Model Evaluation and Uncertainty Analysis of PM2.5 Components over Pearl River Delta Region Using Monte Carlo Simulations

Sulfate, nitrate, ammonium, organic carbon (OC) and black carbon (BC) are the key components of PM2.5, but predicting their concentrations remains a challenge because of high uncertainties in the modeling. Employing the Nested Air Quality Prediction Modeling System (NAQPMS) developed by the Institute of Atmospheric Physics, Chinese Academy of Sciences, this study investigated the uncertainties in Monte Carlo simulations of these aerosols in the Pearl River Delta (PRD) region during 2015. 50 ensemble simulations with a 15 km horizontal resolution were derived by perturbing the emission data for sulfate, nitrate, ammonium, OC and BC from an emission inventory, which is one of the largest sources of uncertainty. Then, surface observations of these species collected from 10 sites across the region for 1 year were used to evaluate the performance of the ensemble simulations. The high correlation coefficients (> 0.74) and low mean biases (< 2 μg m–3) between the mean values of the ensemble and the observation data suggested that the model fairly accurately reproduced spatial and temporal variations in the nitrate, ammonium, OC and BC. However, the predicted sulfate concentrations, which displayed a correlation coefficient of 0.26, were far less reliable, particularly owing to the significant underestimation during winter. Further analysis revealed that uncertainties in the emission data explained most of the discrepancies for the OC and BC, but the mean biases for the sulfate and ammonium, especially during winter, probably stemmed from uncertainties in the heterogeneous reaction modeling.

results suggested that the ensemble mean could well reproduce the spatial and temporal variations 29 of nitrate, ammonium, OC and BC with the correlation coefficients above 0.74 and their mean 30 biases less than 2 −3 . However, the model has poor skills in the sulfate modeling with the 31 correlation coefficients 0.26 and remarkable underestimation in winter. Further analysis for such 32 modeling uncertainties suggested that the uncertainties in emissions can explain most of the 33 modeling uncertainties in OC and BC, but the mean biases in sulfate and ammonium modeling

A C C E P T E D M A N U S C R I P T
7 Heterogeneous reactions are parameterized as a pseudo-first-order irreversible rate constant, which 116 are calculated by the Eq. (1) proposed by Jacob (2000) Where i is the reactant for heterogeneous reactions, r is the mean radius of the particles, is the 121 gas phase diffusion coefficient of reactant i, is the mean molecular speed of the gas, is the 122 uptake coefficient of reactant i, and A is the aerosol surface area per unit volume of air. 123 For different temperature, humidity and particle surface characteristics, the uptake coefficient 124 may vary by several orders of magnitude. Therefore, for some specific particulate and gas phase 125 pollutants, the influence of RH and temperature on the uptake coefficient is considered. Further 126 detailed information on the uptake coefficient ( ) can be found in Li et al. (2012). 127 conform projection, and domain 2 includes 432×339 grid points with a 15km horizontal resolution. 130 8 We did a simulation period from 17 December 2014 to 31 December 2015. The first 15 days 134 were used as the "spin up" time of NAQPMS. Weather Research and Forecasting model (WRFV3.6) 135 was employed to provide the hourly meteorological inputs to NAQPMS. In the daily 136 meteorological simulation, WRF runs have been integrated over individual 36-hour period. Each 137 run included a meteorological "spin-up" time that took place in the first 12 hours of meteorological 138 input and the data of the remaining 24 hours were used for NAQPMS. The meteorological is critical 139 to the pollutant simulation since the meteorological parameter influences the transport process and 140 aerosols formation. The meteorological simulation was evaluated with the daily observations from 141 China Meteorological Administration. Fig. 2 shows a time series comparison of observed and 142 simulated temperature, relative humidity and wind speed at QingYuan site. This site also has been 143 marked on the Fig. 1 to show its position. In general, WRF can reproduce the temporal distribution 144 characteristics of major meteorological factors during simulation period, which can provide reliable 145 input date for NAQPMS. The initial and boundary conditions of WRF were provided by 1° x 1° 146 reanalysis data from the National Center for Atmospheric Research/National Center for the 147 Environmental Prediction (NCAR/NCEP). The parameterized settings of WRF are as follows: The 148 microphysics scheme used WSM3 simple ice scheme, the boundary layer scheme select YSU 149 scheme, the long-wave radiation chose RRTM scheme, the short-wave radiation select Dudhia (3) 202

A C C E P T E D M A N U S C R I P T
12 This study used the surface observations from 10 sites in PRD region for one year to evaluate 205 the simulation of sulfate, nitrate, ammonium, OC and BC. The observation data was from the 206 Secondary Composition Network in PRD region and provided by Environmental Monitoring 207 Center in Guangdong province. Samples were collected every 6 days, and each sampling period 208 was 24 hours. The specific information of monitoring stations is described in Table 2

A C C E P T E D M A N U S C R I P T
15 was -0.48 −3 and 0.56 −3 , respectively. The RMSE was 2.10 −3 and 0.82 −3 , 259 respectively. The MFB and MFE were all in the "excellent" range, and the simulations of OC and 260 BC at the ten stations were also close to the observational results. 261 262

Uncertainty analysis of sulfate, nitrate, ammonium, BC and OC simulations 263
Based on the above assessment, the model showed a good performance for OC and BC modeling, 264 while it displayed a relatively lower skill for SIA modeling, especially for sulfate. Emission as an 265 important input of the air quality model, its uncertainty is a key source of simulation errors. This 266 section will evaluate the impact of emission uncertainty on the modeling of the PM2.5 components 267 in PRD region, and to explore other possible uncertainty sources and factors. 268 Table 3 shows the uncertainties in the simulations of sulfate, nitrate, ammonium, organic carbon 270 and black carbon induced by emission uncertainties, which was calculated by CV. According to 271 Table 3, the modeling uncertainties of sulfate and ammonium in the annual simulation induced by 272 emissions were 5% and 16%, respectively, suggesting that their modeling uncertainties were less 273 sensitive to the emission uncertainties. However, as for nitrate, the emission uncertainty had a 274 greater impact on the simulation results. The modeling uncertainty of nitrate can reach 40%, which

OC and BC 326
Compared with SIA, the uncertainties in simulation of OC and BC induced by the emission 327 uncertainties were larger, which were 74% and 79%, respectively. As can be seen from the box-328 plot in Fig. 8, the uncertainties of OC and BC were the smallest in summer and larger in autumn

A C C E P T E D M A N U S C R I P T
20 simulation error of OC and BC. It is worth noting that the 2010 emission inventory was used to 349 simulate the 2015 concentration in our study. Table 4   Wang, Y., Zhang, Q., Jiang, J., Zhou, W., Wang, B., He, K., Duan, F., Zhang, Q., Philip, S. and Xie, 510

A C C E P T E D M A N U S C R I P T
34  The simulated results are from the simulation with 15km×15km horizontal resolution. 598