Evaluations of Surface PM 10 Concentration and Chemical Compositions in MERRA-2 Aerosol Reanalysis over Central and Eastern China

: The chemical composition dataset of Aerosol Reanalysis of NASA’s Modern-Era Retrospective Analysis for Research and Application, version 2 (MERRAero) has not been thoroughly evaluated with observation data in mainland China due to the lack of long-term chemical components data. Using the 5-year data of PM 10 mass concentrations and chemical compositions obtained from the rou-tine sampling measurements at the World Meteorological Organization the Global Atmosphere Watch Programme regional background stations, Jing Sha (JS) and Lin’An (LA), in central and eastern China, we comprehensively evaluate the surface PM 10 concentrations and chemical compositions such as sulfate (SO 42 − ), organic carbon (OC) and black carbon (BC) derived from MERRAero. Overall, the concentrations of PM 10 , SO 42 − , OC and BC from the MERRAero agreed well with the measurements, despite a slight and consistent overestimation of BC concentrations and a moderate and persistent underestimation of PM 10 concentrations throughout the study period. The MERRAero reanalysis of aerosol compositions performs better during the summertime than wintertime. By considering the nitrate particles in PM 10 reconstruction, MERRAero performance can be signiﬁcantly improved. The unreasonable seasonal variations of PM 10 chemical compositions at station LA by MERRAero could be causative factors for the larger MERRAero discrepancies during 2016–2017 than the period of 2011–2013.


Introduction
The Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2), is NASA's latest reanalysis of the atmospheric environment, with various environmental variables consistently in both temporal and spatial distributions [1]. The MERRA-2 combined the Goddard Earth Observing System, version 5, Earth system model   [2,3], three-dimensional variational data assimilation, and Grid point Statistical Interpolation analysis system [4,5]. In this, "MERRAero" specifically refers to a reanalysis of aerosols in MERRA-2, including bias-corrected aerosol optical depth observations from the Moderate Resolution Imaging Spectroradiometers onboard the Terra and Aqua satellites [6], as well as the Goddard Chemistry-Aerosol-Radiation and Transport model [7,8]. The MERRAero dataset includes the concentrations of the five aerosol species of dust (DS), sea salt (SS), sulfate (SO 4 2− ), black carbon (BC), and organic carbon (OC) over the world with a resolution of 0.5 • latitude by 0.625 • longitude and 72 vertical layers extending up to 0.01 hPa (~80 km) [9]. In the GEOS-5 model, DS and SS emissions are classified into different diameter bins based on the relationship between surface properties and near-surface wind speed [10]. The emission inventories of other aerosol species and their precursors are considered during the simulation, in which, sulfate and carbonaceous aerosol emissions are derived from both natural and anthropogenic sources [1,11]. NASA's Quick Fire Emission Dataset provides the biomass burning emissions after 2009, including primarily OC and BC. Sulfur dioxide (SO 2 , the precursor of SO 4 2− ) anthropogenic emissions come from the Emission Database for Global Atmospheric Research version 4.2 inventory in 2008 [12,13].
The differentiation of the aerosols' chemical compositions in MERRAero provides a valuable dataset for better studying various air quality issues around the world, especially for the regions lacking local particular matter (PM) observations due to unreliable or scarce monitoring [14]. Thus, many studies have emerged trying to estimate surface PM from MERRAero datasets in different regions of the world [12,[14][15][16][17]. Some evaluation studies on surface PM of MERRAero were performed in recent years. In the United States, MERRAero PM 2.5 was closer to observation values during the summer while larger discrepancies were observed during the winter, because of the lack of nitrate emissions in MERRAero and an underestimation of carbonaceous emissions in the Western US [16]. In Europe, the evaluation of PM 10 , PM 2.5 , SO 4 2− and BC concentrations were generally reasonable except OC and SS, in which the wintertime bias is mainly contributed by the anthropogenic sources of PM unresolved by the simulation [15]. The MERRAero PM 2.5 in Taiwan was significantly underestimated, by 42% on average of 2005-2014, because emissions of anthropogenic PM and their precursors were largely uncertain in China [14].
The validation focusing on MERRAero-derived PM 2.5 s reliance and uncertainty by independent measurements has been conducted spatially and temporally in China to support its applications related to climate-aerosol interactions [18][19][20][21]. Generally, MERRAero well captures the spatial distribution and seasonal variations of PM 2.5 mass concentration in China mainland. However, MERRAero produced lower daily mean PM 2.5 concentrations with outstanding bias for high ground PM 2.5 (>75 µg m −3 ) [19]. Significant underestimation of the PM 2.5 mass concentration in MERRAero was revealed across China mainland according to a 5-year evaluation, especially in Central China by 34.6 µg m −3 , followed by 19.8 µg m −3 in East China and 9.1 µg m −3 in South China [20]. Additionally, the bias also exhibited seasonal dependence. The largest biases were observed in winter. Due to the low intensity and weak variations of emission inventory as well as the absence of nitrate, the magnitude and variability of PM 2.5 in MERRAero are both underestimated [20]. Compared with the widely assessments of PM 2.5 product in MERRAero, the chemical speciation of MERRAero has not been thoroughly evaluated with observation data in mainland China due to the lack of long-term chemical components data. In this study, five years of aerosol sample data collected at the World Meteorological Organization's (WMO) Global Atmosphere Watch (GAW) programme's regional background stations, Jing Sha (JS) and Lin'An (LA), in central and eastern China, will be used to evaluate the aerosol concentrations and chemical speciation of MERRAero. Three observation-based methods for reconstructing nitrate will be proposed and evaluated to offset the key absence of MERRAero nitrate. The evaluation results for surface aerosol concentration and especially chemical compositions are necessary for improving MERRAero's applicability in highly polluted regions in China, especially for improving its applicability to the regional background levels of aerosol compositions.
The paper is organized as follows: the observation measurements of chemical components and the methods used for reconstruction and validation are introduced in Section 2; the seasonal and monthly variability of PM 10 and chemical compositions estimated by the MERRAero at two regional background stations are comprehensively validated in Section 3.1; three observation-based methods for making up the nitrate pollutants in the MERRAero are performed and evaluated in Section 3.2; the uncertainty analysis and discussion are presented in Section 3.3; and the assessments are summarized in Section 4.

Observation Sites and PM 10 Sample Analysis
Aerosol samples were collected from 2011 to 2013 and 2016 to 2017 at the WMO GAW atmospheric regional background stations, JS station in Hubei province and LA station in Zhejiang province, representing the regions of Twin Lake Basin and Yangtze River Delta over central and eastern China with high aerosol loading ( Figure 1). JS station (29 • 38 N, 114 • 12 E, 750 m a.s.l.) is located in the east wing of the Twin Lake Basin region, about 105 km south of Wuhan, the capital city of Hubei province. Situated on the peak of Monk mountain, JS station is about 30 km east of Chongming county, which is considered to represent the regional background atmospheric components over the Twin Lake Basin region. LA station (30 • 18 N, 119 • 44 E, 138.6 m a.s.l.) is located in the rural area of Zhejiang Province, which is close to rapidly developing regions in Jiangsu province and the megacity Shanghai. The regional background characteristics of the atmospheric components in Yangtze River Delta can be well represented by LA station [22].  Table 1. Aerosol samples were collected every two or three days by using a MiniVol™ air sampler (Airmetrics, Springfield, OR, USA), operating at a sampling flow rate of 5 L min −1 for 24 h from 09:00 a.m. to 09:00 a.m. (Beijing time) the next day. The 47 mm Whatman quartz microfibre filters (QMA) prefired at 850 • C for 3 h were used for the sampling.
The aerosol mass concentrations of PM 10 were obtained by gravimetric analysis in the Laboratory of the Chinese Academy of Meteorological Sciences with a microbalance (Sartorius, Germany). The water-soluble ions, including nine inorganic ions (F − , Cl − , NO 3 − , SO 4 2− , NH 4 + , K + , Na + , Ca 2+ , Mg 2+ ), were analyzed by using ion chromatography (Dionex 3000 IC and atomic absorption spectrophotometry HITACHI 180-70) at the Laboratory of the Chinese Academy of Meteorological Sciences. The elemental carbon (EC) and organic carbon (OC) were analyzed by using a DRI model 2001A EC/OC analyzer. The methods used were the same as described in Zhang et al. [23].

PM 10 Reconstruction and Evaluation Methods
The MERRAero provides the concentration of five PM species on the 3-hourly basis: SO 4 2− , OC, BC, DS and SS. Based on the available aerosol chemical speciation measurements, PM mass can be reconstructed with equations [24]. The total concentration of PM 10 (PM with diameter < or = 10 µm) is estimated as follows: We use brackets to denote concentrations. Assuming SO 4 2− is fully neutralized by ammonium (NH 4 + ) in the form of ammonium sulfate ((NH 4 ) 2 SO 4 ), [SO 4 2− ] is multiplied by 1.375 to account for sulfate. Other organic compounds found in particulate organic matter (POM) can be represented with [OC] multiplied by 1.8 [14][15][16]. DS 10 and SS 10 can be derived from the first three size bins in MERRAero's DS and SS.
The lack of the concentration of nitrate particles in MERRAero can be problematic in PM mass construction, as nitrate is growing into a bigger proportion in aerosols across China [16,25]. Thus, we use the modified Equation (1)   − ] to [SO 4 2− ] increased during the past few years (Figure 2), indicating the inevitable importance to induce MERRAero-nitrate due to its growing proportion in aerosols. To quantify MERRAero's accuracy, the performance statistics we used include the mean fraction F = C s /C o , (C s : simulated concentration; C o : observed concentration), the mean bias B = C s − C o , the standard deviation of the bias (SD-B) and the correlation coefficient (R). Despite being widely applied, the correlation coefficient R is a criticized index for evaluating model performance since it does not directly compare the simulated results with observed data [27]. Therefore, we adopt a rigorous index to evaluate the MERRAero performance. The proportion of simulated data which falls within a factor of 2 of the observed data (FAC2, i.e., the proportion of the data which satisfies 0.5 ≤ C s /C o ≤ 2.0), such a method can avoid the influence of extreme values and errors. It is considered the simulation performance to be reasonably good if FAC2 > 0.50 [28]. Besides, the index of agreement (IOA) was also used to assess the MERRAero performance in PM 10 against the measurements.
where N is the total number of the samples used for comparisons. The IOA is a standard measure of the degree of model accuracy and ranges from 0 to 1, with 1 showing perfect agreement of the prediction with the observation [29][30][31].

Evaluation of Aerosol Species
The performance statistics of MERRAero aerosol compositions at two regional background stations, LA and JS, are summarized in Tables 2 and 3   All species at LA and JS have reasonably low B values (Tables 2 and 3). However, there is quite a lot of scattering in the data, as reflected by the values SD-B, which are much larger than their respective average bias, accompanied with low R values. The data scattering is noticeable in Figure 3, especially for the PM species at the LA site. Even though, the bulk of data are still well simulated by MERRAero with good linear fitting results. Since all the FAC2 factors at LA and JS exceed 60% (  [15], and reasonable IOA indicating good data consistency with observations. It is worth noting that PM 10 evaluation results in Tables 2 and 3 were calculated with Equation (1) without consideration of nitrate particles.  4 2− ] seems to be the major contributor for [PM 10 ] inconsistency. The most obvious difference between LA and JS existed in the MERRAero performance of [SO 4 2− ] (Tables 2 and 3 and Figure 4). MERRAero overestimated [SO 4 2− ] with the lesser amplitude of monthly variation comparing with the observation in JS. The site JS is located on the remote mountaintop at an altitude of 750 m a.s.l. with the very low anthropogenic emissions of air pollutants and is often influenced by relatively clean air from above the boundary layer when the height of boundary layer is lower in cold season [32]. Air pollutants over the remote mountain are mostly transported from the high emission regions under a favorable synoptic system [33], which could lead to the significant monthly variations of air pollutants. Furthermore, the spatial resolution of 0.5 • latitude by 0.625 • longitude in the MERRAero data is too coarse to accurately present the mountainous topography, which could be responsible for the overestimation and gentle variations of [SO 4 2− ] in the site JS over the remote mountain in central China, while the relative reasonable [SO 4 2− ] was produced in LA over the plain of East China with the relatively high anthropogenic emissions of air pollutants. In both sites, MERRAero's [SO 4 2− ] in the second period, 2016 and 2017, shows larger discrepancy and overestimation. The use of a constant inventory of SO 2 emissions from 2008 [13] and anthropogenic carbonaceous in Aero Com Phase II dataset (HCA0v1) from 2006 [34] in MERRAero are problematic in the long-term simulation, despite that the use of 3DVar technique could partially improve the simulation accuracy [14]. The emissions are constantly changing with time, especially in China where the strict and comprehensive air pollution control policies were implemented by the government from 2013 [25]. Even though the bias of PM 10 species existed along the study period with the regional differences, the decreasing trend of PM 10 and chemical components since 2013 are reproduced by MERRAero, and there is still a potential to improve the performance of reconstructed PM 10 by considering the nitrate contribution to the simulation results.  (1) is lack of the concentration of nitrate particles, which could be the reason for the consistent underestimation in reconstructed PM 10 during the study period. In the following Section 3.2, we will testify and compensate the contribution of nitrate particles to reconstructed PM 10 .

The underestimation of [PM 10 ] is likely caused by a combination of [SO 4 2− ] and [OC] underestimations. For JS site, the lesser extent of monthly variation in [SO
It is noted that a large discrepancy exists between MERRAero and observations in November (N) and December (D) at JS site. The reason is mainly because that the JS station is located on the mountaintop (750 a.s.l.) and often influenced by clean air from the lower free troposphere when the boundary layer is low with wind coming from north west [32]; however, such a case might not be well represented by the MERRAero, and result in the overestimation of chemical compositions from MERRAero in November and December at JS (see Figure 5).

Improvement of PM 10 Reconstruction with Including Nitrate
With a growing proportion of nitrate in aerosols over China, the lack of the concentrations of nitrate particles in MERRAero dataset can make up a considerable portion of the total [PM 10 ] and [PM 2.5 ] loss [16,25]. 2− ] (i.e., from yearly to monthly, and then daily) to estimate MERRAero nitrate concentration. Figure 6 shows that the scattering in the reconstructed [PM 10 ] became less obvious from PM 10 (yearly ratio) to PM 10 (daily ratio) in both LA and JS, which is also confirmed by the declines in SD-B, increases in R, IOA and FAC2 (Table 4). It is worth noting that it is acceptable to include the nitrate concentration with our method for PM 10 construction by averaged observed [NO 3 − ]/[SO 4 2− ] in a period of time considering the difficulties of high temporal-spatial resolution in chemical speciation observation. The nitrate-induced efforts would significantly enhance the MER-RAero's applicability in highly polluted regions, especially with the relatively high nitrate concentrations in ambient air. The daily averages of PM 10

Bias Analysis and Discussion
The statistical analysis results in Table 3 revealed the obvious discrepancy of MER-RAero products in LA during 2016-2017. We examined the individual seasonal variations of MERREAero as well as observed PM chemical components during the 5-year study period (Figure 8). For the years of 2016 and 2017, the MERRAero analysis showed unreasonable seasonal variation at LA with higher concentrations in summer and lower concentrations in winter, which is not in agreement with the observed high concentrations during winter. The possible reasons could be the uncertainty of emissions, the error of data assimilation, and the simulation bias of meteorology, which lead to the significant underestimation of wintertime PM chemical components. Meteorology exerts impact significantly on PM formation and accumulation. In terms of the simulation bias of meteorology, the overestimation of near-surface winds and boundary layer height could cause the PM underestimation with stronger diffusion in the atmosphere [5]. More precipitation enhances the PM removals. Lower humidity and air temperature suppress PM formation. The reasons for the discrepancy in MERRAero analysis in LA during 2016-2017 still need more research. According to the averaged aerosol concentrations in observation and MERRAero during the first period 2011 to 2013, the contributions of different compositions apart from coarse particles to the PM 10 mass concentrations at two sites are identified in Table 5. The major identified aerosol types in PM 10 at LA and JS both were sulfate (as (NH 4 ) 2 SO 4 ), particulate organic matter (POM), nitrate (as NH 4 NO 3 ) and BC, with their contributions of 24%, 22%, 11% and 3% in LA, and 34%, 19%, 10% and 3% in JS, respectively. The MERRAero PM 10 based on daily reconstructed nitrate were similar from observations, with sulfate (33%), POM (26%), nitrate (13%) and BC (6%) in LA and sulfate (38%), POM (26%), nitrate (10%) and BC (5%) in JS. The reconstructed PM 10 with nitrate included is still lower than the observation by around 25 µg m −3 in LA and 14 µg m −3 in JS.  10 12.01 17% 14.03 20% SS 10 3.03 4% 1.14 2% The observation analysis of aerosol types at LA indicated sulfate and soil aerosol contributes a significant proportion [20]. The location of two observation stations is in the inland with less contribution of coarse sea salt aerosol. In the Section 3.1, we attribute the underestimation of sulfate to the constant inventory of SO 2 emissions used. Here we focus on discrepancy in the soil aerosol. According to Figure 9a-d, observed DS 10 rank as the primary components in PM 10 during 2016-2017 both in LA (31%) and JS (31%). The contribution of aerosol types in PM 10 is relatively well reproduced by MERRAero in LA. As for JS station, MERRAero underestimated the proportion of DS 10 by 10%. The model generally had difficulty simulating dust concentrations due to both limitations of the windblown dust emission module and uncertainties in the anthropogenic fugitive dust emissions inventories [35]. The discrepancy of MERRAero's [PM 10 ] could result from the underestimation of coarse particles originating from natural sources and anthropogenic emissions. Natural dust emissions in MERRAero depend on the strong winds exceeding the threshold values, and the threshold wind-speed for emission also depends on soil moisture and other surface properties over the desert regions [36]. The location of observation stations is relatively far away from the large desert area in northwestern China. Thus, in central and eastern China with intensive human activities, the emissions of anthropogenic fugitive dust could be an important factor for the PM 10 concentrations in the sites LA and JS. The discrepancies of DS may lay on excluding the anthropogenic fugitive dust emissions, which is a challenge for further modelling investigation.

Conclusions
We evaluated MERRAero reanalysis of PM 10 concentrations and chemical compositions with five-year observation data at two WMO GAW regional background stations, Jingsha and Lin'An, in central and eastern China. The chemical compositions of MERRAero have been thoroughly validated with observation data in mainland China. Three methods introducing nitrate to MERRAero PM 10 in yearly, monthly and daily timescales were compared to assess the PM 10 improvements. Such efforts could enhance the MERRAero's applicability related to climate-aerosol interactions in pollution regions, especially with the relatively high nitrate concentrations in ambient air. The main conclusions are summarized as follows: