humidity could significantly influence the long-term O₃ trend remains uncertain, given that the synergistic effects of warming and wetting frequently occur in a warming climate.

In recent decades, Northwestern China (NW) has emerged as a critical region for studying O₃ variability due to its unique climatic and environmental conditions. Observational studies report a significant increase in summer surface O₃ concentrations across the region, with trends ranging from 1.5–3.0 µg m⁻³ yr⁻¹ between 2000 and 2020 (Li et al., 2020b; Zhang et al., 2022). This rise has been attributed to intensified anthropogenic emissions (e.g., NO_x and VOCs from energy and industrial activities) and shifting meteorological patterns, including the “warming and wetting” climate regime (Yang et al., 2023). For instance, Ding et al. (2023b) identified a 20 % increase in O₃ levels during dry years compared to wet years in arid regions, highlighting the sensitivity of O₃ to humidity. In this context, O₃ and RH trends in NW over the past decades probably present a unique scenario to help bridge this knowledge gap. NW (Fig. S1), located in the hinterland of the Asian-European continent, is Asia's largest arid and semi-arid region, characterized by limited rainfall and vegetation (Koster et al., 2004; Yang et al., 2023). Meteorological records indicate that from 1960 to 2013, precipitation in NW increased at a rate of 0.55 mm yr⁻¹ (Li et al., 2012; Liu et al., 2013; Yang et al., 2017). During this period, the air temperature rose at a rate of 0.034 °C yr⁻¹, exceeding the warming trends observed across all of China (0.025 °C yr⁻¹) and globally (0.013 °C yr⁻¹). Although RH does not exhibit a significant growth trend during this period, in the recent decade (2008–2017), it increased markedly at a slope of 0.313 in NW and 0.381 in the three westernmost provinces of Xinjiang, Gansu, and Qinghai within NW. Based on these records, NW appears to be entering a warming and wetting phase, accompanied by a rise in precipitation, runoff, and lake levels (Nie et al., 2022; Peng and Zhou, 2017; Shi et al., 2002; Yao et al., 2021).

The present study aims to address the knowledge gap in the complex and long-term relationships between ozone pollution and atmospheric humidity, focusing on NW, where increasing O₃ pollution, air temperature and humidity have co-occurred (Peng and Zhou, 2017; Yang et al., 2017), mainly since the early 2000s. This research provides valuable insights into the broader context of air quality management and climate variability by enhancing our understanding of the long-term relationship between ozone pollution and atmospheric humidity.

2 Method and data

2.1 WRF-Chem model configuration

Since measured O₃ concentration data in China is only available after 2013, we applied the three-dimensional Weather Research and Forecasting with Chemistry (WRF-Chem) version 3.7 model (http://www2.mmm.ucar.edu/wrf/users/wrf_files/wrfv3.7/updates-3.7.html, last access: 26 August 2025) to predict meteorology, O₃ concentrations, and other relevant data in Northwestern and mainland China. The model domain spans the entire mainland China at a 20 km × 20 km resolution, extending vertically from the surface to the lower stratospheric ozone layer at an altitude of 50 hPa with 30 staggered vertical layers. This vertical segmentation facilitates a deepened comprehension of the O₃ dynamics within the region, enabling the capture of its detailed behavior throughout the atmospheric stratifications (Li et al., 2020a; Zhang et al., 2020, 2022). Detailed physical and chemical schemes contributing to O₃ formation in WRF-Chem and emission input followed previous studies (Bei et al., 2018; Bossioli et al., 2016; Liao et al., 2015; Tie et al., 2009; Wild et al., 2000) and are presented in Sect. S1 in the Supplement and Tables S1–S2. The soil NO, NO_x and HONO emissions have been considered a necessary process contributing to surface ozone formation in recent years (Lu et al., 2021). However, given the lack of soil emission inventories of these reactive nitrogen species in China and the world, the present modeling study did not take these processes into consideration. The aerosol feedback mechanisms include the direct and indirect effects, and both are considered in WRF-Chem. The model simulations with a nudging option in WRF were conducted from 1 June to 31 August in each summer from 1998 to 2017, with a time-step of 30 s. A spin-up time of 3 d was adopted before 1 June in each summer. Further details are referred to Sect. S1 in the Supplement.

Located in the upstream of China in terms of dominant prevailing westerly winds in the mid-latitudes of the Northern Hemisphere, it is expected that NW would receive more O₃ and its precursors from foreign sources in Eurasian countries to the west of China compared to Eastern China (Li et al., 2014, 2016). Such an “upstream” effect is implicitly accounted for in meteorology and chemistry lateral boundary conditions under an open boundary approach (https://www2.acom.ucar.edu/wrf-chem, last access: 8 May 2024). These boundary conditions typically come from global chemical transport model outputs (e.g., GEOS-Chem, MOZART) that themselves simulate O₃ on a global scale (including those upstream regions), chemical reactions, wet and dry deposition and atmospheric transport processes. These datasets are usually generated or assimilated from global models and observations. The lateral boundary conditions govern the fluxes and concentration levels of pollutants entering the model domain from external (upstream) regions. They represent the environmental or regional background concentration outside the model domain, which already integrate emissions and transport influences from a broader region (including upstream areas). The lateral boundary conditions of chemical species in our modeling exercise were estimated from MOZART-4 reanalysis driven fields (Emmons et al., 2010; Zhang et al., 2020) on a daily basis. The ERA-Interim data with a 6-hourly time resolution and 0.7° × 0.7° lat/lon spatial resolution provided by the ECMWF (European Centre for Medium-Range Weather Forecasts) were adopted as lateral boundary conditions of meteorology. Further, since the multiple model scenario simulations in our study used the same lateral boundary conditions, this effectively removed the influences of precursor emissions and atmospheric transport from Eurasian countries but focused on the wetting effect on O₃ evolution in the internal model domain (China).

2.2 Trend

The Mann–Kendall (MK) test and Sen's slope (El-Shaarawi and Niculescu, 1992; Kendall, 1949; Mann, 1945) were used to examine the trend of O₃, surface air temperature (SAT) at the 2 m height, and relative humidity (RH, %) at the 2 m height. The MK test and Sen's Slope are both non-parametric statistical methods commonly used for analyzing trends in time series data. The MK Test helps determine if there is a significant trend, and Sen's slope quantifies the strength and direction of the trend (slope value).

Interannual ozone variability in NW is primarily driven by large-scale meteorological shifts (e.g., warming, humidification) and regional precursor emissions, rather than diurnal processes. Previous studies (Cooper et al., 2014; Li et al., 2020a) have revealed that long-term O₃ trends (interannual and interdecadal scales) are robustly represented by seasonal or annual mean concentrations, as diurnal variations associated with daytime photochemistry and nighttime titration are statistically insignificant. For example, Ding et al. (2023b) reported that separating daytime and nighttime O₃ yielded negligible differences (> 95 % correlation) in O₃ trend over a 20-year period in arid seasons in China.

2.3 O₃ Attribution to meteorology and emission

We calculated the fraction (percentage change, %) to account for O₃ attribution to meteorology and emission subject to different model scenarios. For example, we define the fraction

$$\begin{array}{}\text{(1)}& F_{{\mathrm{O}}_{\mathrm{3}}}=(C_{\mathrm{S}\mathrm{3}}-C_{\mathrm{S}\mathrm{1}})/C_{\mathrm{S}\mathrm{1}}\times \mathrm{100}\mathit{\%},\end{array}$$

as the attribution of O₃ concentration to emissions, where C_S1 and C_S3 are the summer surface O₃ air concentrations from the base (S1) and fixed meteorology (S3) scenarios.

2.4 Data

To initialize and prescribe boundary conditions in the WRF model from 1998 to 2017, meteorological data were sourced from the ERA-Interim reanalysis from the European Centre for Medium-Range Weather Forecast (ECMWF, https://rda.ucar.edu/datasets/d627000/, last access: 26 August 2025). This allows the adoption of the open boundary conditions to dynamically account for cross-regional O₃ and precursor transport, and to resolve both local photochemical production and contributions from transboundary sources, ensuring a comprehensive representation of O₃ dynamics.

Anthropogenic emissions were collected from the Emissions Database for Global Atmospheric Research (EDGAR) version 4.3 on a spatial resolution of 0.1° × 0.1°, which was available when we started this modeling investigation. A grid-mapping program implemented the emissions into the WRF-Chem (Crippa et al., 2018; Mo et al., 2017). Biogenic emissions were simulated using MEGAN version 2.1 (Guenther et al., 2006, 2012). A recent study suggests that temperature can alter precursor emissions (Pfannerstill et al., 2024). However, since available emission inventories worldwide, including the widely-used EDGAR inventory, did not account for the influence of meteorology, typically the air temperature, on emissions but only considers emissions from anthropogenic and natural sources, the emissions used as input in the present modeling might be subject to additional uncertainties. The reduction of these uncertainties will depend on improved inventories that take the meteorological effect into account.

We replaced WRF land-use change (LUC) data with the Chinese land use/land cover remote sensing monitoring database provided by the Data Center for Resources and Environmental Sciences, Chinese Academy of Sciences (RESDC) (https://www.resdc.cn/, last access: 26 August 2025), which significantly improved China's LUC in our investigation of the effect of urbanization on O₃ pollution in China (Zhang et al., 2022). The new LUC data have a spatial resolution of 1 km × 1 km, which is extrapolated to the WRF-Chem grid cells (20 km × 20 km).

Considering that the warming and wetting trends in NW might persist in the coming decades, we also projected future O₃ and meteorology in 2019, 2030 and 2060 in the WRF-Chem model framework and simulations, respectively. We collected future anthropogenic emission data from the Dynamic Projection Model for Emissions in China (DPEC Inventory – MEICModel, http://meicmodel.org.cn, last access: 26 August 2025), with a spatial resolution of 0.25° × 0.25° longitude and latitude (Cheng et al., 2021, 2023; Tong et al., 2020). The meteorological data were collected from Shared Socioeconomic Pathways (SSPs) and Representative Concentration Pathways (RCP) climate scenarios under the CMIP6 framework (https://aims2.llnl.gov/search/cmip6/, last access: 26 August 2025), with the MPI-ESM1-2-HR model (Eyring et al., 2016). Compared with other models, MPI-ESM1-2-HR has higher resolution and can simulate the complex interactions within the climate system, including all climate scenarios (SSP2-RCP4.5 and SSP5-RCP8.5) targeted by this study, and has been widely used in WRF simulations.

2.5 Model evaluations

Rigorous comparisons between simulated O₃ concentrations and measured data were conducted and presented in Figs. S2–S7 and Sect. S2 in the Supplement. Overall, the simulated O₃ concentrations agree reasonably well with measured data, demonstrating a high level of reliability of modeling results.

2.6 Model scenario set-up

We established several model scenarios to distill signals of meteorological conditions, primarily SAT and RH, in modeled O₃ concentrations. Detailed configurations are provided in Table 1. The first scenario (S1) serves as the baseline, incorporating summer-specific meteorological variables and emissions from 1998 to 2017 into WRF-Chem simulations, and is thus considered a realistic scenario. In the second scenario (S2), we fixed emissions from 1998 throughout the period from 1998 to 2017, aiming to assess the significance of the meteorological effect on O₃ variation. In the third scenario (S3), we fixed meteorology from 1998 during the same period, thereby emphasizing the importance of emissions in O₃ variation. To evaluate the sensitivity of scenario S3, we also conducted an additional WRF-Chem run by fixing SAT and RH only in 1998 (scenario S4) and compared the modeled O₃ concentrations with those simulated in scenario S3 for 2004. Overall, we found minor differences (two tails T-test = 0.1065) between model scenario S3 and the results from the fixed SAT and RH model run (Fig. S8) but slightly stronger O₃ pollution levels subject to fixed RH and SAT scenario than S3 results can be discerned in the Northern China Plain. ozone concentrations under fixed SAT and RH in 1998 seemed higher than fixed all meteorology in 1998 (S3 scenario). We further calculated the differences of the planetary boundary layer heights (PBLH) between the two scenarios (fixed RH and SAT scenario minus scenario S3). The results are illustrated in Fig. S9. Negative PBLH differences can be seen in many areas across China, including the North China Plain (NCP), indicating that fixed SAT and RH scenario reduced PBLH, which often associates with stronger air pollution within the PBLH. By calculating the differences and fractions, we can discern the contributions of meteorology and emissions to O₃ formation and destruction. We also established two additional model scenarios to project the responses of O₃ concentrations to future meteorology and emissions in NW based on CMIP6 pathways SSP2-RCP4.5 and SSP5-RCP8.5, specifically for 2019, 2030, and 2060, respectively. Detailed descriptions of these two pathways and the corresponding data are presented in Sect. S3 in the Supplement.

Table 1 Model scenarios.

3 Results and discussion

3.1 Summer ozone trends

Figure 1a and b present MK-test statistics and Sen's slope estimated temporal trends of summer O₃ concentrations from 1998 to 2017 under scenario S1 (base scenario) and S3 with fixed meteorology in China. In the S1 scenario (Fig. 1a), O₃ concentrations exhibit statistically significant declining trends in the northernmost regions, spanning from Northeastern Xinjiang to Northeastern China, with the most significant decline occurring in Inner Mongolia. Increasing O₃ concentrations are observed along Tianshan Mountain in Xinjiang, western Tibet, and central China. The fixed meteorology scenario (S3, Fig. 1b) shows a monotonically increasing trend across China, indicating continuously rising O₃ levels for the past two decades due to growing precursor emissions, except for some grid cells in the Yangtze River Delta (YRD), the Pearl River Delta (PRD), the Beijing-Tianjin metropolitan area and other sporadic regions, where the MK-test and Sen's slope identified statistically significant negative O₃ trends. Previous studies have also noted the negative trends over the past decade in these regions (Hu et al., 2024; Li et al., 2022b; Wang et al., 2020; Yu et al., 2021). The Beijing-Tianjin area, YRD and PRD are three critical regions under China's joint efforts to prevent and control air pollution. Regional emission mitigation in these areas may lead to varying reductions in precursor emissions and declines in O₃ concentrations. Overall, increasing precursor emissions have dominated the O₃ trend in China, as meteorology remained constant during this period. This is evidenced by the linear trends of NO_x and NMVOC (non-methane VOC) emissions in China from 1998 to 2017 (Fig. S10). When comparing the MK-test and Sen's slope results between the two scenarios model runs (Fig. 1c), the meteorological conditions from 1998 to 2017 overshadowed the effects of emissions, primarily contributing to the declining O₃ levels (negative trends) in the northernmost regions of China, extending from Inner Mongolia to Northeastern China. Our modeling result shows that mean summer O₃ concentration, averaged over Inner Mongolia, reached the maximum in 2007 and slightly declined thereafter (Fig. S11). This can be partly attributed to relatively lower precursor emissions resulting from the sparse population and limited industrial activities in this part of China, though having a growing trend (Zhang et al., 2022, 2023).

Figure 1 MK-Test and Sen's slope estimated summer O₃ concentration trend and O₃ attribution (fraction or percentage change, %) under baseline (S1) and fixed meteorology (S3) scenarios from 1998 to 2017, respectively. (a) MK-Test and Sen's slope estimated trend from the base scenario S1, the trends of concentrations are well within the MK-Test statistically significant range at 95 % confidence level in terms of $\left|Z\right|$ > 1.96 criteria; (b) same as panel (a) but for fixed meteorology scenario S3; (c) fraction of O₃ concentration trend between the model scenarios 1 and 3 from 1998 to 2017; (d) mean differences of summer mean O₃ concentrations averaged from 1998 to 2017 between the model scenarios 1 and 3, defined by $\stackrel{\mathrm{‾}}{\mathrm{\Delta }{\mathrm{O}}_{\mathrm{3}}}={\stackrel{\mathrm{‾}}{\mathrm{O}}}_{\mathrm{3},\mathrm{S}\mathrm{1}}-{\stackrel{\mathrm{‾}}{\mathrm{O}}}_{\mathrm{3},\mathrm{S}\mathrm{3}}$, where ${\stackrel{\mathrm{‾}}{\mathrm{O}}}_{\mathrm{3},\mathrm{S}\mathrm{1}}$ and ${\stackrel{\mathrm{‾}}{\mathrm{O}}}_{\mathrm{3},\mathrm{S}\mathrm{3}}$ stand for mean summer O₃ concentrations averaged over 1998 to 2017, respectively. $\stackrel{\mathrm{‾}}{\mathrm{\Delta }{\mathrm{O}}_{\mathrm{3}}}$ in three westernmost provinces of NW is shown by yellow bars and blue bars for three easternmost provinces of NW. Error bar in Fig. 1d stands for ±1 SD.

To discern potential influences of meteorology on O₃ concentrations at the provincial level, we estimated the differences of summer mean O₃ concentrations averaged from 1998 to 2017 between model scenarios 1 (S1) and 3 (S3), defined by $\stackrel{\mathrm{‾}}{\mathrm{\Delta }{\mathrm{O}}_{\mathrm{3}}}={\stackrel{\mathrm{‾}}{\mathrm{O}}}_{\mathrm{3},\mathrm{S}\mathrm{1}}-{\stackrel{\mathrm{‾}}{\mathrm{O}}}_{\mathrm{3},\mathrm{S}\mathrm{3}}$. Figure 1d illustrates $\stackrel{\mathrm{‾}}{\mathrm{\Delta }{\mathrm{O}}_{\mathrm{3}}}$ in each province of NW. Since the mean O₃ levels under scenario S1 were greater than those under scenario S3 with fixed meteorology in 1998, although in some years we also identified negative ΔO₃ in the three westernmost provinces (Xinjiang, Gansu, and Qinghai), we would expect positive $\stackrel{\mathrm{‾}}{\mathrm{\Delta }{\mathrm{O}}_{\mathrm{3}}}$, as shown in Fig. 1d. In this context, the values of $\stackrel{\mathrm{‾}}{\mathrm{\Delta }{\mathrm{O}}_{\mathrm{3}}}$ indicate the extent of meteorological influence on O₃ concentrations. Specifically, the smaller $\stackrel{\mathrm{‾}}{\mathrm{\Delta }{\mathrm{O}}_{\mathrm{3}}}$, the stronger meteorological influences, as a smaller $\stackrel{\mathrm{‾}}{\mathrm{\Delta }{\mathrm{O}}_{\mathrm{3}}}$ suggests that O₃ levels with varying meteorology (S1) are closer to those with fixed meteorology in 1998 (S3). We highlighted the three provinces with smaller $\stackrel{\mathrm{‾}}{\mathrm{\Delta }{\mathrm{O}}_{\mathrm{3}}}$ (< 2 µg m⁻³) using yellow bars, all located in westernmost NW, indicating a weaker increase in O₃ concentrations in these three westernmost provinces compared to others, aligning well with the RH trends in these provinces, as compared to the other three provinces located in the east of NW (Inner Mongolia, Ningxia and Shaanxi), as mentioned in the Introduction. As shown in Fig. S12, modeled concentration fractions under the S2 model scenario with fixed emissions in 1998 in the three westernmost provinces of NW are negative, indicating that meteorology tended to inhibit the increase in summer ozone concentrations, particularly in Xinjiang, Gansu and Qinghai, confirming the result in Fig. 1d. In the other three provinces to the east of NW, we observe larger $\stackrel{\mathrm{‾}}{\mathrm{\Delta }{\mathrm{O}}_{\mathrm{3}}}$ values, suggesting that meteorology, along with precursor emissions, tended to enhance O₃ levels from 1998 to 2017. Consequently, we would expect that meteorology favored O₃ growth in these three eastern provinces of NW.

3.2 Associations of summer ozone with wetting and warming

Figure 2a illustrates the MK-Test statistics of gridded summer O₃ concentrations from 1998 to 2017 subject to the model scenario S2 with fixed emissions and annually varied meteorology, aiming to highlight the effect of meteorology on O₃ evolution. A notable decline in summer O₃ ozone concentrations is evident in most areas of mainland China, except for the Northern China Plain, central China, the Sichuan Basin and the Guanzhong Plain (inner figure on the left-low corner of Fig. 2a). The declining O₃ trends are also evident in most regions across NW with the most substantial decrease occurring in west Gansu, northwest Inner Mongolia and Xinjiang. With fixed emission, such an O₃ falling trend is induced primarily by meteorology (climate). Considering that air temperature and humidity are two key meteorological factors influencing O₃ fluctuations, we present the MK-test statistics of the trends of summer RH (%) at the 2 m height in Northwestern and mainland China (illustrated in the lower-left corner of Fig. 2b), along with their annual variations depicted in a line chart in the lower-right corner of Fig. 2b. We also present the SAT (K) in Northwestern and mainland China in a line chart in the lower-right corner of Fig. 2c. Compared to many places across Eastern and Central China where RH has negative or drying trends (left-low corner of Fig. 2b), RH in NW showed a marked surge, suggesting pronounced wetting, as reported before (Peng and Zhou, 2017). This can also be seen in the annual changes in the mean RH averaged over NW (the line chart on the right-low corner of Fig. 2b) from 1998 to 2017, dominated by an overall upward RH trend with a mean growth rate of 1.9 % yr⁻¹. Similarly, NW also experienced warming during the same period, as shown in Fig. 2c, particularly in the western part of NW, including Xinjiang, western Gansu Province and Qinghai Province. NW has been undergoing one of the most significant warming trends in the nation. Between 2005 and 2017, NW, particularly the Xinjiang region, has warmed by over 0.5 K with a rate of 0.04 K yr⁻¹. In contrast, eastern and Central China show a cooling trend. The rising SAT trend and falling O₃ trend are in contrast to the common knowledge that increasing SATs should enhance O₃ levels. The result thus implies that SATs did not dominate the O₃ trend in NW.

Figure 2 (a) MK-Test statistics for the trends of summer O₃ concentrations from 1998 to 2017 subject to model scenario 2 (S2) with fixed emissions in 1998, (b) MK-Test statistics for summer RH from 1998 to 2017, the line chart on the low-right corner illustrates summer RH in NW from 1998 to 2017 under scenario 2, (c) MK-Test statistics for trend of summer SATs from 1998 to 2017, the line chart shows summer SATs averaged over Northwest China from 1998 to 2017, under scenario 2. Contour figures on the low left corner of panels (a)–(c) illustrates MK-Test statistics for O₃ trends under fixed emission (model scenario 2) and modeled RH and SAT trends in mainland China from 1998 to 2017. The trends are well within the statistically significant range at 95 % confidence level in terms of $\left|Z\right|$ > 1.96 criteria.

Figure 3 (a) Pearson correlation (R) between summer O₃ concentration and RH from 1998 to 2017 under S2 scenario, (b) R between summer OH radical concentration and O₃ concentration from 1998 to 2017 under S2 scenario, (c) summer surface precipitable water (PW) in the bounded area of NW from 1998 to 2017, error bar stands for ±1 SD and red solid line stands for linear trend, (d) summer O₃ concentrations under Scenarios 1 (dark blue solid line with triangles) and 3 (red solid line with circles) scaled on the left y axis, and their fractions $F_{{\mathrm{O}}_{\mathrm{3}}}=(C_{\mathrm{S}\mathrm{1}}-C_{\mathrm{S}\mathrm{3}})/C_{\mathrm{S}\mathrm{3}}\times \mathrm{100}$ % (red dashed line) in the three westernmost provinces of NW (Xinjiang, Qinghai, and Gansu), where C_S1 and C_S3 are the summer (June to August) surface O₃ air concentrations from the base (S1) and fixed meteorology (S3) scenarios, scaled on the right y axis, the blue dashed line in panel (d) is a linear fitting line from 1998 to 2004 and 2004 to 2017, respectively. The green solid line represents the zero-value line of the fraction.

As mentioned earlier, since rising temperatures favor O₃ formation while increasing humidity is detrimental to O₃ growth, it is intriguing to investigate which of these two meteorological factors contributes more to the O₃ trend in NW. We first calculated linear correlation coefficients between RH and O₃ concentrations modeled from the S2 scenario (fixed emissions) from 1998 to 2017 across Northwestern and mainland China (bottom left corner of Fig. 3a). Strong negative correlations are evident in most regions of Xinjiang, Qinghai, and Western Gansu. These negative correlations indicate declining O₃ concentrations with rising RH. To highlight the influences of RH and precipitation on summer O₃ variation, Fig. 3d illustrates modeled mean summer O₃ concentrations averaged over the three westernmost provinces of NW (Xinjiang, Qinghai and Gansu) under the base scenario S1 with annually varying summer meteorology and emissions, and S3 with fixed meteorology in 1998 and annually varying emissions averaged over NW, along with the concentration fractions (%, red dashed line) between the two scenarios from 1998 to 2017. O₃ concentrations exhibit the same increasing trend primarily driven by rising precursor emissions. However, the S1 scenario shows more significant annual fluctuations (solid dark blue line in Fig. 3d) compared to the S3 scenario (solid red line in Fig. 3d). These stronger fluctuations can be attributed to the influence of meteorology. Since scenarios 1 and 3 shared the same emissions and chemistry, the concentration fractions quantify the contributions of meteorology to the O₃ trend alone. As shown, the fractions increased from 1998 to 2004, decreased thereafter, and turned negative after 2009 (the red dashed lines), indicating that meteorology has led to declining O₃ concentrations since 2005 in NW. Among the two key meteorological factors influencing O₃ formation, increasing SAT (Fig. 2c) tends to elevate O₃ levels, which contradicts the decreasing and negative O₃ fractions. Consequently, RH probably plays a more significant role in the O₃ trend in NW, as rising RH aligns well with the decreasing and negative O₃ fractions.

Figure 4 Summer RH, O₃, and OH radical fractions between 1998 and 2017 in six provinces of NW under the fixed emission scenario (S2). (a) RH and O₃ fractions, O₃ fractions are scaled on the left y-axis, and RH fractions are scaled on the right y-axis, (b) OH radical fractions. Error bars denote ±1 SD.

Figure 3c displays the mean summer surface precipitable water (PW, kg m⁻²) over the region bounded between 75–100° E in NW from 1998 to 2017, collected from the NCAR reanalysis (https://psl.noaa.gov/cgi-bin/data/composites/printpage.pl, last access: 13 June 2024). One can identify a slow decrease in PW from 1998 to 2004 and a rapidly growing PW thereafter. Since warming exacerbates drought in arid and sub-arid NW (Lian et al., 2021), growing precipitation in NW implies that increasing humidity might play a more significant role than warming in the changing climate, which acts to reduce O₃ levels via precipitation washout and chemistry. A close look at Figs. 2b, c and 3a indicate that more substantial warming and wetting occurred in Xinjiang, Gansu (including westernmost of Inner Mongolia) and Qinghai provinces. We compared the fractions of provincial mean summer RH and summer O₃ concentrations from the fixed emissions and annually varying meteorology scenario (S2) between 1998 and 2017 (Fig. 4a), defined by $f=(Y_{\mathrm{2017}}-Y_{\mathrm{1998}})/Y_{\mathrm{1998}}\times \mathrm{100}$ %, where f is the concentration or RH fractions and Y₁₉₉₈ and Y₂₀₁₇ stand for O₃ concentrations or RH in 1998 and 2017, respectively, thereby to highlight the net effect from meteorology. The most significant fraction of O₃ concentration was identified in Inner Mongolia at 9.9 %, followed by Ningxia (5.7 %), Shaanxi (5.5 %), Gansu (2.8 %), Xinjiang (1.6 %) and Qinghai (−1.5 %), respectively. Among the six provinces, Xinjiang and Gansu accounted for smaller O₃ fractions, and Qinghai accounted for a negative fraction (Fig. 4a), indicating that O₃ levels in these three provinces inclined less significantly or even declined (Qinghai) between 1998 and 2017 as compared to the other three provinces. Interestingly, more significant O₃ fractions, meaning growing O₃ levels for the two decades in Inner Mongolia, Ningxia and Shaanxi, correspond to smaller (Ningxia) or negative (Inner Mongolia and Shaanxi) RH fractions, meaning falling RH. In contrast, more diminutive and negative O₃ fractions in Xinjiang, Gansu, and Qinghai provinces are associated well with the stronger increasing trend of summer RH by MK-test (Figs. 2b and 4a) at the growing rate of 9.8 % in Xinjiang, 8.9 % in Gansu and 9.9 % in Qinghai from 1998 to 2017, respectively. While the weak or falling summer O₃ levels in three provinces respond to both growing RH and SAT trends (Fig. 2b and c), since growing SATs tend to enhance O₃ concentrations but modeled O₃ respond weakly to rising SAT in these three provinces, the result demonstrates again that the wetting might inhibit O₃ formation and increment for the past decades in NW.

Note that model scenario 2 with fixed emissions also considers chemistry, in which OH radicals and water vapor can interact. The primary source of OH radicals in the atmosphere is the reaction between water vapor (H₂O) and excited oxygen atoms (O(¹D)), which are formed by the photolysis of ozone (O₃) by ultraviolet (UV) radiation. The OH radicals react with ozone to form oxygen (O₂) and the excited oxygen atom O(¹D), which can further react with water vapor to regenerate two OH radicals:

$$\begin{array}{}\text{(R1)}& {\displaystyle }{\mathrm{O}}_{\mathrm{3}}+\mathrm{OH}\to {\mathrm{O}}_{\mathrm{2}}+\mathrm{O}{(}^{\mathrm{1}}\mathrm{D}),\text{(R2)}& {\displaystyle }\mathrm{O}{(}^{\mathrm{1}}\mathrm{D})+{\mathrm{H}}_{\mathrm{2}}\mathrm{O}\to \mathrm{2}\mathrm{OH}.\end{array}$$

This reaction does not lead to a net reduction in OH radicals but contributes to ozone removal by initiating reactions that lead to the formation of less reactive species. Figure 3b is a correlation diagram between modeled summer O₃ concentrations and OH radicals in NW and mainland China (the left-low corner of Fig. 3b) subject to the fixed emission scenario (S2). The spatial pattern of correlation coefficients in NW is, to some extent, similar to that between O₃ concentrations and RH (Fig. 3a), showing the negative associations between OH radicals and O₃ concentrations, particularly in Xinjiang and Qinghai with the most significant wetting trend, suggesting that increasing OH radicals reduce O₃ levels. Figure 4b displays the fractions of provincial mean summer OH radicals from the fixed emissions and annually varying meteorology scenario (S2) between 1998 and 2017. Relatively larger positive fractions can be discerned in Xinjiang, Qinghai and Gansu, meaning increasing OH radicals in these provinces as a result of increasing RH (Eq. 1), which corresponds to higher RH fractions and lower O₃ fractions. An exception occurs in Inner Mongolia, where the large positive OH radical fraction (Fig. 4b) is associated with negative RH and large O₃ fractions (Fig. 4a). OH radicals show an increasing trend (Fig. S13) which agrees with the positive correlation between OH and O₃ concentrations shown in Fig. 3b. This anomaly from the modeling results requires further investigations.

There are several primary mechanisms of OH radicals leading to O₃ reduction, including the reaction with VOCs to form peroxyl radicals (RO₂), the reaction of RO₂ with nitrogen oxides (NO_x = NO + NO₂) to form nitrogen dioxide (NO₂) and other compounds, and photodissociation of NO₂. Figure S10a and b show the MK test estimated NO_x and NMVOC emission trends from 1998 to 2017. While increasing NO_x is observed in NW (Fig. S10a), its emission levels are considerably lower than in Eastern and Southern China (Fig. S10c). Both NMVOC emission levels and trends in NW are significantly smaller than in Eastern and Southern China due to lower populations and less developed economies (Fig. S10b and d). Hence, it is likely that the reaction of water vapor (H2O) with excited oxygen atoms (O(¹D)) produced by the photodissociation of ozone by UV light plays a specific role in the negative correlation between O₃ and OH radicals.

A multiple regression model (MRM) was established to further quantitatively assess the contribution of RH and SAT to summer O₃ concentrations in NW. The MRM has been applied in a number of investigations to quantify meteorological effects on ozone evolution in China (Li et al., 2019, 2020b, 2021a; Yan et al., 2024). For SAT and RH, the MRM model can be defined as

$$\begin{array}{}\text{(2)}& \ln \left({\mathrm{O}}_{\mathrm{3}}\right)=\mathit{\alpha }+{\mathit{\beta }}_{\mathrm{1}}\ln \left(\mathrm{SAT}\right)+{\mathit{\beta }}_{\mathrm{2}}\ln \left(\mathrm{RH}\right)+\mathit{\epsilon },\end{array}$$

where α, β₁ and β₂ are model coefficients, and ε is a random error. In the first instance, we considered three provinces with significant wetting trends: Xinjiang, Gansu and Qinghai. We combined annual summer O₃, RH, and SAT from three provinces into one time series of each dependent and independent variable from 1998 to 2017, yielding 60 samples for each variable. Tables S4–S6 list regression coefficients and statistics of the MRM. To examine the multicollinearity in the MRM, we estimated the variance inflation factor (VIF, Tables S4–S6). The VIFs for RH and SAT are far less than 10, indicating no multicollinearity in the model. The contribution of RH and SAT to O₃ is calculated by:

$$\begin{array}{}\text{(3)}& {\mathrm{Contribution}}_i={\displaystyle \frac{\left|{\mathit{\beta }}_i\times \frac{{\mathrm{SD}}_i}{{\mathrm{SD}}_{\mathit{\alpha }}}\right|}{\mathrm{SUM}}},\end{array}$$

where $\mathrm{SUM}=\sum _i|{\mathit{\beta }}_i\times {\mathrm{SD}}_i/{\mathrm{SD}}_{\mathit{\alpha }}|$, and β_i denotes the regression coefficient of independent variable i (namely, RH and SAT), SD_i is the standard deviation of regression coefficient of independent variable i, and SD_α represents the standard deviation of coefficient of intercept.

The MRM analysis reveals that RH contributes 77.5 %, 59.4 % and 46.2 % to O₃ variations in Xinjiang, Qinghai and Gansu, respectively. Accordingly, SAT contributes 22.5 %, 40.6 % and 53.8 %, respectively, to O₃ fluctuations in these three provinces from 1998 to 2017. The results indicate that RH has become a major contributor to long-term change in O₃ concentrations, overwhelming or equivalent to SAT's contributions to the O₃ trend for the past two decades, probably via interactions among increasing precipitation washout and OH radical, O₃ concentration, and water vapor.

Figure 5 Contribution of PBLH (planetary boundary-layer height), RH, SAT, precipitation, and wind to O₃ evolution. (a) Northwest China, (b) Mainland China.

We further assess the contributions of precipitation, RH, SAT, winds and planetary boundary-layer height (PBLH) to O₃ evolution in NW using an extended MRM model. The results confirm that RH and SAT made the most significant contributions to long-term ozone trends than other meteorological variables under the long-term perspective. As seen from Fig. 5, the contribution of RH (49.5 %) and SAT (47.6 %) to O₃ evolution in NW was almost equivalent (Fig. 5a), but the SAT (92.4 %) dominated in entire mainland China (Fig. 5b). Whereas the PBLH only contributed approximately 2 % to O₃ variations because the PBLH varies on a daily and hourly basis and affects insignificantly (summer) seasonal changes in O₃ concentrations.

Note that since the wetness also helps growing vegetation, which causes increasing biogenic VOC (BVOC) emission and hence rising O₃ concentrations, the contribution of BVOCs to O₃ fluctuations under increasing wetness should also be taken into consideration. To examine this hypothesis, we designed an extra model scenario (S5) in which we used the updated LUC data (Sect. 2.4) and fixed LUC in 1998 in the WRF-Chem simulation and ran the model from 1998 to 2017. Since the LUC defines vegetation coverage, the fixed LUC suggests no vegetation change during this period. We then estimated the fractions between modeled O₃ concentration with fixed LUC and the baseline scenario S1 with annually-varying LUC, thereby eliminating the effects from emissions, meteorology and chemistry but highlighting the LUC (or greenness) effect only. The result is illustrated in Fig. S14. As shown, while the fraction fluctuates positively (Xinjiang) and negatively (Northern Shaanxi and Eastern Inner Mongolia), indicating increasing and decreasing O₃ levels induced by LUC or greening, the magnitudes of provincial mean concentration fractions are very small, mostly in the range −1 % to 0.3 %, suggesting that the LUC under the wetting trend in NW plays an insignificant role in O₃ fluctuation.

3.3 Projected future climate impacts on summer ozone

Given the significant impacts of wetting on O₃ evolution in NW over the past two decades, concerns have also been raised regarding O₃ responses to future wetting and warming. We estimated the projected O₃ concentrations, RH and SAT fractions between the baseline and fixed meteorology scenarios under the SSP2-4.5 and SSP5-8.5 pathways (Sect. 2.6) for the years 2019, 2030, and 2060, respectively (Figs. 6 and 7). As mentioned earlier, since both scenarios implemented the same emissions and chemistry but the baseline scenario accounted for variable meteorology, the fractions (or differences) between the two scenarios eliminated the effects of emissions and chemistry, thereby isolating meteorological signals only.

Figure 6 (a) Summer O₃ concentration fractions of 2019 to 2030 under SSP2-4.5scenario, estimated by $F_{\mathrm{SSP}\mathrm{245}}=(C_{\mathrm{2030}}-C_{\mathrm{2019}})/C_{\mathrm{2019}}\times \mathrm{100}$ %, where C₂₀₁₉ and C₂₀₃₀ are the summer surface O₃ air concentrations in 2019 and 2030, respectively, (b) same as panel (a) but for O₃ fractions of 2019 to 2060, (c) same as panel (a) but for summer RH fractions of 2019 to 2030, (d) same as panel (a) but for summer RH fractions of 2019 to 2060, (e) fractions of RH, SAT, and O₃ of 2019 to 2030 and 2019 to 2060 (shown in the same color bar as 2030 but with oblique lines) in the six provinces of NW under the SSP2-4.5 scenario.

Figure 7 (a) Summer O₃ concentration fraction of 2019 to 2030 under the SSP5-8.5 scenario, estimated by $F_{\mathrm{SSP}\mathrm{585}}=(C_{\mathrm{2030}}-C_{\mathrm{2019}})/C_{\mathrm{2019}}\times \mathrm{100}$ %, where C₂₀₁₉ and C₂₀₃₀ are the summer surface O₃ air concentrations in 2019 and 2030, respectively, (b) same as panel (a) but for O₃ fraction of 2019 to 2060, (c) same as panel (a) but for summer RH fraction of 2019 to 2030, (d) same as panel (a) but for summer RH fraction of 2019 to 2060, (e) fractions of RH, SAT, and O₃ of 2019 to 2030 and 2019 to 2060 (shown in the same color bar as 2030 but with oblique lines) in the six provinces of NW under the SSP5-8.5 scenario.

Detailed discussions on the associations between projected O₃ evolution and trends of RH and SAT under the SSP2-4.5 and SSP5-8.5 pathways from 2019 to 2030 and from 2019 to 2060 are presented in Sect. S4 in the Supplement and Figs. 6–7 and S15, respectively. Overall, from 2019 to 2030, we observe a stronger negative association between O₃ and RH under SSP5-8.5 than under SSP2-4.5 across NW, suggesting that climate conditions under the SSP5-8.5 scenario will more effectively suppress O₃ growth compared to SSP2-4.5 during this period. However, from 2019 to 2060, the SSP5-8.5 tends to project higher O₃ levels than SSP2-4.5. From a wetting trend perspective, we must determine which pathway would project a more realistic future humidity trend in NW. China has been making significant strides in transitioning to renewable energy sources, reducing its carbon footprint, and committing to achieve a carbon peak before 2030 and carbon neutrality by 2060. These commitments align more closely with the mitigation efforts outlined in SSP2-4.5, which, as manual interventions, will overshadow the impact of climate change on O₃ fluctuations. There have also been debates regarding ongoing wetting trends in NW based on CMIP6 projections under SSP5-8.5 and historical data analysis (Huang et al., 2017). Historically, China's rapid economic growth has heavily depended on fossil fuels, which may render SSP5-8.5 more applicable in China. The responses of O₃ evolution to RH under SSP5-8.5 are more akin to those observed from 1998 to 2017 (Sect. 3.2) than to SSP2-4.5. In light of this, while China may achieve its carbon peak and neutrality goals, the wetting trend in NW will probably persist in the coming decades, exacerbating continuous O₃ pollution in NW, in addition to manual interventions. Under SSP5-8.5, the projected increase in humidity (e.g., 34.9 % in Xinjiang by 2030) may enhance O₃ wet deposition, partially offsetting the warming-driven O₃ rise. This aligns with Li et al. (2014), who identified humidity as a critical factor modulating cross-regional ozone transport efficiency.

4 Conclusions

This study conducted extensive sensitivity model simulations to explore the surface O₃ responses to past and future climate change in NW. The results demonstrated that significant wetting since the 21st century in this region, particularly in the three westernmost provinces (Xinjiang, Qinghai and Gansu) of NW, has slowed O₃ growth. We show that in these provinces, the wetting influence on O₃ evolution has outweighed the warming effect and played an almost equal role to warming across the entire NW. Both factors contributed approximately 50 % to summer O₃ variations from 1998 to 2017, among the major meteorological factors causing changes in O₃ concentrations. This finding contrasts with the dominant role of warming in the meteorological effects on O₃ evolution in China. We linked the increasing OH radicals, which interact with water vapor, to the declining O₃ trend. Given that humidification in NW is projected to continue in the coming years and that the mitigation of O₃ precursor emissions in China will still be implemented under the nation's air pollution control strategies, the impact of humidification on O₃ trends in NW will probably become more significant. Such an impact should be considered in future policymaking for the reduction of O₃ pollution in NW. While this study advances our understanding of ozone-climate interactions in NW, several uncertainties warrant attention. First, the reliance on static emission inventories (e.g., EDGAR) may underestimate dynamic sources such as soil NO_x emissions, which are critical in arid region. Second, the 20 km resolution of WRF-Chem may inadequately resolve fine-scale topographic effects on local meteorology (e.g., valley breezes in mountainous NW), potentially biasing ozone simulations. Future work should integrate high-resolution emission inventories to improve accuracy. In addition, ground-based ozone measurements in remote NW remain sparse, limiting robust model validation. Expanding monitoring networks will be essential to constrain uncertainties.