Numerical Sensitivity Tests of Volatile Organic Compounds Emission to PM2.5 Formation during Heat Wave Period in 2018 in Two Southeast Korean Cities

A record-breaking severe heat wave was recorded in southeast Korea from 11 July to 15 August 2018, and the numerical sensitivity simulations of volatile organic compound (VOC) to secondarily generated particulate matter with diameter of less than 2.5 μm (PM2.5) concentrations were studied in the Busan and Ulsan metropolitan areas in southeast Korea. A weather research and forecasting (WRF) model coupled with chemistry (WRF-Chem) was employed, and we carried out VOC emission sensitivity simulations to investigate variations in PM2.5 concentrations during the heat wave period that occurred from 11 July to 15 August 2018. In our study, when anthropogenic VOC emissions from the Comprehensive Regional Emissions Inventory for Atmospheric Transport Experiment-2015 (CREATE-2015) inventory were increased by approximately a factor of five in southeast Korea, a better agreement with observations of PM2.5 mass concentrations was simulated, implying an underestimation of anthropogenic VOC emissions over southeast Korea. The simulated secondary organic aerosol (SOA) fraction, in particular, showed greater dominance during high temperature periods such as 19–21 July, 2018, with the SOA fractions of 42.3% (in Busan) and 34.3% (in Ulsan) among a sub-total of seven inorganic and organic components. This is considerably higher than observed annual mean organic carbon (OC) fraction (28.4 ± 4%) among seven components, indicating the enhancement of secondary organic aerosols induced by photochemical reactions during the heat wave period in both metropolitan areas. The PM2.5 to PM10 ratios were 0.69 and 0.74, on average, during the study period in the two cities. These were also significantly higher than the typical range in those cities, which was 0.5–0.6 in 2018. Our simulations implied that extremely high temperatures with no precipitation are significantly important to the secondary generation of PM2.5 with higher secondary organic aerosol fraction via photochemical reactions in southeastern Korean cities. Other possible relationships between anthropogenic VOC emissions and temperature during the heat wave episode are also discussed in this study.


Introduction
Particulate matter with diameter of less than 2.5 µm (PM 2.5 ) concentrations are affected by both local emissions and the long-range transport of air pollutants [1][2][3]. Under stagnant meteorological conditions, local emissions of precursors primarily affect the formation of high PM 2.5 concentrations via the accumulation of both directly emitted air pollutants and secondary formation [4,5]. In Korea, stagnant atmospheric conditions typically occur with weak horizontal and vertical mixing processes. Further, meteorological variables such as wind speed, temperature, precipitation, boundary layer effects on air quality in the BMA and UMA during the record-breaking heat wave period in 2018. Other discussions, such as the possible relationship between anthropogenic evaporative VOC emissions and temperature during the heat wave period, are also given in this study.

Study Area and Geographic Features
BMA and UMA are both located in southeast Korea. Busan, the second largest city in Korea, is located on the southeastern tip of the Korean Peninsula, with an area of 769.89 km 2 and a current population of approximately 3.48 million [42]. Busan has complex terrain including an irregular coastline and moderately high mountains. The urban center is situated in a valley away from the coastline; the valley floor is covered with tall buildings and roads. The BMA has various pollutant emission sources such as vehicles, industrial facilities, ships, and urban activities. The highest emissions occur in the industrialized area and the downtown area about 4 km from the coastline. In the BMA in 2015, total annual emissions of approximately 43,755 tons of NO X and 42,207 tons of VOCs were reported from point, line, and area sources [43]. A large portion of these emissions were estimated to originate from heavy traffic. There were 1,255,722 registered vehicles in Busan in 2015 [42].
Ulsan currently has a population of approximately 1.18 million people, occupies an area of 1,061.18 km 2 , and has one of the largest metropolitan areas in Korea [44]. Ulsan was designated as Special Industrial District in 1962 and became a metropolitan city in 1997. Ulsan represents one of Korea's largest industrial clusters for the automotive, shipbuilding, maritime, and petrochemical industries, located at the southeastern end of the Korean Peninsula. The UMA is therefore a typical seaport town with a power plant and industrial areas. According to the statistics for 2015, the UMA had total annual emissions of approximately 47,506 tons of NO X and 98,781 tons of VOCs from point, line, and area sources [43], and large portions of these emissions were estimated to originate from heavy traffic; there were 525,092 registered vehicles in Ulsan in 2015 [44]. This vehicle density implies that the air in both the BMA and UMA is highly polluted, related to potentially secondary air pollutants.
Since the early 1990s, both the BMA and UMA have experienced substantial increases in surface O 3 concentrations [45]. The prevailing synoptic winds over the two metropolitan areas are generally north-westerlies in winter and south-westerlies in summer. During the warm season (i.e., from April to September), a well-developed land-sea breeze is a prominent feature over both areas, frequently emerging under weak synoptic conditions.

Model, Input Data, and Simulation Experiments
For a diagnostic modeling study of the two metropolitan areas, version 3.8.1 of the WRF-Chem model was employed to simulate the impact of heat wave occurrence on the formation of secondary PM 2.5 . The employed model, WRF-Chem, is an online-coupled meteorology-chemistry-aerosol model that is able to simulate trace gases and aerosols simultaneously with the meteorological fields. The meteorological component, WRF, is a fully compressible and non-hydrostatic meso-scale numerical weather prediction model [46]. The WRF model physics options were pre-selected for this study. We employed identical transport and physics schemes for both the grid scale and sub-grid scale, and equally applied them to the air quality component and meteorological component [47,48]. The main configurations for the physical and chemical schemes adopted in this study are listed in Table 1. Figure 1 provides the WRF-Chem grid definition. In the WRF-Chem model, all grids were defined on a Lambert conformal conic (LCC) projection centered at 38 • N, 126 • E with true latitudes at 30 • N and 60 • N. Figure 1a shows the modeling area with three domains: one mother domain (Domain 1: 174 × 126 grids with 27 km resolution) with two nested domains (Domain 2: 96 × 135 grids with 9 km resolution, and Domain 3: 195 × 213 grids with 3 km resolution). The 31 vertical layers used varied in thickness with 1 hPa at top height. There are 10 layers within 1.5 km height in the WRF-Chem model. Figure 1b displays the WRF-Chem 3 km resolution modeling grid (i.e., Domain 3) with the locations of air quality monitoring sites and meteorological measurement sites, and Figure 1c focuses on the locations of sites in both the BMA and UMA in Korea.  Meteorological observations within both the BMA and UMA were obtained from the KMA. Meteorological variables, including wind speed/direction, precipitation, cloud cover, and temperature were observed at the sites indicated in Figure 1, including a site located nearest to the sea in the BMA and the central urban site in the UMA. Air quality measurements, including hourly PM2.5 mass concentrations, were gathered from the air quality monitoring sites indicated in Figure 1,  Meteorological observations within both the BMA and UMA were obtained from the KMA. Meteorological variables, including wind speed/direction, precipitation, cloud cover, and temperature were observed at the sites indicated in Figure 1, including a site located nearest to the sea in the BMA and the central urban site in the UMA. Air quality measurements, including hourly PM 2.5 mass concentrations, were gathered from the air quality monitoring sites indicated in Figure 1, which are operated by the Korean Ministry of Environment (MOE). There are 18 air quality monitoring sites in BMA and 14 in UMA; these cover downtown, commercial, traffic, industrial, harbor, and residential areas. The observed PM 2.5 concentration data were used to analyze the movement of polluted air and to compare against the simulated surface concentrations from WRF-Chem modeling.
WRF-Chem requires hourly, gridded, and speciated VOC emissions as an input data. The latest version of the Comprehensive Regional Emissions for Atmospheric Transport Experiment-2015 (CREATE-2015) emissions dataset was used in this study. The anthropogenic emissions of sulfur dioxide (SO 2 ), nitrogen oxides (NO X ), carbon monoxide (CO), VOCs, black carbon (BC), organic carbon (OC), PM 10 , and PM 2.5 were based on the CREATE-2015 emission data. CREATE-2015 includes the Multi-resolution Emission Inventory of China (MEIC) [49], Clean Air Policy Support System (CAPSS) [50], and Regional Emission Inventory in Asia (REAS) version 2 [51]. Biogenic emissions were calculated online using a model based on the Model of Emissions of Gases and Aerosols from Nature (MEGAN) inventory [41]. MEGAN has been fully coupled into WRF-Chem to enable the online calculation of biogenic precursor emissions subject to vegetation cover and meteorological conditions such as temperature and solar radiation at the time of the calculation [47]. Dust, sea salt, and dimethyl sulfide (DMS) emissions were not included in this study.

Case Description
As a case study, we focused on the heat wave advisory days with no consecutive precipitation days under stagnant synoptic condition, and the heat wave alert/warning period was carefully selected. We analyzed the meteorological features and PM 2.5 concentrations observed during the case study period and compared them with those observed over the last four years (from 2015 to 2018, since PM 2.5 concentration data has been officially collected from January 2015, in Korea). Figure 2 shows the time series of daily mean PM 2.5 concentrations for both the heat-wave alert and wet precipitation periods. We also denoted a period governed by a stationary front located over the Korean Peninsula, called the "Changma" period, from 2015 to 2018. Changma is generally observed in July and August over the Korean Peninsula with an immense amount of precipitation. Daily average concentrations exceeding 35 µg/m 3 (the criterion for a PM 2.5 'bad' grade defined as unhealthy level by Korean operational air quality prediction agency) are highlighted in Figure 2. Shaded bars represent precipitation days, and the "Changma" period is indicated. As shown in Figure 2, the rainy period (Changma period) was much shorter in 2018 than in other years, and the number of precipitation days in July, 2018 was recorded to be much lower than that in normal years ( Table 2). The black line in Figure 2 indicates the days of heat wave alerts/warnings. Large numbers of heat wave alerts/warnings were issued in both cities (Busan and Ulsan) in both 2016 and 2018, but they were more frequent and continuous in 2018 than in 2016 in both cities. In particular, heat wave alerts/warnings persisted for a month in both cities in 2018, which broke the record duration set in 1973 in Korea [10]. In this regard, the period from 12 to 21 July 2018 was selected as a case study period (indicated in Figure 3) during which a heat-wave alert was consecutively issued with no precipitation under stagnant synoptic atmospheric conditions. As shown in Figure 3, a center of high pressure continuously overlaid the Korean Peninsula during the study period, and these stagnant atmospheric conditions offered sufficient circumstances to cause the accumulation of locally emitted pollutants and to generate secondary pollution. In this regard, the period from 12 to 21 July 2018 was selected as a case study period (indicated in Figure 3) during which a heat-wave alert was consecutively issued with no precipitation under stagnant synoptic atmospheric conditions. As shown in Figure 3, a center of high pressure continuously overlaid the Korean Peninsula during the study period, and these stagnant atmospheric conditions offered sufficient circumstances to cause the accumulation of locally emitted pollutants and to generate secondary pollution.
The sensitivity of secondary aerosol formation depends on both atmospheric NOx and VOC emissions. In a previous study, NOx emission evaluation was reported [52] during the Korea-US Air quality (KORUS-AQ) Campaign based on a top-down assessment using an ozone monitoring instrument (OMI) whereby NO 2 suggested an underestimation of 53% during the WRF-Chem simulations (NOx emissions by a factor of 2.13). In addition to Goldberg et al., [52], due to the remaining uncertainties, Miyazaki et al., [53] suggested the possibility of underestimation of NO 2 emission when using top-down approaches, despite employing spatially more continuous and temporally more frequent data than bottom-up approach. However, anthropogenic VOC emission evaluation and OA generation from anthropogenic VOC emissions are poorly understood in modeling studies [54][55][56][57][58]. VOC emission evaluation has been attempted in several studies [54,55] by employing the diagnostics modeling through analyzing the discrepancies between modeled and observed concentrations. This process is of great importance for model performance because SOA formation is associated with VOC emissions and is known to be enhanced by active photochemical reaction conditions under high temperature, which leads to an increase in PM 2.5 concentrations.
Atmosphere 2020, 11, x FOR PEER REVIEW 8 of 21 The sensitivity of secondary aerosol formation depends on both atmospheric NOx and VOC emissions. In a previous study, NOx emission evaluation was reported [52] during the Korea-US Air quality (KORUS-AQ) Campaign based on a top-down assessment using an ozone monitoring instrument (OMI) whereby NO2 suggested an underestimation of 53% during the WRF-Chem simulations (NOx emissions by a factor of 2.13). In addition to Goldberg et al., [52], due to the remaining uncertainties, Miyazaki et al., [53] suggested the possibility of underestimation of NO2 emission when using top-down approaches, despite employing spatially more continuous and temporally more frequent data than bottom-up approach. However, anthropogenic VOC emission evaluation and OA generation from anthropogenic VOC emissions are poorly understood in modeling studies [54][55][56][57][58]. VOC emission evaluation has been attempted in several studies [54,55] by employing the diagnostics modeling through analyzing the discrepancies between modeled and observed concentrations. This process is of great importance for model performance because SOA formation is associated with VOC emissions and is known to be enhanced by active photochemical reaction conditions under high temperature, which leads to an increase in PM2.5 concentrations. In this study, as the scaling factor for NOx emission has been applied in CREATE-2015 emission inventory, we only conducted VOC emission sensitivity study because the uncertainty of VOCs likely remains high, especially during heat wave alert periods such as the current heat-wave study period. For a better investigation of the impact of VOC on the secondary formation of PM 2.5 during the heat wave period, three more experiments were conducted: doubling, five-fold increase, and ten-fold increase of VOC emission, in addition to a control experiment. This is because the effects of high temperature on SOA generations have not been quantified, and thus VOCs-related chemical mechanism is enhanced via active photochemical reaction under high temperature during the heat-wave period. Here, the relationship between VOC emissions and their sensitivity to the generation of SOA was investigated in the BMA and UMA in Korea when the temperature was highly elevated.

Sensitivity Simulations of VOC Emission and SOA Generation
In the BMA and UMA, one of the major uncertainties is the lack of knowledge on the source distributions of disaggregated VOC emissions, partly because of the numerous species of VOCs and their diverse origins. Therefore, sensitivity simulations were designed and executed to investigate potential causes of model biases in VOC emissions. When independent evidence of model performance evaluation is not readily available, these sensitivity tests are good tools for diagnosing model performance on which inputs affect model results and what magnitude of input adjustment can reduce discrepancies between model results and observations.
Multiple WRF-Chem simulations were carried out for the period covering 5-21 July 2018 with 7 days as a spin-up period, with a base case (CASE 1) and three more sensitivity tests doubling (twofold increase), quintupling (fivefold increase), and decupling (tenfold increase) the VOC emissions (CASE 2-CASE 4), respectively. Analyzing the differences between the four cases facilitates a diagnosis of the optimal VOC emissions under atmospheric conditions relevant to the BMA and UMA areas. CASE 4 with decupling (tenfold increase) the VOC scaling can be regarded as an extreme case of VOC enhancement from heat wave effects. Using these four WRF-Chem simulations, the enhanced PM 2.5 concentrations were evaluated and the potential contribution of heat wave phenomena to PM 2.5 formations over the BMA and UMA in Korea was explored.
Over the selected study period, modeling was carried out with the various emissions to specifically diagnose the impact of the heat wave. The WRF-Chem simulation was run for 433 h, with the first 7 days (168 h) discarded as model spin-up for meteorology-chemistry interactions during this period. As a result, this study was based on the assumption that the emission of precursors increased, and the photochemical reaction conditions led to an increase in the secondary generation of O 3 and PM 2.5 concentrations.

Results and Discussion
An overview of the monthly mean PM 2.5 concentrations and the number of days with daily mean PM 2.5 concentrations higher than 35 µg/m 3 (unhealthy grade in the PM 2.5 forecasting system in Korea) are listed in Table 2, for June, July, and August from 2015 to 2018 in the two cities. The differences in concentration between years were not obvious for the four years in both cities, but the occurrence of days exceeding 35 µg/m 3 PM 2.5 was higher in both 2015 and 2018 for Busan and Ulsan. Hourly maximum PM 2.5 concentrations were highest in both 2015 and 2018. In 2015, 1-h maximum PM 2.5 concentrations were 74.8 µg/m 3 and 83.2 µg/m 3 for Busan and Ulsan, and 72.8 µg/m 3 and 84.2 µg/m 3 for Busan and Ulsan, respectively, in 2018. We investigated the seasonal characteristics of backward trajectories and synoptic weather chart (not shown here) and found out that frequent long-range transport (LRT) processes dominated, originating from China inland, and extremely high PM 2.5 were frequently observed in 2015. On the contrary, in 2018, our high episodic case during heat wave period occurred under consecutive stagnant synoptic conditions without LRT contributions, indicating short-term peaks in two areas were correlate well with the heat-wave alert period.
In order to find out the weather features during our high episodic cases, we investigated the synoptic situations at the 850 hPa level over the Korea Peninsula for the period from July 12 to July 20 (see Figure 3). The two cities were dominated by a long-lasting stagnant high-pressure system during the whole study period, as indicated in Figure 3. Therefore, these synoptic conditions with a weak prevailing wind speed made it possible for local circulations to develop within the two metropolitan areas, suggesting that the local air pollutants could be well characterized without transportation from other regions during the study period. Therefore, this study period provides good meteorological conditions to examine the accumulated air pollutants originating from local emissions and the generation process of secondary pollutants.
The time series of simulated and observed daily mean PM 2.5 concentrations and meteorological variables in the two metropolitan areas were given in Figure 4, during the study period from 12 to 22 July 2018. As described above, meteorological conditions during the study period showed low wind speeds (which favored accumulation of PM 2.5 ), high temperatures with a daily mean over 25 • C, and consecutive no-precipitation days with little or no cloud cover compared with other days in July, 2018. Overall, with the exception of small decreases on 16 and 18 July 2018, daily mean PM 2.5 concentrations increased steadily from 12 to 19 July 2018; this may be related to the influence of accumulated air pollutants. We selected the period from 00:00 LST 12 July 2018 (15:00 UTC 11 July) to 00:00 LST 21 July 2018 (15:00 UTC 20 July) to examine, in detail, the impact of the heat wave period under long-lasting stagnant atmospheric conditions over both the BMA and UMA.  Before analyzing the air quality simulation results, we evaluated the performance of the WRF meteorological model. This study focused on the impact of high temperature on secondarily generated air pollutants during the heat wave period; therefore, we compared simulated temperatures against actual temperature observations. Figure 5 shows the time series of hourly observed and simulated temperatures, and scatter plots of simulations versus observations in the cities. As shown in Figure 5, the observed daily average temperatures were 27.99 • C (in Busan) and 29.14 • C (in Ulsan) for 12-21 July 2018, and the simulated temperatures were 27.75 • C (in Busan) and 29.71 • C (in Ulsan). These were similar to the observations during the same period ( Figure 5). We also employed statistical parameters such as the correlation coefficient (R) and index of agreement (IOA), which are widely used to analyze model performance. High R values of 0.88 and 0.95 were found for Busan and Ulsan, respectively, indicating low discrepancies between observations and simulations ( Figure 5). Wind speed and relative humidity showed R values of 0.77 and 0.82 for Busan, and 0.65 and 0.76 in Ulsan.     Table 3 provides a statistical summary of model performance for the latter. Busan had the highest R (0.83) and IOA (0.86) in CASE 4, whereas Ulsan had the highest R (0.75) and IOA (0.78) in CASE 3. These results indicate that the appropriate increase in VOCs was different for each region, and imply that scaling approximately by a factor of 5 may give the best PM 2.5 simulation results. Elevated ozone was also simulated reasonably in CASE 3, showing that the maximum hourly O 3 concentrations is elevated up to approximately 100 ppb that is very close to the observation (max. 108 ppb) (not shown here). Thus, VOC emissions in the BMA and UMA may be underestimated by factors of more or less than five in the two cities. In terms of the root mean squared error (RMSE), normalized mean bias (NMB), and mean bias error (MBE), CASE 4 was found to be the most appropriate simulation of the lower NMB for Busan, within ± 30%, according to the criteria suggested by Emery et al. [59]. In the case of Ulsan, CASE 3 was found to be the best simulation, implying a VOC emission scaling factor of more than five for the simulation of PM 2.5 under heat-wave period emissions. We also analyzed the scatter diagrams of PM 2.5 concentrations from the four sensitivity tests of the WRF-Chem simulations for both areas in Korea. The results indicated that CASE 4 is shown to best fit the y = x line for Busan (y = 1.1986x − 11.6723), and CASE 3 fits best for Ulsan (y = 1.0446x − 10.4250), corresponding well with the observed results indicated in Figure 6 and Table 3.     Figure 7 represents the time series of the anomalies of mass concentrations of PM 2.5 for best fitted case during the study period, and the variations showed that the increase in VOC emissions resulted in day-time mass concentrations with peak time with the exception of 20 July 2018, but bias showed ranges less than 1 µg/m 3 with the remarkable agreement of IOA (0.94) and RMSE (0.179).  The chemical components during the heat wave episodes and their differences to the non-heat wave period are also important characteristics associated with SOA formation. We therefore contrasted the model results between four cases in two cities. Figure 8 illustrates the simulated chemical components for each of the four cases. As shown in Figure 8, SOA formation became active as VOC emissions increased, and their mass concentration was simulated to increase accordingly. It is also interesting that the increase in VOCs did not necessarily increase SOA generation in a linear manner, but increased it by only 30% even when VOC emissions increased by a factor of 10. Nevertheless, the proportions of SOAs during this period were 42.3%-66.6% (in Busan) and 34.3%-56.6% (in Ulsan) for the four cases. As no measurements of chemical components were available, we The chemical components during the heat wave episodes and their differences to the non-heat wave period are also important characteristics associated with SOA formation. We therefore contrasted the model results between four cases in two cities. Figure 8 illustrates the simulated chemical components for each of the four cases. As shown in Figure 8, SOA formation became active as VOC emissions increased, and their mass concentration was simulated to increase accordingly. It is also interesting that the increase in VOCs did not necessarily increase SOA generation in a linear manner, but increased it by only 30% even when VOC emissions increased by a factor of 10. Nevertheless, the proportions of SOAs during this period were 42.3-66.6% (in Busan) and 34.3-56.6% (in Ulsan) for the four cases. As no measurements of chemical components were available, we could not compare the model results directly with observations. Therefore, further monitoring and modeling studies are needed for the diagnosis over the two cities in Korea.
observation data in Busan and Ulsan metropolitan areas.
The chemical components during the heat wave episodes and their differences to the non-heat wave period are also important characteristics associated with SOA formation. We therefore contrasted the model results between four cases in two cities. Figure 8 illustrates the simulated chemical components for each of the four cases. As shown in Figure 8, SOA formation became active as VOC emissions increased, and their mass concentration was simulated to increase accordingly. It is also interesting that the increase in VOCs did not necessarily increase SOA generation in a linear manner, but increased it by only 30% even when VOC emissions increased by a factor of 10. Nevertheless, the proportions of SOAs during this period were 42.3%-66.6% (in Busan) and 34.3%-56.6% (in Ulsan) for the four cases. As no measurements of chemical components were available, we could not compare the model results directly with observations. Therefore, further monitoring and modeling studies are needed for the diagnosis over the two cities in Korea. It is widely known that higher ratios of PM 2.5 /PM 10 are induced by photochemical reactions originating from anthropogenic sources, whereas smaller PM 2.5 /PM 10 ratios result from the generation of large particles predominantly by natural emissions [60,61]. According to the Busan Air Quality Assessment Report [62], the annual PM ratio in Busan ranged from 0.5 to 0.6 throughout the whole year, which contrasts with the current WRF-Chem results during the study period. We investigated the PM 2.5 /PM 10 ratios simulated during the heat wave episodes. The time series of ratios of PM 2.5 to PM 10 for the four simulations in the two metropolitan areas are shown in Figure 9. In both areas, the PM 2.5 to PM 10 ratios were about 60% (min. scenario) to 92% (max. scenario). CASE 4, which showed the best model performances, averaged about 71% and 78%, with maxima of 91% and 92% in the two cities, respectively. On 19-20 July, the PM 2.5 to PM 10 ratio reached a maximum of 91-92% when the highest PM 2.5 concentration was simulated with the highest temperature in Ulsan (and second highest temperature in Busan) during the study period. It was therefore inferred that photochemical reactions were most active on 20 July, 2018, driven by the high temperature, and actively generated SOA during the heat wave period in 2018.
Considering these simulated PM 2.5 /PM 10 ratios, both the BMA and UMA tended to show considerably higher fractions during the heat wave period, compared with other season or previous years. Furthermore, the fraction increased overall as temperature increased, and the highest fraction occurred on 20 July, the day with the highest temperature and highest PM 2.5 concentrations during the study period. Thus, these results strongly suggest that the notably high fraction was induced during the heat wave period when extremely high temperatures were observed in these urban areas of Korea. the best model performances, averaged about 71% and 78%, with maxima of 91% and 92% in the two cities, respectively. On 19-20 July, the PM2.5 to PM10 ratio reached a maximum of 91%-92% when the highest PM2.5 concentration was simulated with the highest temperature in Ulsan (and second highest temperature in Busan) during the study period. It was therefore inferred that photochemical reactions were most active on 20 July, 2018, driven by the high temperature, and actively generated SOA during the heat wave period in 2018. Considering these simulated PM2.5/PM10 ratios, both the BMA and UMA tended to show considerably higher fractions during the heat wave period, compared with other season or previous years. Furthermore, the fraction increased overall as temperature increased, and the highest fraction occurred on 20 July, the day with the highest temperature and highest PM2.5 concentrations during the study period. Thus, these results strongly suggest that the notably high fraction was induced In order to investigate the possible relationship between temperature vs. VOC emission, we further carried out the same sensitivity tests for non-heat wave periods; pre-and post-heat wave period (refer to as pre-HWP and post-HWP hereafter) during the entire summer season in 2018. We defined the period of 12-21 July as HWP, 1 June-11 July as pre-HWP, and 22 July-31 August as post-HWP, respectively, and the same scaled VOC emission applied by a factor of five was considered. The simulated SOA concentrations were all examined and inter-compared for three different periods.
The calculated mean differences (MD = OBS (avg) − Model (avg)) of SOA over each of the three periods were 8.84 µg/m 3 , 9.40 µg/m 3 , and 9.29 µg/m 3 , for pre-HWP, HWP, and post-HWP, in Busan, and 3.97 µg/m 3 , 2.56 µg/m 3 , and 5.57 µg/m 3 , in Ulsan, respectively. The averaged concentrations for two non-heat wave periods: pre-and post-HWP, were 9.06 µg/m 3 (for Busan) and 4.79 µg/m 3 (for Ulsan), and the differences between heat wave and non-heat wave period were about 1.47 µg/m 3 (for Busan) and 2.24 µg/m 3 (for Ulsan), respectively. The differences in our results appeared due to the higher emissions influenced by higher temperature during the heat wave period. Our assumption is that the emission of precursors was increased during the record-breaking period, and photochemical reaction conditions were enhanced as well, which elevated the probability of high PM 2.5 concentrations. For example, the anthropogenic emission of VOCs is believed to be enhanced under the hot weather condition because of its variability: as the temperature increases, anthropogenic VOC's emissions will increase. It appears, in this study, that the emission of anthropogenic VOCs is enhanced by high temperatures via both anthropogenic and natural processes. However, in our comparative simulation results between three different periods, namely heat wave, pre-and post-heat wave periods, it is also difficult to quantify these differences with specific factors, partly because of the differences in weather conditions such as precipitation, atmospheric boundary layers, and atmospheric stabilities in each of the three periods. It is also doubtful how significant these differences of simulated SOA formations were between three periods. In this respect, it seems difficult to determine this conclusion with our result alone.
It seems unlikely that we are able to draw a robust relationship with our result alone between increase in temperature vs. precursor's emissions of PM 2.5 . One of the major difficulties is that we are unable to differentiate clearly between temperature effects and scaling effects from the underestimated VOCs emissions. Therefore, we are not suggesting that anthropogenic VOC emissions will clearly increase as the temperature increases. Instead, the changes in secondary PM 2.5 concentrations were given via sensitivity simulations of the actual heat wave period, particularly focusing on interpreting the contributions of disaggregated VOCs to the enhancement of SOA over the main cities in southeast Korea.

Summary
The BMA and UMA experienced an unprecedented heat wave during 11 July-15 August, 2018, shortly after the "Changma" rainy season. During this period, PM 2.5 concentrations were significantly elevated in the two metropolitan areas, with higher temperatures and weaker wind speeds than before and after the study period. The daily average PM 2.5 concentrations observed during the period in both the BMA and UMA mostly exceeded 135 µg/m 3 in Ulsan.
To analyze the effects of the heat wave on secondarily generated PM 2.5 concentrations during the study period, we carried out analysis of both measurements and simulations. As a modeling study, we employed WRF-Chem to simulate PM 2.5 concentrations and CREATE-2015 emissions over the BMA and UMA. As there were no robust VOC emission data for these extremely high temperature conditions, VOC emissions were determined and adjusted by carrying out several sensitivity simulations. Based on these "sensitivity run and adjustment" approaches, we evaluated the enhanced PM 2.5 concentrations and explored the potential contribution of heat wave phenomena to PM 2.5 secondary formations through photochemical reactions over the BMA and UMA in southern Korea.
The results showed that VOC emissions in the BMA and UMA were underestimated approximately by factors of five by CREATE-2015, and it was suggested that a considerable factor for scaling of VOC emission of CREATE-2015 yielded reasonable PM 2.5 simulations for the heat wave period in the BMA and UMA. The simulated PM 2.5 mass concentrations were compared with observations from the last four years, and it was concluded that the observed and simulated PM 2.5 increased mainly because of photochemical reactions in these southern urban cities during the heat wave period in Korea. The simulated chemical components of PM 2.5 , SO 4 2− , NH4 + , OC, EC, POA, and SOA, showed increases during the study period, and the SOA components were most dominantly (but not linearly) enhanced as VOC emissions increased during the heat wave period. The PM 2.5 to PM 10 ratios were about 74% (min. scenario) to 86% (max. scenario), with averages of 69% and 74% in Busan and Ulsan, respectively. The PM 2.5 to PM 10 ratio reached a maximum of 89%, when the highest PM 2.5 concentrations were simulated with the highest temperature in Ulsan (and second highest temperature in Busan) during the study period. This suggested that photochemical reactions mostly induced the generation of SOAs during the heat wave in 2018. Considering the general ratios of PM 2.5 /PM 10 reported by the Busan Air Quality Assessment Report [62], the annual mean ratios in Busan ranged from 0.5 to 0.6, contrasting considerably with the high PM 2.5 /PM 10 ratios seen during the study period. During the study period, the ratios of PM 2.5 /PM 10 increased as temperature increased, and the highest fraction occurred on the day with the highest temperature and the highest PM 2.5 concentrations. These results suggest that extremely high temperatures with no precipitation are of significant importance in the secondary generation of PM 2.5 masses in cities in southeastern Korea.
In addition, we further carried out the same sensitivity tests for non-heat wave periods, and investigated the possibility of anthropogenic VOC emissions influenced by higher temperature during the heat wave period. We found a small but detectable effect from our simulations. With these limited simulations carried out in our study alone, we are not suggesting the most-likely possibility of higher-than-normal anthropogenic VOC emission strengths partly originating from extremely high temperature, but it can be a potential relation between them, particularly around industrial areas such as Ulsan metropolitan area. As a future research, a highly complicated and well-designed intensive air quality monitoring campaign would be a prerequisite for this purpose. Together, more complicated sensitivities of changes in temperature to biogenic emissions should also be addressed over the areas in southeast Korea.