A numerical simulation of PM 2.5 concentration using the WRF-Chem model during a high air pollution episode in 2019 in Jakarta, Indonesia

Jakarta, as a megapolitan city, is always crowded with thousands of vehicles every day which results in decreased air quality due to combustion emissions and may have a significant impact on human health. Particulate matter (PM 2.5 ) is a pollutant that has an aerodynamic diameter of fewer than 2.5 micrometers and very easy to enter the human respiratory system so it can affect health. In the dry season, rain as the main natural mechanism for reducing PM 2.5 occurs very rarely, causing an accumulation of PM 2.5 concentrations in the atmosphere. The weather research and forecasting model coupled with chemistry (WRF-Chem) model is a dynamic model that works with atmospheric chemistry combined with meteorological variables simultaneously. This study aims to simulate the concentration of PM 2.5 in Jakarta during the high air pollution episode from 20 to 29 June 2019 with the WRF-Chem model based on the T1-MOZCART chemical scheme. Spatial analysis was conducted to determine the distribution of PM 2.5 concentrations during high air pollution episodes in Jakarta. Validation of the simulation model was based on three observation sites, one in South Jakarta and two in Central Jakarta. The results showed that the highest correlation is 0.3 and the lowest root mean square error (RMSE) is 26.4, while the simulations still tend to overestimate the PM 2.5 concentration.


INTRODUCTION
Jakarta is one of the most polluted metropolitan cities in the world with quite poor air quality and high particulate concentrations [1]. Poor air quality is one of the causes of premature death in the world and exposure to fine particles such as particulate matter (PM2.5), which is an important element of air pollution in cities, is associated with increased cardiovascular disease and premature death [2]. PM2.5 is a mixture of primary components which can consist of a mixture of heavy metals, organic carbon (OC), elemental carbon (EC), and secondary components such as sulfate, nitrate, ammonium, and secondary organic aerosol (SOA) [3].
According to Statistics Indonesia, in addition to being the center of the economy, Jakarta is also a trade center that has a large population and increasing purchasing power, which causes the use of vehicles to  [4]. This large number of vehicles contributes to the high concentration of PM2.5 in Jakarta.
According to Lestari et al. [5], land transportation and the industrial sectors account for 46% and 43% of PM2.5 emissions in Jakarta with emissions from heavy-duty vehicles still the highest contributor. Jakarta's air quality is typically worse during the dry season than during the rainy season [6]- [9]. The study by Kusumaningtyas et al. [7] showed that the maximum concentration of particulate matter occurs from June to September and then decreases from December to February. Based on previous studies, the concentration of particulate matter can also be influenced by meteorological variables such as rainfall, air temperature, and wind speed [10]- [14]. Rainfall can reduce atmospheric particulate pollution, including PM2.5 [15]. On days where there has been no rain for a long time, the air that does not fluctuate too much, sunny weather, the presence of an inversion layer of temperature, or wind speeds that are close to calm allow pollutants to remain in the atmosphere of an area and increase in concentration.
The weather research and forecasting model coupled with chemistry (WRF-Chem) is a model that coupled the meteorological model and atmospheric chemistry models [16], [17]. WRF is often used to simulate or forecast meteorological events that influence the variability of the concentration of pollutants in the atmosphere [18]. Meanwhile, the WRF-Chem has been used to simulate atmospheric chemistry based on the atmospheric model so that it can be taken into consideration how the meteorological process influenced the composition of atmospheric chemistry and pollutants [19], [20]. Based on previous research, WRF-Chem has been widely used to estimate the concentration of PM in subtropical regions [21]- [24], but research in tropical regions like Indonesia is still limited and usually related to wildfire cases [25], [26]. Research on air pollution in Indonesia using WRF-Chem in Indonesia is still limited due to the complexity of the precise parameterization for specific areas in Indonesia which have complex atmospheric conditions and there is not enough reference for this, however, this research must be continuously developed. Based on a study by Liu et al. [27], the WRF-Chem model may simulate the PM2.5 concentration with an overestimated output, but the model error is not significant.
The WRF-Chem uses several parameterization schemes that are selected based on the conditions of an area to be modeled or analyzed for simulating air pollutants. The choice of parameterization scheme will affect the model output [28]. In this study, the parameterization that will be used refers to Liu et al. [27], to simulate the PM2.5 during the 2019 high air pollution episode in Jakarta from 20 to 29 June.

RESEARCH METHOD
The research location chosen was the Special Capital District of Jakarta has coordinated 5°19'12"S to 6°23'54"S and 106°22'42"E to 106°58'18"E with an area of 740.3 km 2 . Jakarta is often suffered from a low air quality problem with vehicle emissions being a major factor in declining air quality in Jakarta. We used PM2.5 concentration datasets from 2 monitoring stations in Central Jakarta and 1 station in South Jakarta owned by BMKG and US-AirNow which locations as shown in Figure 1. parts of WRF-Chem treat transport processes (progressive, convective, and diffusion), wet and dry deposition, chemical transformation, emissions, photolysis, aerosol chemistry, and dynamics (including inorganic and organic aerosols) [29]. We used the global forecast system (GFS) dataset as input for meteorological parameters. WRF-Chem is a model of air pollution that combines meteorological factors and atmospheric chemistry together (online coupled). Each region has a characteristic and unique atmosphere that cannot be compared to other regions, the parameterization scheme in WRF-Chem is expected to be able to mathematically simulate the uniqueness of the region by choosing the right parameterization. With a process that is too small or physically complex, parameterization is used to obtain a more accurate prediction result which is represented in a simpler model [30]. This study used 3 domains in the WRF-Chem process the 3 rd domain covers Special Capital Region of Jakarta (DKI Jakarta) as shown in Figure 2. Parameterization is used as a representation of small-scale weather processes affecting larger scales. The parameterization of weather modeling consists of microphysics, cumulus or convection, surface land model, planetary boundary layer (PBL), atmospheric radiation, and physical interactions. In a study by Chen et al. [31], a combination of Yonsei University (YSU) PBL, Goddard SW, and geophysical fluid dynamics laboratory (GFDL) LW schemes showed the greatest consistency between simulated and observed PM2.5 values. Although the PBL scheme has a dominant impact on the simulation of meteorological variables, the selection of the LW and SW schemes is equally important. In other research, Lin Microphysics, Grell 3D Cumulus, Mellor-Yamada-Janjic (MYJ) PBL, rapid radiative transfer model (RRTMG LWR), and RRTMKG SWR were used in the parameterization that simulated PM concentration in Jakarta [28]. In this study, we used the WRF-Chem configuration as in Table 1. T1-MOZCART presents an update to the MOZART-4 chemical gas phase mechanism in the chemical option (chem_opt) option in the WRF-Chem scheme. T1-MOZCART has 142 gas-phase species compared to 81 gas-phase species in MOZCART. In addition, there is an increased understanding of the volatile organic compound (VOC) oxidation process through laboratory measurements, as well as the need to better represent secondary organic aerosol precursors. Recent field measurements of increasing amounts of isoprene oxidation products, as well as individual aromatic hydrocarbons and terpenes, allow a more precise evaluation of the model [32]. We used the Pearson correlation coefficient and root means square error (RMSE) in modeling validation. Figure 3 illustrates a comparison of surface temperature observations from the meteorological station of Kemayoran, Central Jakarta (BMKG headquarter), and the simulation of WRF-Chem using the parameterization in Table 1 during high air pollution episode in Jakarta. The diurnal variations in surface temperature can be simulated well as shown by the correlation coefficient of 0.96 and RMSE of 2.6 which is not much different from the standard deviation of the observational data which is 2.2. Simulation of the surface temperature performs better results in simulating temperature from morning to noon and is less accurate in the late afternoon to night time. The simulation of surface wind speed at the BMKG headquarter, Central Jakarta also shows better results in the morning to noon and performs poor results in the afternoon to early morning as shown in Figure 4. In general, surface wind speeds during periods of high air pollution episodes in Jakarta can be simulated well by WRF-Chem which is indicated by a correlation coefficient of 0.76 and an RMSE of 1.8 which is still smaller than the standard deviation of surface wind observations of 2.8.   Figures 5 and 6 shows that the model simulates a higher PM2.5 concentration in the eastern and northern parts of Jakarta.

RESULTS AND DISCUSSION
Based on Figure 7, is a graph of hourly PM2.5 concentrations averaged in all observation sites and an average of the hourly concentration from the simulation with T1-MOZCART. The observation shows an increase of PM2.5 during the late afternoon before evening, this high concentration state will last until 8 am LT. After 8 am LT is the time the PM2.5 concentration starts to decrease until 5 pm. The higher PM2.5 concentration observed at nighttime to early morning compared to daytime is due to changes in the boundary layer height at nighttime due to the cooling of the near-surface atmosphere so that PM2.5 will be concentrated near the surface [31]. The hourly PM2.5 simulation follows the observation well on average from 1 am to 4 pm LT, although it is higher than the observed value.  Figure 8(b). The simulation can simulate up to 120 µg/m 3 with a minimum value that is still above the observation. The peak period of the highest PM2.5 concentration according to observations occurred on 25 June while the model shows on 27 June, but the decline after 27 June can be well simulated. Figure 8(c) shows a graph of the simulated and observed PM2.5 concentrations with a three-hour resolution at the BMKG headquarter in Central Jakarta whose observation data also has a three-hour resolution. Similar to the observations at the other two sites, observed PM2.5 concentrations at the BMKG headquarter in Central Jakarta have also experienced a decline in trend since June 27, which the decline can be simulated by the model although with lower variability and tends to be closer to the average. Furthermore, the highest concentration at the observation site at the BMKG headquarter occurred on June 25, while in the model it occurred on June 27. Some data are blank and data that are too low at the observation site at the BMKG headquarter occurs due to daily periodic maintenance from midnight to morning on the equipment used. We evaluated the PM2.5 simulation based on observation datasets from three observation sites in Jakarta using three parameters, i.e. correlation coefficient (r), RMSE, and Bias as in Table 2.  Although the correlation coefficients were low, the highest PM2.5 concentration correlation with the observational data is shown in the simulation at the AirNow observation site in Central Jakarta, while the lowest is shown in the simulation at the BMKG headquarters in Central Jakarta. RMSE at the two AirNow observation sites is not much different and better than at BMKG headquarters. Meanwhile, the smallest bias is shown by the PM2.5 simulation at the AirNow observation site in South Jakarta, while the simulations at two other observation sites are quite overestimated the observation. This overestimate PM2.5 simulations might come from the WRF-Chem parameterization schemes used in this research, that also based on a study by Liu et al [27].

CONCLUSION
We analyzed the spatial distribution of the output of the PM2.5 concentration simulation using T1-MOZCART scheme in Jakarta during a high air pollution episode from 20 to 29 June 2019. The WRF simulation performs better in simulating surface wind speeds but tends to underestimate the surface temperature. Meanwhile, a simulation of PM2.5 concentration shows that during the peak of the high pollution episode, the average PM2.5 concentrations are more than 100 µg/m 3 at 08.00 am. The lowest PM2.5 concentration at 08.00 pm is in the southern and western parts of Jakarta. A look at how the simulation changes over time showed that it tends to get higher at night and get lower afternoon. We validated the simulation of PM2.5 concentration using T1-MOZCART scheme based on observation data and found that the simulation shows better performance in correlation, RMSE, and bias at two AirNow observation sites than at BMKG headquarters. Overall, the simulation shows an overestimate at all three observation sites.