Spatial inter-comparison of Top-down emission inventories in European urban areas

25 This paper presents an inter-comparison of the main Top-down emission inventories 26 currently used for air quality modelling studies at the European level. The comparison is 27 developed for eleven European cities and compares the distribution of emissions of NO x , 28 SO x , VOC and PPM 2.5 from the road transport, residential combustion and industry sectors. 29 The analysis shows that substantial differences in terms of total emissions, sectorial emission 30 shares and spatial distribution exist between the datasets. The possible reasons in terms of 31 downscaling approaches and choice of spatial proxies are analysed and recommendations are 32 provided for each inventory in order to work towards the harmonisation of downscaling and proxy calibration, in particular for policy purposes. The proposed methodology may be useful for the development of consistent and harmonised European- wide inventories with the aim of reducing the uncertainties in air quality modelling activities.

the SNAP nomenclature has been made to allow comparing with the other databases. The 175 simplified version of the mapping scheme from IPCC to SNAP codes is included in the SI 176 (Table 3)  An overview of the spatial proxies and ancillary data used for the spatial distribution of 235 emissions from the considered sectors is shown in Table  1. 236  For NO x , the share for road transport varies from 37% (EDGAR) to 43% (INERIS), while 256 differences are between 1% and 2% for the residential combustion and the industrial sectors. 257 The most noticeable difference for NO x is in the EDGAR emission inventory, as it assigns 258 more emissions (~ 7%) to SNAP01 (Combustion in energy and transformation industries), 259 which is compensated by a lower share of emissions in SNAP08 (Non-Road transport). 260 For PPM 2.5 , the share of emissions from SNAP02 (Residential) ranges from 38% (EDGAR) 261 to 48% (GAINS, on which JRC07 is based) with the exception of EMEP, which assigns 262 much more importance to this sector (54%). This difference between EMEP and the other For SNAP07 (Road Transport) and SNAP34 (Industry), we find the same pattern reported 268 for NO x , with EDGAR assigning to industry 5% higher emissions than MACC2 and ~10% 269 higher than the other inventories, while reporting a ~6% lower share of the road transport 270 sector. Similar variations are also seen for the agricultural sector (SNAP10). 271 In the case of SO 2 , no major difference is observed between the inventories, although it has to 272 be noted that the road transport sector is of negligible importance. Looking at the target sectors for VOC, while there is a good agreement for SNAP02 and 283 SNAP07, EDGAR has higher emissions for the industrial sector for EDGAR. This is most 284 likely an allocation issue, since this difference is partially compensated by an underestimation 285 in SNAP06 (Solvents and other Products use). This compensation between sectors may indicate a potential inconsistency in the mapping of industrial activities related to the use of 287 solvents (e.g. pharmaceutical products, paint manufacturing). This inconsistency highlights 288 that differences in the original mapping and linking tables used in each inventory to match 289 specific pollutant activities to an official reporting format (e.g. NFR to SNAP) may have a 290 large impact when re-mapping activities from one reporting nomenclature to another. 291 292 3.2. Comparison at regional/urban scale 293 294 We focus here on the regional allocation of emissions, i.e. on the fraction of the sum of 295 national emissions from SNAP02, SNAP34 and SNAP07 which is assigned to a particular 296 city ( Figure 2; as in the following figures, the cities are ordered on the x axis by degree of 297 longitude, West to East). All inventories perform similarly for NO x with an exception in 298

Emission totals
Budapest to which EDGAR assigns almost 30% of the national totals, almost twice the 299 percentage assigned by the other inventories. Budapest consistently shows the largest 300 differences between the inventories for all compounds. Large differences are also observed 301 for Paris, for all pollutants and especially for EDGAR and MACCII, and in Bucharest and 302 Sofia, for SO 2 and PPM 2.5 . The higher emission share in Paris according to MACCII could be 303 explained by an over-allocation of industrial emissions (SNAP34) to urban areas. Emissions 304 from the industrial sectors that cannot be linked to a specific point source are merged in 305 MACCII and gridded based on total population (Table 1) of view, EDGAR tends to allocate a larger fraction of the national totals to urban areas than 311 the other inventories, in particular for PPM 2.5 and SO 2 . The higher estimation ranges between 312 factors 1.5 and 2. The behaviour of INERIS in Bucharest NO x and SO 2 follows the average 313 trend while, for VOC and PPM 2.5 , it is outlying. As it appears from the analysis at sectorial 314 level in the next chapters, these higher values are likely due to higher emissions from the 315 industrial sector which represents the ~70% of the total emissions, a share much higher than 316 the ones reported for the other inventories (~5% -~40%). 317 In general, it is clear that the spatial disaggregation methods applied in each inventory work 318 differently in terms of urban areas and pollutants. 319 In order to better understand why the spatial allocation differs between inventories, we 325 compare the way each inventory spatially allocates the regional emission in terms of macro-326 sectors, more specifically, transport (SNAP07), industry (SNAP34) and residential 327 combustion (SNAP02). Some uncertainties could be present due to the way different 328 countries might convert sectors between the different nomenclatures (NFR, SNAP, IPCC). 329 For each urban area, the contribution of each macro-sector, which will partly depend of the 330 characteristics of the selected study site (Table 2, SI), it is assessed in terms of percentage of 331 the total city emission. The regional/city macro-sector percentages (C) are computed as represents the total city emission for a pollutant "p" and macro-sector "m" and M 335 is the total number of sectors (3 in our case). 336 In general, NO x and SO 2 show the most robust trend among the four pollutants, while it is not 337 possible to identify a consistent pattern for VOC in terms of cities or in terms of sectors. 338

360
The residential sector (SNAP02) shows good agreement among the inventories for NO x and, 361 in particular, there is no difference between TNO-MACCII and TNO-MACCIII ( Figure 5)   The need to select a reference pollutant is a disadvantage of this methodology as discussed in 395 Thunis et al. (2016b). However, in this work we follow an alternative approach that does not 396 require a reference pollutant. We assume that the activity and emission factor ratios behave as 397 random variables with probability distributions following a Gaussian law centered around 1. 398 These distributions are then used to estimate the probability that the activity and emission 399 factors ratios take specific values within given intervals, while satisfying the known 400 constraint on total emission ratios. The activity and emission factor ratio are then those 401 characterised by the highest probability. The activity and emission factor ratio are used as X 402 and Y coordinates in the "diamond" diagram, where each sector-pollutant couple is 403 represented by a specific point ( Figure 6). As a result of the construction, the diagonals (slope 404 = −1) provide information on the overall under-/over-prediction in terms of total emissions. 405 We can define a diamond shaped area where activity, activity shares and total emissions all 406 remain within given degrees of variation. For example, the red diamond indicates ratios of 407 activity, emission factor and total emissions all within 100% (or a factor 2) differences, while 408 the green diamond indicates ratios within 50% (or a factor 1. It is noteworthy to remark that, in this work, urban emission totals are further scaled by their 428 country totals as explained in the methodology. This step is made to ensure that all urban 429 inventories originate from similar country totals and that the observed differences in the 430 diamond approach focus on the differences in terms of spatial allocation of the emissions 431 rather than on country scale biases. In this particular case, the value on the X axis is now an 432 indication of the differences in terms of activity shares rather than in terms of emission 433 factors. 434

Analysis in terms of sector 435
Transport Sector -There is an overall agreement between the inventories both in 436 terms of activity intensity and sectorial share as indicated by the fact that most points are 437 concentrated within the diamond shape ( Figure 6). This is probably explained by the fact that 438 similar proxies are used for the spatial and sectorial disaggregation from the country totals, 439 allowing to allocate similar amounts of emissions to the considered study areas. Indeed, the indicating that most of the differences between inventories tend to originate from differences 449 in country total estimates rather than from the spatial disaggregation proxies. improvements. In particular, EDGAR consistently has higher emissions from this sector, 513 while INERIS and EMEP assign lower values (Figure 9). There are in general differences 514 between all the inventories and for all the pollutants that appear to be due to discrepancies 515 both in terms of activity levels and shares, as indicated by the wide horizontal and vertical 516 spreads of the points in Figure 9. included in this sector. Especially for small countries, the existence of threshold makes the 532 PRTR dataset less valuable and it requires additional data for point sources falling below the 533 threshold. Hence, the diffuse fraction has to be spatially allocated according to different 534 proxies that may greatly contribute to the inconsistencies among inventories (Table 1) The other two sectors, in particular the industrial sector, highlight problems with both activity 553 levels and activity shares. It is also interesting to note that in general the problems are similar 554 for all cities in each inventory. This might mean that specific parameters of each urban area, 555 such as land use, population density and degree of urbanization, play an important role in 556 emission distribution. 557

558
If we sum-up the emissions from the three sectors and use the diamond approach, we observe 559 greater consistency between the inventories for all pollutants than for single macro sectors 560 ( Figure 10). This consistency results from the compensation effects of higher and lower 561 estimations in the individual macro-sectors. This is particularly notable for the EDGAR 562 inventory, where the estimates of traffic emissions, which are lower when compared to the 563 other datasets, are compensated by higher ones from the industrial and residential sectors. 564 The largest consistencies are mostly observed for NO x and VOC and the lowest for PPM 2.5 565 and SO 2 . For SO 2 , the discrepancy mostly lies in the sectorial share as indicated by the large 566 horizontal spread. It is interesting to note the differences for SO 2 between MACCII and 567 MACCIII which are important in cities like Budapest, but small in others like Paris. These 568 differences can be attributed to changes in the proxies used to distribute industrial emissions 569 resulting in differences in terms of share. The proxies for industrial activities were in fact a It has also to be noticed that the uncertainty is very low and similar for all pollutants, 600 including PPM 2.5 and VOC which, differently from NOx and SOx, have a significant 601 contribution from non-exhaust emissions. This entails that even if non-exhaust emissions M A N U S C R I P T

A C C E P T E D ACCEPTED MANUSCRIPT
from resuspension is a major source of uncertainty in national inventories, they are spatially 603 distributed in the same way by the different dataset we analysed. 604 Furthermore, it is also interesting to note the overall good agreement for Utrecht and 605 Barcelona. When looking at the combination of all the considered sectors, uncertainties are 606 generally reduced for all cities and pollutants, due to a compensation effect, although they are 607 still very high for SO 2 and some cities such as Paris, Bucharest, Budapest and Sofia. 608 Hence, while for the most important cities, bottom-up inventories often do exist providing 625 more accurate information at a higher spatial resolution, for extensive air quality modelling it 626 is still of utmost importance to be able to rely on consistent and harmonised European-wide 627 inventories. In order to assess the potential impact of the choice of a specific inventory for air 628 quality modelling, we analysed their spatial patterns of behaviour looking at representative 629 urban areas over Europe. 630

Conclusions
A distinctive outcome of the work presented in this paper is the significant difference 631 between regional emission inventories due to the choices made in terms of disaggregation 632 bottom-up estimates to better calibrate the spatial patterns of emissions from wood and coal 656 burning, in order to reflect the significant variations between countries. Furthermore, city-657 specific features such as district heating should be taken into account; in these cases, a much 658 lower share of residential emissions would be expected over the city compared to individual 659 heating sites. At the same time, the traditional proxies used for gridding residential emissions 660 (e.g. population density) would not be any more relevant. 661 Based on the differences highlighted in this analysis, we list the main aspects for each 662 inventory that could be important to review. It has to be noted that these issues have been Further work will be needed in order to provide a deeper insight into emission spatial patterns 708 through a comparison at a finer scale with local bottom-up inventories, which rely on massive 709 and detailed spatial information such as point sources, detailed censuses and traffic statistics 710 or, as alternative, with the national grids at 0.1*0.1 degrees resolution recently reported to 711 EMEP by many European countries. Such a comparison would help calibrate proxies at a 712 regional/local scale rather than using common ones for such diverse and extended areas. 713 Finally, and considering that one of the main aims of the analysed inventories is to provide 714 emission inputs for air quality modelling, future work should also consider the influence of 715 uncertainties in proxy-based emission inventories when they are used in atmospheric 716 chemistry models.