Evaluating the sensitivity of fine particulate matter ( PM 2 . 5 ) simulations to chemical 1 mechanism in Delhi 2

14 Elevated levels of fine particulate matter (PM2.5) during winter-time have become one of the most important 15 environmental concerns over the Indo-Gangetic Plain (IGP) region of India, and particularly for Delhi. Accurate 16 reconstruction of PM2.5, its optical properties, and dominant chemical components over this region is essential to 17 evaluate the performance of the air quality models. In this study, we investigated the effect of three different 18 aerosol mechanisms coupled with gas-phase chemical schemes on simulated PM2.5 mass concentration in Delhi 19 using the Weather Research and Forecasting model with the Chemistry module (WRF-Chem). The model was 20 employed to cover the entire northern region of India at 10 km horizontal spacing. Results were compared with 21 comprehensive filed data set on dominant PM2.5 chemical compounds from the Winter Fog Experiment 22 (WiFEX) at Delhi, and surface PM2.5 observations in Delhi (17 sites), Punjab (3 sites), Haryana (4 sites), Uttar 23 Pradesh (7 sites) and Rajasthan (17 sites). The Model for Ozone and related Chemical Tracers (MOZART-4) 24 gas-phase chemical mechanism coupled with the Goddard Chemistry Aerosol Radiation and Transport 25 (GOCART) aerosol scheme (MOZART-GOCART) were selected in the first experiment as it is currently 26 employed in the operational air quality forecasting system of Ministry of Earth Sciences (MoES), Government 27 of India. Other two simulations were performed with the MOZART-4 gas-phase chemical mechanism coupled 28 with the Model for Simulating Aerosol Interactions and Chemistry (MOZART-MOSAIC), and Carbon Bond 5 29 (CB-05) gas-phase mechanism couple with the Modal Aerosol Dynamics Model for Europe/Secondary Organic 30 Aerosol Model (CB05-MADE/SORGAM) aerosol mechanisms. The evaluation demonstrated that chemical 31 mechanisms affect the evolution of gas-phase precursors and aerosol processes, which in turn affect the optical 32 depth and overall performance of the model for PM2.5. All the three coupled schemes, MOZART-GOCART, 33 MOZART-MOSAIC, and CB05-MADE/SORGAM, underestimate the observed concentrations of major 34 aerosol composition (NO3 , SO4 2, Cl , BC, OC, and NH4 + ) and precursor gases (HNO3, NH3, SO2, NO2, and O3) 35 over Delhi. Comparison with observations suggests that the simulations using MOZART-4 gas-phase chemical 36 mechanism with MOSAIC aerosol performed better in simulating aerosols over Delhi and its optical depth over 37 the IGP. The lowest NMB (-18.8%, MB = -27.4 μg/m 3 ) appeared for the simulations using MOZART-MOSAIC 38 https://doi.org/10.5194/acp-2020-673 Preprint. Discussion started: 27 August 2020 c © Author(s) 2020. CC BY 4.0 License.


42
The industrial activities in India have escalated to new heights in the past three decades which 43 consequently have led to multiple urban environmental issues, especially deteriorating air quality due to 44 suspended particulate matter of aerodynamic diameter smaller than 2.5 µm (PM 2.5 ) (Ghude et al., 2016;Ghude https://doi.org/10.5194/acp-2020-673 Preprint. Discussion started: 27 August 2020 c Author(s) 2020. CC BY 4.0 License.
distribution. In the WRF-Chem model, one can choose between four and eight aerosol size bins, which are with a diameter between 2.5 and 10 μm. Therefore, when four aerosol bins are used, three bins are assigned to 159 https://doi.org/10.5194/acp-2020-673 Preprint. Discussion started: 27 August 2020 c Author(s) 2020. CC BY 4.0 License. aerosols less than 2.5 μm in diameter. When eight aerosol bins are used, seven bins are assigned to aerosols with 160 diameters within this range. Usually, it is sufficient to use the four-bin simulation option to which the focus is 161 on air quality and it also reduces computational complexity (Georgiou et al., 2018).

CB05-MADE/SORGAM (CMS):
In the third experiment, we conducted simulations using the Carbon Bond 5 163 (CB-05) gas-phase mechanism (Yarwood et al., 2005,) which includes 51 chemical species and 156 reactions.  (Guenther et al., 2006) and dust emissions are based on the online observations values below 10 µg/m 3 and above 1500 µg/m 3 at a given site if other sites in the network do not

198
show values outside this range. The purpose of this step is to eliminate any short-term local influence that 199 cannot be captured in the models and to retain the regional-scale variability. Second, we removed single peaks Delhi as a part of the WiFEX field campaign at Delhi international airport (Ghude et al., 2017;Acharja et al., 207 2020). The quality assurance and control process applied to the measurement of the chemical ion is given at 208 Acharja et al., (2020). The meteorological observation data used in this study are taken from the Indian

210
The focus of the model evaluation was mainly to assess whether the model is able to effectively To quantitatively evaluate the model performance for basic meteorological parameters, the data for the 227 temperature at 2m (T 2m ), relative humidity at 2m (RH 2m ), and wind speed at 10m (WS 10m ) from six stations over

228
Delhi, India is used. Statistical metrics are derived by comparing the output of the three model simulations to

229
hourly measurements averaged over all ground stations. Table 1 shows the correlation coefficient (r), mean bias

230
(MB), and root mean squared error (RMSE) between observed and modeled temperature at 2m (T 2m ), relative 231 humidity at 2m (RH 2m ), and wind speed at 10m (WS 10m ) over Delhi, India. Modelled T 2m is in good agreement 232 with observations (NMB = 2 to 5 %) but shows higher RSME values (8.84 to 8.92 o C) for all three mechanisms.

233
The statistic for RH 2m indicates that the model has dry bias during winter for all the three mechanisms and speed and poor correlation could be due to the poor representation of surface drag exerted by the unresolved 236 topography, other smaller-scale terrain features, and building morphology (Mar et al., 2016;Zhang et al., 2013). under-predicts PM 2.5 within 70% at all stations, possibly due to lack of NO 3 and secondary organic aerosols

277
(SOA) in the GOCART model. We find that simulated mean NO 3 and SOA together contributed ~44 µg/m 3 278 with the MM mechanism, which is about 30% of total PM 2.5 mass concentration simulated during the winter 279 period.

280
We also examined the model performance of MG, MM and CMS chemical schemes over the Punjab,

281
Haryana, Uttar Pradesh, and Rajasthan ( Figure 3), which are the neighboring states of Delhi and often influences 282 the air quality in NCR region (Kumar et al., 2015;Kulkarni et al., 2020).

310
We further examined how the differences between coupled chemical mechanisms translate in 311 simulating Aerosol Optical Depth (at 550 nm) over the model domain. Figure 4 shows the spatial distribution of  Figure 5i shows the surface NO 3 concentration simulated by the MM and CMS mechanism. Since the MG mechanism does not simulate nitrate aerosols, NO 3 from the MG epxeriment is not shown here. Mean

343
NO 3 concentration was generally higher in the MM experiment than in the CMS experiment, particularly the 344 magnitude of NO 3 over central and eastern IGP region is larger. Similarly, as shown in Figure 5j, the magnitude 345 of mean NH 4 + concentration was also higher in the MM experiment over central and eastern IGP. On the other 346 hand, mean HNO 3 concentration was found highest in the MG experiment, followed by the CMS experiment, 347 and the lowest was found in MM (Figure 5g) experiment. The highest HNO 3 concentration observed in the MG the lack of aerosol thermodynamics in the MG mechanism means that HNO 3 stays in the gas-phase and does not 350 partition to particle-phase. The main precursor for NO 3 is HNO 3, and the equilibrium between nitrate and HNO 3,

351
and gas-phase NH 3 and HNO 3 can convert to aerosol NH 4 NO 3 . This indicates that the gas-particle partitioning 352 from HNO 3 to NH 4 NO 3 is more efficient in the MM experiment than in the CMS experiment. While, higher  Table 2 shows that all three chemcial mechanisms underestimate PM 2.5 concentrations in Delhi. The

375
lowest NMB appears for the MM mechanism (NMB = -18.8%, MB = -27.4 µg/m 3 ), whereas the NMB for the and gas-phase compounds that lead to secondary inorganic aerosols. Figure 6 presents the box-whiskers plot for 388 components of PM 2.5 from the observations and simulated by the model for the different coupled aeorsol 389 mechanisms at Delhi during the study period. It should be noted that nitrate is absent in GOCART; therefore, 390 ammonium and nitrate are not shown in Figure 6. The MG mechanism does not simulate NH 4 but multiplies