Model Analyses of Changes in Spring Surface Ozone Concentrations over Shandong Province in the Period of 2014‒2017

The concentrations of surface ozone (O 3 ) in eastern China have increased significantly in recent years, resulting in the earlier appearance of more serious O 3 pollution. Measurements at 14 stations in Shandong Province showed that the monthly mean O 3 concentrations in the late spring (May) increased by 22.2 µ g m –3 from 2014 to 2017. To investigate the reasons of the increase of O 3 in springtime from 2014 to 2017, the changes of O 3 concentrations due to meteorological conditions and emissions in May were studied based on ambient measurements and simulations with the RAMS-CMAQ modeling system. By analyzing the observed data, it was found that the variations in wind field were conducive to the accumulation of O 3 , while the effects of other meteorological parameters on O 3 concentrations were different at the same site between years. Further to perform a series of simulations with only the meteorological conditions changed in May from 2014 to 2017, the results showed that the effects of variations in meteorological conditions had become more and more important in the changes of O 3 concentrations in May between years, especially the factors that affected the photochemical generation of O 3 . For example, the percentage of the sites where the changes of O 3 concentrations were dominated by the variations of the meteorological conditions increased from 28.6% to 78.6% over the region in May between years. Besides, the changes in NO x emissions had a close relationship with the variations in O 3 concentrations when the changes of O 3 were dominated by the emission variations.


INTRODUCTION
Tropospheric ozone (O3), a major oxidant, is mainly produced by the photochemical oxidation between the volatile organic compounds (VOCs) and nitrogen oxides (NOx). If ozone concentration in the troposphere exceeds the natural level, it has adverse effects on human health (Canella et al., 2016), vegetation (Feng et al., 2015;Van Dingenen et al., 2009) and climate (Stevenson et al., 2013;Worden et al., 2008). According to the available monitoring data from 1950-1979 until 2000-2010 for the Northern Hemisphere, surface O3 concentration has increased globally during the 20 th century with an increase of 1-5 ppbv per decade (Sun et al., 2019). Thus, attentions have been attracted to tropospheric O3 worldwide (e.g., Marais et al., 2014;Monks et al., 2015;Zeng et al., 2018).

Observational Data
We collected the measured surface O3, NO2 and NOx concentrations at 14 monitoring sites ( Fig. 1) in Shandong Province from January 2014 to December 2017 from the CNEMC. The corresponding meteorological factors (including temperature [T], relative humidity [RH], wind speed [WS] and wind direction [WD]) were from the Meteorological Information Comprehensive Analysis and Process System (MICAPS). The measured data of cloud fraction (CF) was obtained from the MODerate Resolution Imaging Spectroradiometer (MODIS) through https://ladsweb.modaps. eosdis.nasa.gov/search/ (last accessed 19 April, 2021).

Model Description
The version of CMAQv4.7.1 coupled with the gas-phase photochemical mechanism SAPRC99 (1999 Statewide Air Pollutant Research Center) (Carter, 2000) was used to simulate and trace the evolution of the concentrations of pollutants. Different inventories were combined to make the emission sources. The anthropogenic emissions over China were from the Multi-resolution Emission Inventory for China (MEIC) of 2014 (0.25° × 0.25°) (www.meicmodel.org, last accessed 17 November 2019), while the emissions of surrounding countries were from the MIX inventory (Li et al., 2017). The biogenic emissions provided by the Model of Emissions of Gases and Aerosols from Nature (MEGAN) (Guenther et al., 2012) were derived from the Emissions of atmospheric Compounds and Compilation of Ancillary Data (ECCAD) database (https://eccad3.sedoo.fr, last accessed 8 October, 2021), namely MEGAN-MACC Biogenic emission inventory (0.5° × 0.5°, the year of 2010). The open biomass burning emissions of 2014 were collected from the Global Fire Emissions Database, Version 4 (Randerson et al., 2015). The monthly soil emissions of NOx were derived from the Regional Emission inventory in ASia, Version 2.1 (0.25° × 0.25°, the year of 2008) (Kurokawa et al., 2013), while the monthly lightning NOx emissions were obtained from the Global Emissions Inventory Activity (1° × 1°, the year of 2000) (Benkovitz et al., 1996). According to Han et al. (2004) and Gong et al. (2003), the online calculation of dust and sea salt emissions were added in the model. The output of RAMS provided the meteorological inputs to drive CMAQ. The initial and lateral boundary conditions used for the RAMS were from the National Center for Environmental Prediction reanalysis datasets. Fig. 1 presents the two nested domains used in our simulations. The outer domain (D01), covering most of East Asia, was divided into 105 × 86 grid cells with a horizontal resolution of 64 km. The center of D01 was located at (35°N, 110°E). The nested domain (D02) was a 16 km × 16 km horizontal resolution domain that covered the North China Plain (1504 km × 1440 km) with the center located at (40°N, 116°E).   Four sensitivity experiments (Table 1) were designed to analyze the effects of emissions and meteorological conditions on pollutant concentrations over Shandong Province in May during the time period 2014-2017 with the RAMS-CMAQ modeling system.

Sensitivity Experiments
We used the changes in the observed pollutant concentrations between different time spaces as the total changes (Eq. (1)). We kept the emissions of the CMAQ simulations unchanged between years and only changed the meteorological inputs. Thus, the changes in the predicted pollutant concentrations were due to differences in the meteorological conditions (Eq. (2)). The changes caused by different emissions were obtained by excluding the changes due to meteorological conditions from the total changes (Eq. (3)). The scheme was proved to be reasonable in Wang et al. (2019b).
∆Ei,j = ∆Oi,j -∆Mi,j where i represents the pollutant; j and k represent the year (j > k); ∆Oi,j is the total changes in the observed concentration of pollutant i from year k to j; Obsi,j and Obsi,k represent the observed concentration of pollutant i in year j and k, respectively; ∆Mi,j is the change in the simulated concentration of pollutant i in year j due to changes in the meteorological conditions compared to year k; Simi,j and Simi,k represent the modeled concentration of pollutant i in year j and k, respectively; and ∆Ei,j is the difference in the simulated concentration of pollutant i between the year j and k due to the changes in emissions. The contribution of the meteorological conditions and the changes in emissions cannot be completely differentiated as a result of the complex atmospheric processes and therefore our results were only approximations. This analysis was used to give an overview of Shandong Province.

Model Evaluation
To evaluate the reliability of the modeled results, we compared the modeled and observed meteorological factors (including temperature, relative humidity, wind direction and wind speed) in May of each year and the hourly concentrations of pollutants (including O3, NO2 and NOx) in May 2014 derived from 14 monitoring sites over Shandong (Tables 2 and 3). Table 2 shows statistical results of the meteorological parameters, where: N presents the total number of samples; Csim and Cobs are the averaged results of simulations and observations, respectively; MB is the mean bias; RMSE and GE are the root-mean-square and gross error, respectively; R is the correlation coefficient between the measured and simulated values; P22.5° and P45° represent the percentages of compared points at which the absolute biases between the modeled and measured wind directions are within 22.5° and 45°, respectively.
As shown in Table 2, the model reproduced the variation and magnitude trend of the temperature and relative humidity quite well according to the statistical results. For T, though the absolute GE values were a little higher than the benchmark (2.0), the absolute MB values were within the benchmark (0.5) suggested by Emery et al. (2001) and the values of R were no less than 0.8. Besides, the RMSE of T were comparable to that in Gao et al. (2016), also suggesting a reasonable simulation of T. While for RH, MB and R values were no more than 2.5% and no less than 0.7, respectively. Compared to Wang et al. (2019b), RMSE and GE values for RH in this study were even smaller. All these indicated the good reproduction of the relative humidity by the model. For WS, both the values of GE and MB met the benchmarks (GE ≤ 2.0 and |MB| ≤ 0.5), suggesting a better simulation than Feng et al. (2016). The values of RMSE for WS were also within the  benchmark (≤ 2.0) for a good performance. For WD, P22.5° and P45° were larger than 40% and 60%, which was comparable to Li et al. (2019b), indicating the reasonable simulation of wind directions. Thus, the model was performed well to provide a reasonable meteorological field.
The evaluated results for the modeled O3, NO2 and NOx concentrations are shown in Table 3. MB values of O3, NO2 and NOx in May of 2014 were small (5.9, -1.8 and -4.5 µg m -3 ) with R values equaling 0.5, 0.6 and 0.6, respectively. Besides, the normalized mean bias (NMB) and the normalized mean error (NME) values of O3 were comparable to those reported by Wang et al. (2019a), while NMB and NME values of NO2 were smaller than the results of Wang et al. (2019a). Thus, the simulated pollutant concentrations were reliable.

RESULTS AND DISCUSSION
According to previous studies (e.g., Atkinson, 2000;Meleux et al., 2007), temperature, relative humidity, wind speeds and cloud fraction played important roles in O3 formation. Thus, we first examined the relationships between the observed 90th percentile of hourly O3 concentration (O3-h_90) and the measured temperature, relative humidity, wind field and cloud fraction in May at the 14 monitoring sites during the time period 2014-2017. Fig. 3 shows the variations of observed O3-h_90, temperature, relative humidity and CF, while Fig. 4 shows the statistical results for the observed wind direction and wind speed.    Fig. 3(b) shows that the temperature decreased and the relative humidity either increased or remained stable in May 2015 compared with 2014. According to the previous studies (e.g., Atkinson, 2000;Hu et al., 2008;Zhang et al., 2015), lower T and higher RH were unfavorable for ozone production. The rate of ozone formation was repressed in May 2015 compared with 2014 due to the increase in CF, which can reduce the intensity of surface illumination (Fig. 3(c)). We therefore concluded that the temperature, relative humidity and CF in May 2015 reduced the formation of ozone compared with 2014 over Shandong.
O3-h_90 in May 2016 decreased at most sites compared with 2015 (except Dongying, Laiwu, Tai'an, Zibo, Qingdao and Weihai) (Fig. 3(a)). A significant increase in the CF (Fig. 3(c)) led to a decrease in O3 at all sites in May 2016 compared with 2015. Different from the CF, the effects from the changes in temperature and relative humidity were not consistent at all sites. However, it was easy to find that both the two factors had positive effects on the formation of O3 at most sites in May 2016 compared with 2015 ( Fig. 3(b)) due to the decrease of relative humidity and slight increase of temperature.
O3-h_90 in May 2017 ( Fig. 3(a)) significantly increased at most sites compared with 2016 (except Weihai and Zibo). Cloud fraction can affect the surface temperature and photochemical production of surface ozone due to its impacts on the amount of insolation (Meleux et al., 2007;Lee et al., 2015); thus the decreased cloud fraction (Fig. 3(c)) at most sites (except Weihai) contributed to the increase of O3 production. Besides, the decrease in the relative humidity ( Fig. 3(b)) at most sites in May 2017 compared with 2016 was also conducive to the formation of O3.
As shown in Fig. 4, the observed distribution of winds in May was similar between the years with the southerly winds as the main winds. The continuous reduction in the mean wind speed (i.e., 2.7, 2.4, 2.3, and 2.3 m s -1 ) of May from 2014 to 2017 can result in the less air mass of O3 taken away from Shandong Province. Thus, the changes in wind field were conducive to the accumulation of O3 in May during the time period 2014-2017.
In conclusion, the variations of meteorological parameters had different effects on the formation of O3 in May between years. It was easy to identify the individual effect of the changes in each meteorological factor on O3 production at each site, but the combined effects from all the meteorological conditions could not be determined. It was necessary to use the modeling system to evaluate the integrated effects.
Thus, secondly, as described in Section 2.3, four sensitivity experiments (Table 1)  Combined with the results for ∆Oozone, it showed that changes in emissions dominated the variations in total O3 concentrations in May 2015 over most regions of Shandong. For example, ∆Oozone was positive and ∆Mozone was negative at Dongying and therefore the final changes in the concentration of O3 in May 2015 were dominated by the effects of emissions. As another example, both ∆Oozone and ∆Mozone were positive at Liaocheng, but ∆Mozone contributed < 30% of ∆Oozone. Thus, the dominant effects were still from the changes in emissions. There were 10 such sites and most of the sites were in northwestern and central Shandong. While O3 variations at other four sites (28.6%) were dominated by the changes of meteorological conditions. In Fig. 5(b), ∆Mozone and ∆Eozone in May between 2016 and 2015 varied from -34.1 to 43.2 µg m -3 and from -47.6 to 123.3 µg m -3 , respectively. By contrast with 2015, the variations in meteorological conditions in May 2016 resulted in a decrease in O3 (-34.1 to -4.6 µg m -3 ) at most sites (except Qingdao, Weihai and Rizhao). Similarly, according to the analyzed results from Figs. 3 and 4, except the cloud fraction, the changes of temperature, relative humidity and wind field were conducive to the increase of O3 at most sites over Shandong. We therefore concluded that the cloud fraction played a dominant role in the effects of changes in meteorological conditions on O3 concentration through its impacts on the photochemical generation of O3, especially the regions over northwestern, central and most parts of southern Shandong. Combined with the results for ∆Oozone, this showed that there were eight sites (57.1%) at which the variations in the concentration of O3 were dominated by the changes in emissions. Compared with May 2015, the effects of meteorological conditions on the concentration of O3 increased over Shandong Province in 2016 (6 sites, 42.9%). Fig. 5(c) showed that ∆Mozone and ∆Eozone in May between 2017 and 2016 varied from -20.4 to 43.1 µg m -3 and from -35.7 to 28.6 µg m -3 , respectively. The differences in meteorological conditions in May 2017 compared with May 2016 resulted in an increase in O3 at most sites (7.6-43.1 µg m -3 ) (except Weihai and Rizhao). Combined with the results for ∆Oozone, it was clear that the differences of meteorological conditions contributed more to the changes of O3 concentration than that of emissions at most sites (11 sites, 78.6%) in May 2017. ∆Mozone was positive at 10 of the 11 sites. According to the analyzed results from Figs. 3 and 4, except the relative humidity, the changes of temperature and cloud fraction had positive effects on O3 formations at nearly all the NOx is one of the photochemical precursors of O3. Previous studies (e.g., Wang et al., 2019a) presented that changes in the formation of O3 were closely related to variations in NOx emissions over eastern China, thus we further investigated the effects of meteorological conditions and emissions on NOx concentrations, respectively. Fig. 6 shows the changes in the monthly mean NOx concentrations (∆ONOx) in May between two adjacent years resulting from changes in the meteorological conditions (∆MNOx) and emissions (∆ENOx) during the time period 2014-2017.
As shown in Fig. 6(a), the concentration of NOx decreased (-33.9 to -0.7 µg m -3 ) at most sites in May 2015 compared with May 2014, which was dominated by the effects of changes in NOx emissions over most regions of Shandong (|∆ENOx| > |∆MNOx| at 10 sites). Combined with the analysis from Fig. 5(a), ∆Eozone dominated the changes in total O3 concentrations in May 2015. There were seven sites where both ∆Eozone and ∆ENOx contributed more to corresponding changes in the concentration of pollutants (e.g., Dongying, Tai'an and Qingdao). By comparing ∆Eozone and ∆ENOx at the seven sites, the variation tendency was same at five sites between ∆Eozone and ∆ENOx, suggesting that the variations in O3 were closely related to the changes in NOx emissions in May 2015 compared to 2014.
As presented in Fig. 6(b), the changes in NOx emissions still dominated the variations in NOx concentrations at most sites (11 sites) in May 2016 compared to 2015. Combined with the results shown in Fig. 5(b), ∆Eozone also dominated the variations in O3 formation in May 2016 compared with 2015. There were seven sites where both ∆Eozone and ∆ENOx contributed more to corresponding changes in the concentration of pollutants (e.g., Dongying, Zibo and Rizhao). By comparing ∆Eozone and ∆ENOx at the seven sites, the variation tendency was same at six sites between ∆Eozone and ∆ENOx, suggesting that the variations in O3 were still closely related to the changes in NOx emissions in May 2016 compared to 2015.
In Fig. 6(c), NOx concentrations decreased at most of the sites (except Heze) in May 2017 compared with 2016, which was still dominated by the changes in NOx emissions. However, combined with the results shown in Fig. 5(c), the variations in meteorological conditions that affected the photochemical generation of O3 dominated the changes of O3 concentrations. Therefore, the variations in NOx emissions may have relatively small effects on the changes in O3 concentrations.
In conclusion, the changes in NOx emissions had a close relationship with the variations in O3 concentrations when the changes in emissions dominated the variations of O3 concentrations in Shandong.

CONCLUSIONS
The changes of surface O3 concentrations due to meteorological conditions and emissions in the late springtime (May) of Shandong during the time period 2014-2017 were discussed based on ambient measurements and simulations with the RAMS-CMAQ modeling system. Based on the measurements, an increase was observed in O3 pollution over Shandong during the time period 2014-2017. The peak period of O3 pollution gradually changed from a broad (May-August) in 2014 to a narrow (May and June) range in 2017, while the observed monthly mean O3 concentrations during the peak period changed from 90.1-91.7 µg m -3 in 2014 to 112.3-117.8 µg m -3 in 2017. There was a significant and unexpected increase of O3 in May 2017 compared to 2016, indicated that more serious ozone pollution appeared in the late springtime over Shandong Province. Further to analyze the effects of each meteorological factor on O3 concentrations in May between years, it was found that the changes in wind field were conducive to the accumulation of O3, while the effects of other meteorological factors on O3 concentrations were different at the same site between years.
In order to identify the effects of the changes in O3 concentrations due to meteorological conditions and emissions in May during the time period 2014-2017, four sensitivity simulations were designed and performed with RAMS-CMAQ. The simulation showed that the effects of changes in meteorological conditions became more and more important in the variations of O3 concentrations in May during the time period 2014-2017, especially the factors that affected the photochemical generation of O3. For instance, the percentage of the sites where O3 variations were dominated by the changes of the meteorological conditions increased from 28.6% to 78.6% over the region in May between years. The effects of meteorological parameters on O3 concentrations were different and were positive (11.8-35.2 µg m -3 ), negative (-34.1 to -4.6 µg m -3 ) and positive (7.6-43.1 µg m -3 ) at most sites in May between years. By analyzing the effects of the changes in NOx concentrations due to meteorological conditions and emissions, it was found that the changes in NOx emissions had a close relationship with the variations in O3 concentrations when the changes of O3 were dominated by the emission variations in Shandong.