Emission inventory of air pollutants and chemical speciation for specific anthropogenic sources based on local measurements in the Yangtze River Delta region, China

A high-resolution air pollutant emission inventory for the Yangtze River Delta (YRD) region was updated for 2017 using emission factors and chemical speciation based mainly on local measurements in this study. The inventory included 424 non-methane volatile organic compounds (NMVOCs) and 43 fine particulate matter (PM2.5) species from 259 specific sources. The total emissions of SO2, NOx , CO, NMVOCs, PM10, PM2.5, and NH3 in the YRD region in 2017 were 1552, 3235, 38 507, 4875, 3770, 1597, and 2467 Gg, respectively. SO2 and CO emissions were mainly from boilers, accounting for 49 % and 73 % of the total. Mobile sources dominated NOx emissions, contributing 57 % of the total. NMVOC emissions, mainly from industrial sources, made up 61 % of the total. Dust sources accounted for 55 % and 28 % of PM10 and PM2.5 emissions, respectively. Agricultural sources accounted for 91 % of NH3 emissions. Major PM2.5 species were OC, Ca, Si, PSO4, and EC, accounting for 9.0 %, 7.0 %, 6.4 %, 4.6 %, and 4.3 % of total PM2.5 emissions, respectively. The main species of NMVOCs were aromatic hydrocarbons, making up 25.3 % of the total. Oxygenated volatile organic compounds (OVOCs) contributed 21.9 % of the total NMVOC emissions. Toluene had the highest comprehensive contribution to ozone (O3) and secondary organic aerosol (SOA) formation potentials, while other NMVOCs included 1,2,4trimethylbenzene, m,p-xylene, propylene, ethene, o-xylene, and ethylbenzene. Industrial process and solvent-use sources were the main sources of O3 and SOA formation potential, followed by motor vehicles. Among industrial sources, chemical manufacturing, rubber and plastic manufacturing, appliance manufacturing, and textiles made significant contributions. This emission inventory should provide scientific guidance for future control of air pollutants in the YRD region of China.

ammonia (NH3), and particulate matter (PM) defined by diameters that are generally 45 9 waste treatment and disposal, and agricultural) were all calculated using the EF method. 215 The emission estimation method of this study has been improved on the basis of 216 our previous study (the latest version was for 2014) Ni et al., 2020). 217 Table 1 shows the differences between the methods and data sources of this study and 218 the previous. First, the source category has been refined from the third-level 135 219 categories to the fourth-level 2812 categories. Among them, large point sources such as 220 iron & steel and petroleum refining sectors were further subdivided into different 221 emission segments. Secondly, in addition to the environmental statistics data, the 222 activity data has been refined through local investigations on the removal technologies 223 and efficiencies, operating hours, and working conditions of industrial and mobile 224 sources including motor vehicles and non-road machinery; emissions from ships and 225 aircrafts, which were not considered in our previous study, were estimated based on 226 dynamic activity data like AIS provided by local department. In terms of the EFs, most 227 of them were corrected based on local measurements. 228   The method used to determine the PM2.5 and NMVOC source profiles followed 295 that used for the EFs and data was preferentially selected as follows: local 296 measurements>measurements from previous domestic studies>USEPA  Table S3. For those species which could not be determined by 312 analytical methods, the mass fraction data was supplemented with that obtained from 313 the SPECIATE database. Figure 2 shows the PM2.5 and NMVOC speciation profiles for 314  cci.org/). Emissions from port and factory machinery, and airport ground handling 330 equipment were assigned according to their geographical coordinates. Emissions from 331 residential sources were assigned from population distribution data with a 1 km 332 resolution. Emissions from agricultural sources were allocated from farmland areas in 333 the land use data (ibid.). 334

Uncertainty analysis 335
Uncertainty was mainly derived from the activity data and EFs. The coefficients 336 of variation of the activity data and EFs for each source were classified into seven 337 grades in the range of 2%-100% using expert judgment. The coefficient of variation for 338 the activity data was determined according to the data source. Environmental statistical 339 data with specific source information was assigned the lowest coefficient of uncertainty 340 (2%), while activity data estimated from the statistical yearbooks, such as biomass 341 burning, was assigned the highest uncertainty value (98%). The coefficients of 342 uncertainty for other activity data sources were assigned to be 18%, 34%, 50%, 66%, 343 and 82% in turn. The principle for assignments of the coefficients of variation for EFs 344 was the same as the activity level. EFs derived from local measurements in the YRD 345 region with large samples were assigned the lower coefficients of uncertainty (18%), 346 while those from USEPA or EMEP/EEA datasets were assigned higher coefficients 347 (98%). Then the uncertainty of each pollutant from each emission source can be 348 combined by Eq. (3-5). A detailed description of the analytical methods used can be 349 found in our previous study (Huang et al., 2011). 350 where, CV is the coefficient of variation of the emission rate, E is the emission rate, U 351 is the uncertainty of the emission source, Ca is the uncertainty of activity data, Cf is the 352 uncertainty of EF, j and k represent for pollutant and emission source, respectively. 353

Model configurations 354
To verify the reliability of the EI, we used CMAQ (version 5.3) to simulate the 355 concentrations of SO2, NO2, PM2.5, PM10, O3, and CO in the YRD region for January 356 and July 2017, and compared these with the observation data for each city in the region. 357 The meteorological field for the CMAQ model was obtained from the WRF (version 358 3). The EI developed in this study was then used to produce an emission system for the 359 YRD region while emissions beyond the YRD were obtained from the MEIC 2016. The 360 anthropogenic data was then combined with biogenic data obtained from the Model for 361 Emissions of Gases and Aerosol from Nature modelling system (version 2.10) as the 362 final input for the EI of the model. Figure S1 and Table S6 show the domain and settings 363 for the model system. Detailed information is provided in Section 6 of the Supporting 364 information. 365 2.10 Estimation of O3 and SOA formation potentials 366 To characterize the regional O3 and SOA formation contributions of different 367 NMVOC species and their sources, we used the O3 formation potential (OFP) and SOA 368 formation potential (SOAP) methods of estimation. OFP and SOAP were obtained from 369 the sum of the individual NMVOC species emissions multiplied by the maximum 370 incremental reactivity (MIR) and SOA yield, respectively. MIR and SOA yield for where, OFPi and SOAPi are the ozone formation potential and SOA formation potential 374 of source i, respectively, Ei,j is the VOC emission of species i, MIRj is the maximum 375 increment reactivity for the jth chemical species, Yj is the SOA yield for the jth chemical 376 species. 377  Table S1 of the 385 supporting information. 386 Table 2  study were derived from local measurements which were generally lower than those in 396 previous studies, so the NOx emissions from the power sector were 47% lower than 397 those from the MEIC. CO emissions were higher than the MEIC results but similar to 398 those reported by Sun et al. (2018a). NMVOC emissions for key sources in this study 399 were individually estimated using the "bottom-up" method, so the estimates were lower 400 than those using the "top-down" approach. In addition, most of the EFs selected in this 401 study were detailed into different process segments, which were generally lower than 402 the comprehensive EFs used for whole industrial sectors in previous studies. Since dust 403 sources were not included in the MEIC inventory, PM10 and PM2.5 emissions estimated 404 in this study were 1.7 and 0.5 times higher, respectively. A previous study also showed 405 that NH3 emissions in China were underestimated, mainly due the application of lower 406 emission rates from fertilizer applications and livestock and the omission of other 407 sources (Zhang et al., 2017). Therefore, we used local measured NH3 EFs for fertilizer 408 application and some of the livestock breeding sources in the YRD region. 20%, and 24%, respectively, which were consistent with the trends of regional air 417 quality improvement (SO2 44 %; NO2 5%; PM10 22%; PM2.5 27%). However, it should 418 be noted that the approach of emission estimation in this study has made a number of 419 localized corrections in terms of emission factors and activity data. For example, CO, 420 NMVOC, and NH3 emissions have increased significantly compared to 2014, which 421 mainly because more point sources were included in this study and more localized EFs, 422 which were generally higher than those in previous studies, were applied to estimate 423 NOx, CO, NMVOC, and NH3 emissions from solvent-use, motor vehicle, non-road 424 machinery, and agricultural sources. Next, it is necessary to estimate the emission 425 inventories by the same approach for different years to evaluate the changes in air 426 pollutant emissions in recent years. 427  and 12% of mobile source SO2, NOx, PM10, and PM2.5 emissions, respectively while 473 heavy-duty trucks contributed 31%, 37%, and 36% to mobile source NOx, PM10, and 474 PM2.5 emissions respectively. Light-duty vehicles contributed significantly to CO, 475

Results and discussion
NMVOCs, and NH3 emissions, accounting for 61%, 46%, and 90%, respectively. Non-476 road machinery accounted for 27%, 18%, 12%, 21%, and 22% of NOx, CO, NMVOCs, 477     emissions is consistent with that obtained from the on-site surveys in Jiangsu Province 519 (Zhao et al., 2017); the distribution of NH3 emissions is also consistent with the results 520 using dynamic emission factors and localized information (Zhao et al., 2020). 521 Compared with the national-scale inventory like the MEIC, this study has improved the 522 distribution along the Yangtze River and Hangzhou Bay where large point sources were 523 denser, and it also reduced the misjudgment of NOx and NMVOC emission hotspots in 524 the northern and southern areas, as shown in Figure S1. The distribution of NH3 525 emissions was also improved in the northern areas of the region and in the city centers 526 with more localized EFs of mobile and agriculture sources. 527

Uncertainty assessment 528
The EI was compiled using the bottom-up approach based on local EFs and 529 activity data from the region. The activity data for industrial sources, including fuel 530 consumption, sulfur content, ash content, raw material used, and control efficiency, 531 were collected from the Environmental Statistics Database. EFs from some key sources, 532 such as coal-fired power plants and boilers, iron and steel manufacturing, gasoline and 533 diesel vehicles, non-road machinery, catering, and agricultural sources, were modified 534 based on the local measurements. These measured helped to reduce the uncertainty of 535 the emission estimates. Table 3 shows the uncertainties of major sources at the 95% 536 confidence interval in this EI. The average uncertainties of emissions from the YRD 537 region were estimated as -29 to 36% for SO2, -28 to 33% for NOx, -42 to 75% for CO, 538 -44 to 68% for NMVOCs, -36 to 62% for PM10, -30 to 46% for PM2.5, and -58 to 117% 539 for NH3. The overall uncertainties were lower compared with our previous EI for the 540 YRD region (Huang et al., 2011). 541 The uncertainty assessment indicated that emissions from stationary combustion 542 sources, such as power plants and boilers, were more reliable, because the estimates 543 were based on detailed activity data and local measurements. The uncertainties for 544 emissions from major industrial sectors, such as ferrous metal manufacturing, non-significantly improved by the detailed emission estimation approach for the different 547 process segments. However, large uncertainties remained for emissions from chemical 548 manufacturing due to the many uncategorized processes and emissions for that sector. 549 The uncertainties for emissions from vehicles and non-road machinery in this study 550 were mainly from the activity data. Although their population could be obtained from 551 the statistical yearbooks, VMT and working hours could not be estimated accurately. 552 Dust emissions from construction and roads dust had much higher uncertainties as their 553 activity data lacked detail and fewer EFs were available. Most of the area sources, such 554 as residential and agricultural, were estimated from activity data obtained from the 555 statistical yearbooks, resulting in higher uncertainties in their emission estimates. 556 Despite these limitations, the emission estimation approach, based on refined process 557 segments and local measurements, reduced the overall uncertainties of the EI. However, 558 more comprehensive activity data and accurate EFs are still required to improve the 559 quality of EIs in the future. 560 3.2 PM2.5 and NMVOC species emissions 562 3.2.1 PM2.5 species 563 Figure 6 shows the major source contributions and species comprising PM2.5 in 564 the EI. OC, Ca, Si, PSO4, and EC were top five components of primary PM2.5 in the 565 YRD region, accounting for 9.0%, 7.0%, 6.4%, 4.6%, and 4.3% of PM2.5 emissions 566 respectively. There were large differences in the emission contributions for the different 567 PM2.5 species. Among the industrial sources, the non-metallic mineral manufacturing 568 sector made the largest contributions to Ca, Si, and Al emissions, accounting for 51.6%, 569 15.9%, and 18.8% of these species respectively. Ferrous metal manufacturing was the 570 main source of Fe emissions, accounting for 57.9%. Vehicles were major contributors 571 to OC and EC emissions at 18.0% and 43.5% respectively. K and Cl emissions mainly 572 came from biomass burning, accounting for 50.4% and 78.5% respectively. 573 Construction dust was also an important source of PM2.5 species, accounting for 15. Overall, the refinement of the source profiles helped to assess the impact of 606  Table S7 of the  625 supporting information. Figure 9 shows the mean fractional error (MFE) and the mean 626 fractional bias (MFB) between the simulated and observed daily average concentrations 627 in the cities of the region. Overall, the MFB and MFE of simulation and observation 628 results of all the pollutants in January and July were all within the criteria (MFB ≤ ±60%, 629 MFE ≤ 75%) of model performance recommended by Boylan and Russell (2006) most of them were with the performance goals (MFB ≤ ±30%, MFE ≤ 50%), which 631 indicated that the EI in this study could reflect the air pollution in winter in the YRD 632 region. In July, the MFB and MFE of O3 and PM2.5 model performance all fell within 633 the criteria range. However, the simulation results of primary pollutants like SO2, NO2, 634 PM10 and CO were somewhat underestimated. Especially for SO2 and CO, nearly half 635 of the cities had MFBs lower than -60%, and the cities with large deviations were 636 mainly concentrated in peripheral areas of the YRD region (such as Huangshan, 637 Chizhou, Xuancheng, Lishui, etc.). These cities generally had higher contributions of 638 area emissions from residential and agriculture sources instead of large point industrial 639 sources. The activity data of these sources usually had higher uncertainties and would 640 easily cause the deviation of emission estimation. For example, the underestimation of 641 the amount of residential coal combustion would undoubtedly lead to a severely low 642 estimate of SO2 and CO emissions. However, since PM2.5 and O3 pollution were more 643 regional, their simulation results were less affected by insufficient local activity data in 644 these cities. Conducting more detailed on-site investigations to obtain more accurate 645 activity data is the key to further improving the performance of EI in the future. 646  exceeded that of the industrial process sources. The contribution of motor vehicles to 667 regional OFP and SOAP were 13.9% and 13.5% respectively, which was close to those 668 from residential solvent-use sources. These two sources were the major contributors to 669 O3 and SOA formation in urban areas. 670 Four major industrial sectors could account for most of the OFP and SOAP in the 671 YRD region: The chemical manufacturing sector contributed 16.4% and 14.8% to OFP 672 and SOAP respectively; the rubber & plastic manufacturing sector had a SOAP 673 contribution rate of 11.8% (its OFP was relatively low at 1.2%); the appliance 674 manufacturing and textile sectors, accounted for 10.5% and 10.4% of OFP and SOAP 675 contributions respectively. 676 Based on the above, it was concluded that the reduction of aromatic hydrocarbon 677 emissions from industrial and vehicular sources were of greatest importance for the 678 YRD region. In particular, the high reactivity species, such as toluene, xylene, and 679 trimethylbenzene should be given priority in NMVOCs pollution control measures for 680 the region.

Conclusions 685
A high-resolution air pollutant EI for the YRD region was updated using EFs 686 derived mainly from local measurements. In addition to the conventional pollutants, 687 424 NMVOCs and 43 PM2.5 components were also included in the inventory. The EI 688 was refined into four main categories comprising 259 specific sources. The results 689 indicated that the total emissions of SO2, NOx, CO, NMVOCs, PM10, PM2.5, and NH3 690 in the YRD region in 2017 were 1,552, 3,235, 38,507, 4,875, 3,770, 1,597, and 2,467 691 Gg, respectively. Overall, SO2 and NOx emissions estimated in this study were lower 692 than the previous EIs such as the MEIC. Substantial reductions in emissions from power 693 plants and boilers in recent years were a significant factor. The NMVOC emissions 694 were also slightly lower than the results of previous studies. This was mainly due the 695 use of EFs refined to the specific sectors in this study, which were generally lower than 696 the comprehensive EFs used elsewhere. PM10 and PM2.5 emissions were respectively 697 1.7 times and 0.5 times higher than the MEIC, due to the inclusion of dust sources. The 698 NH3 emissions from this study, estimated using localized EFs, were significantly higher 699 than those of previous studies. 700 SO2 and CO emissions were mainly from boilers in the region, accounting for 49% 701 and 73% of the total. Mobile sources dominated NOx emissions from anthropogenic 702 sources, accounting for 57% of the total. NMVOC emissions were mostly from 703 industrial sources, accounting for 61% of the total. The main contributing industrial 704 sectors were chemical manufacturing and solvent-use sources like furniture 705 manufacturing, appliance manufacturing, textiles, packaging and printing, and 706 machinery manufacturing. Dust sources were responsible for 55% and 28% of PM10 707 and PM2.5, respectively. Agricultural sources accounted for 91% of NH3 emissions. 708 The main PM2.5 species emitted from anthropogenic sources in the YRD region 709 were OC, Ca, Si, PSO4 and EC, which accounted for 9.0%, 7.0%, 6.4%, 4.6% and 4.3% 710 of total primary PM2.5 emissions respectively. The main species of NMVOCs are relatively high proportion of NMVOC, accounting for 21.9% of the total. Among these, 713 aldehydes, ketones, alcohols, and esters accounted for 5.0%, 4.4%, 9.0% and 3.5%