On the impact of excess diesel NO X emissions upon NO2 pollution in a compact city

NO X emissions from diesel light-duty-vehicles (LDV) largely exceed the Euro emission standards in real-world driving conditions. Recent studies have quantified their impact upon air quality and human health primarily based on air quality models at mesoscale and large-scale resolutions. Here, we show that these approaches can significantly underestimate the impact of diesel LDV excess NO X emissions upon NO2 pollution in cities, particularly in the more compact and heavily trafficked ones. We compare an air quality mesoscale model at both 4 and 1 km resolution with a street-scale model in Barcelona, a compact city where the EU annual NO2 limits are repeatedly exceeded and a large share of passenger cars are diesel (65%). We compare consistently two emissions scenarios: a business-as-usual scenario where diesel LDV emit NO X in excess, and a counterfactual standard limits scenario where emissions are compliant with the Euro emission standards. We first show that in contrast to the mesoscale model, the street scale model is able to largely represent the observed NO2 concentration gradients between traffic and background stations in the city. In a second step, we find that the mesoscale model strongly underestimates the impact of diesel LDV excess NO X emissions upon NO2 pollution both in absolute terms (by 38%–48%) and relative terms (by 10%–35%). We argue that such underestimated impacts should be considered when assessing NO X excess emissions by LDV in cities. Using the street scale model, we find that diesel LDV excess NO X emissions are associated with about 20% of NO2 levels in the city, contributing substantially to an increased number of citizens exposed to high NO2 pollution in Barcelona.


Introduction
In 2017, 16 European Union (EU) countries reported NO 2 exceedances of the annual limit value enforced by the European Air Quality Directive (40 µg m −3 ), with 86% of them occurring at urban traffic monitoring stations (EEA 2019). Exceedances are mostly due to emissions from vehicles, especially diesel light-duty-vehicles (LDV) (Ntziachristos et al 2016, Degraeuwe et al 2017 which represent about 40% of the European fleet, and whose emissions fail to meet the Euro emission standards under real-world driving conditions (Weiss et al 2011, Thompson et al 2014, Lewis et al 2015. Hereafter we will refer to diesel emissions above the Euro emission standards as excess diesel NO X emissions. Assessments of excess NO X emissions impacts on air quality and health have relied on air quality models at mesoscale or large-scale resolutions and have mostly focused on PM 2.5 and O 3 . For instance, Barrett et al (2015) tackled specifically the impact of Volkswagen diesel LDV equipped with emission defeat devices (Thompson et al 2014). Based on estimated PM 2.5 increases calculated with a chemistry-transport model at 50 km resolution, this study attributed to the so-called 'dieselgate' scandal ∼59 premature deaths in the US. Anenberg et al (2017) attributed as much as ∼38 000 premature deaths to excess diesel NO X Figure 1. CALIOPE domains for EU, IP, and CAT (a-c), and the CALIOPE-Urban domain over Barcelona city (d). The panels depict NO2 concentration at surface level. The black dots in (d) represent the location of the monitoring stations used in this work. The lines in (d) depict areas mentioned in this study: solid-white is Barcelona municipality, dashed-white is harbour, dashed-grey are ring roads and solid-grey is Eixample district in the city center. emissions worldwide via increases in PM 2.5 and O 3 based on a model at 2 • by 2.5 • downscaled to 0.1 • by 0.1 • resolution. Von Schneidemesser et al (2017) estimated the impact of excess emissions upon NO 2 concentrations at 16 measurement sites in Berlin based on a model at 1 km resolution and observations, and found that NO X traffic emissions would be reduced by 30%-55% if diesel LDV would comply with the regulatory standards. Using a chemical transport model at 0.1 • by 0.1 • , Jonson et al (2017) estimated ∼5000 premature deaths from PM 2.5 and O 3 in the adult population due to excess diesel NO X emissions in EU28, Norway and Switzerland in 2013. Chossière et al (2018) showed that a large fraction of the health impacts from changes in PM 2.5 and O 3 in Europe are trans-boundary.
We identify two gaps in previous studies. First, they omit the effect of NO 2 upon health. It has been argued that the relationship between NO 2 and health is not as well-established as for PM 2.5 (Jonson et al 2017). Although the role of NO 2 as a surrogate of other measured or unmeasured pollutants cannot be completely ruled out, a variety of studies are consistent with long-(e.g. Faustini et al 2013, Sunyer et al 2015, Atkinson et al 2018, Achakulwisut et al 2019 and short-term (Samoli et al 2006, Chiusolo et al 2011 NO 2 effects upon health. Given that NO 2 is prominent in populated cities, omitting its effect may substantially bias the estimated health impact of excess NO X emissions. Second, health effect assessments can be strongly affected by the methods for evaluating exposures; in this sense air quality model-based assessments using models at mesoscale or large-scale resolutions tend to underestimate pollutant levels and the associated health impacts (Li et al 2016, Korhonen et al 2019 as they do not capture intra-urban and near-roadway exposure gradients (Greco et al 2007, Karner et al 2010, Borge et al 2014. This is particularly important in compact cities for NO 2 and other species within the PM 2.5 fraction. Our study focuses on the second gap. Mesoscale air quality models cannot represent NO 2 gradients and underestimate NO 2 levels in compact cities (Borge et al 2014, Duyzer et al 2015. We provide here a robust quantification of the impact of excess diesel NO X emissions upon NO 2 levels in a compact city. To the best of our knowledge, this study presents the first such assessment using a streetscale air quality model. We focused on Barcelona city (Spain) (figure 1(d)), a densely populated and trafficked urban area (about 5500 vehicles km −2 ) where the annual NO 2 limit value set up by the European Air Quality Directive (40 µg m −3 ) has been exceeded uninterruptedly since year 2000, the majority of passenger cars are diesel (65%) (Barcelona City Council 2017a) and road transport is the main contributor to the chronic NO 2 exceedances. According to the local Public Health Agency about 70% of its 1.6 million citizens were exposed to NO 2 concentration levels above the annual air quality limit value in 2017. These NO 2 exposure levels were associated with about 929 premature deaths in Barcelona city in 2017 (ASPB 2018).
Our underlying hypothesis is that prior modeling studies at mesoscale and large-scale resolutions have substantially underestimated the impact upon NO 2 in cities, particularly in the more compact and heavily trafficked ones. We compare modeled NO 2 concentrations calculated consistently at 4 km, 1 km and street-level resolutions using an air quality multiscale modeling system fed by a state-of-the-art bottom-up emission model for two scenarios: business-as-usual (BAU) and standard limits (SL) diesel LDV NO X emissions. The BAU scenario represents real-world driving conditions and the SL scenario represents diesel LDV emissions complying with Euro emission standards.

Domain and period of study
Barcelona city covers an area of 101 km 2 (figure 1(d)). The simulation period for the case study is from the 9th to the 25th of November 2017. This period is selected because it is representative of late autumn air quality levels in Barcelona, including both an episode of high NO 2 concentrations and days that are representative of the observed annual mean daily cycle in the city. During this period the NO 2 hourly limit set up by the European Air Quality Directive (200 µg m −3 ) was exceeded two times at the Gràcia-Sant Gervasi (i.e. Gràcia) traffic monitoring station in Barcelona. The high concentration levels recorded were due to a strong temperature inversion causing stagnant air.

Multiscale air quality modeling: from mesoscale to street-level
We used the CALIOPE-Urban street-scale modeling system which consists of the CALIOPE air quality mesoscale modeling system (Baldasano et al 2011)  . CALIOPE was run over a domain covering Europe at a 12 km by 12 km horizontal resolution (figure 1(a)), Iberian Peninsula at 4 km by 4 km ( figure 1(b)), hereafter referred to as IP-4 km, and the Catalonian domain, including Barcelona, at 1 km by 1 km, hereafter referred to as CAT-1 km (figure 1(c)). CALIOPE results have been evaluated in detail elsewhere (e.g. Pay et al 2014). For the mesoscale model, simulations were initialised with the ECMWF reanalyses (ERA-Interim), boundary conditions for chemistry come from the CAMS reanalysis of atmospheric composition (Inness et al 2019) and pollutant emissions are obtained from HER-MESv3. For the European parent domain (EU-12 km) HERMESv3 was run using the TNO-MACIII inventory (Kuenen et al 2014) for European countries and the HTAPv2.2 inventory (Janssens-Maenhout et al 2015) for countries outside Europe. For the nested mesoscale domains covering the Iberian Peninsula (IP-4 km) and Catalonia (CAT-1 km) the same approach was applied except for Spain, where high resolution detailed emissions were estimated using the bottom-up module of HERMESv3 (Guevara et al 2020). The coupling with R-LINE estimates local traffic dispersion driven by channeled street winds and vertical mixing with background air taking into account atmospheric stability and street morphology (e.g. aspect ratio). R-LINE applies the Generic Reaction Set (GRS) to resolve simple NO to NO 2 chemistry (Valencia et al 2018). Further information regarding CALIOPE-Urban's methodology and its evaluation using NO 2 observed concentrations can be found in Benavides et al (2019). To obtain highresolution concentration maps for the entire Barcelona city, hereafter referred to as BCN-20 m, we set the domain over Barcelona city as the minimum rectangle where Barcelona municipality is contained and extended it by 250 m buffers that include the highways surrounding the city (figure 1(d)). This domain is covered by a regular receptor grid of 20 m resolution.

Road traffic emissions
Road traffic emissions were estimated using the bottom-up traffic emission module from HERMESv3 (Guevara et al 2020) for the IP-4 km and CAT-1 km mesoscale domains as well as for the BCN-20 m urban domain. This allows to maintain consistency between the different scales and therefore obtain comparable results. Estimated hourly road link-level vehicle emissions were conservatively mapped onto the gridded mesoscale domains (IP-4 km and CAT-1 km) and adapted to the requirements of BCN-20 m. HERMESv3 estimates hot and cold exhaust road transport emissions combining the tier 3 method described in the 2016 EMEP/EEA air pollutant emission inventory guidebook (EMEP/EEA 2016), which is fully incorporated in the COPERT 5 software, with a digitised traffic network. This traffic network contains specific information by road link for daily average traffic, mean speed circulation, temporal profiles and vehicle fleet composition profiles. For the city of Barcelona, this information was obtained from multiple sources, including the local automatic traffic counting network (Barcelona City Council, personal communication), fleet composition from a remote sensing device campaign in 2017 (Barcelona City Council 2017a) and TomTom historical average speed profiles (TomTom 2019) based on GPS data gathered from circulating vehicles between 2015 and 2016. In HERMESv3, we further extended the degradation factors for diesel vehicles taking into account mileage reported by COPERT 5 for gasoline vehicles. For EURO 2 and EURO 3, this extension considers a deterioration of tailpipe NO X emissions of 22% and 10% respectively, as suggested by Chen and Borkenkleefeld (2016). HERMESv3 disaggregates the calculated NO X emissions into NO and NO 2 using vehicledependent speciation factors that are extracted from EMEP/EEA (2016) (chap. 1.A.3.b.i-iv, table 3.87) and the investigations of Rappenglueck et al (2013) and Carslaw et al (2016). HERMESv3 also estimates the emissions released from other anthropogenic sources, which are considered in the mesoscale model and their contributions are taken into account in the street-scale model as background concentrations.

Emission scenarios
We defined two scenarios to quantify the effect of excess diesel LDV NO X emissions in Barcelona City. The BAU scenario is our best estimate of NO X emissions under real-world driving conditions for Barcelona city in 2017. It incorporates emissions representing real-world driving conditions, which is supported by (a) the use of COPERT 5 realworld adjusted NO X emission factors for EURO 5 and 6 diesel vehicles, (b) the inclusion of a mileage correction factor for diesel vehicles and (c)   2). As shown in figure 2, the combination of COPERT 5 emission factors with the degradation factors is in close agreement with the RSD-derived NO X emission factors for the prevalent EURO categories. COPERT-derived emission factors were calculated considering an average speed of 28 km h −1 , the same speed at which RSD measurements were performed. This speed value is within the range of the common vehicle's average speed within the urban area of Barcelona (Barcelona City Council 2017b). Table 1 shows the two vehicle fleet distributions used in HERMESv3 for Barcelona city. The profiles divide the road network within the city into two main zones, ring roads and urban streets within these ring roads, to account for observed differences in the composition of the circulating vehicles. In Eixample district, the percentage of mopeds and motorbikes is increased up to 25% to better represent the observed fleet (Barcelona City Council 2017a) while the ratio among other vehicle types is kept the same as in the urban streets. Diesel LDV predominate in both the urban fleet (60.2%) and in the ring roads (61.6%) vehicle fleet distributions. The main difference is that diesel trucks are more frequent in the ring roads (8%) than in the inner city (1%), while mopeds are not allowed to circulate in the ring roads (0% versus 2.6%). The counterfactual scenario is the SL scenario, which assumes that diesel LDV emissions comply with the Euro emission standards (European Commission 2007). It therefore considers that emission  1(d)) during the period of study (9-25th November 2017). Standard statistics are described in Chang and Hanna (2004) and are computed with model hourly results of CALIOPE-Urban, CALIOPE-1 km and CALIOPE-4 km systems. FAC2 refers to the fraction of model results within a factor of 2 of observations, MB is the mean bias, RMSE is the root-mean-square error and r is the correlation coefficient. Bold numbers represent model results with better performance for each statistic and site.

Model evaluation
We first provide an evaluation of the BAU scenario using hourly NO 2 concentrations reported by the official monitoring network in Catalonia (XVPCA) for the year 2017 in the only two traffic stations available in Barcelona (i.e. Gràcia and Eixample) and in one of the urban background stations (Palau Reial), which is considered to be representative of the models behaviour at this site type, as shown in figure 1(d). The model performance at the other urban background sites can be found in the supplementary material (available online at stacks.iop.org/ERL/16/024024/mmedia) and for a more complete evaluation near traffic in Barcelona we refer to Benavides et al (2019). Table 3 shows standard statistics computed using the modeled and measured concentrations and figure 3 depicts the NO 2 average daily cycle. All the simulations tend to slightly underestimate NO 2 and behave similarly well in the urban background station. Yet, IP-4 km and CAT-1 km are strongly biased in the traffic stations.
As expected BCN-20 m is able to better reproduce the observed values at the traffic stations. At the Gràcia site, it reduces the underestimation by ∼50% (∼−16 vs ∼−33 µg m −3 ) and at the Eixample site, by a factor ∼5 (∼4 vs ∼−20 µg m −3 ). Between the traffic sites, we relate the distinct behaviour at each site to the relative influence of local traffic in each site, to the influence of the main trafficked areas and to the specific micro-meteorological patterns. Table 4 shows the annual total NO, NO 2 and NO X traffic emissions estimated for Barcelona city for the BAU and the SL scenarios. NO X traffic emissions decrease by ∼27% on average when diesel LDV are assumed to comply with EU standard limits. The decrease in primary NO 2 emissions is much stronger (∼49%) than in NO emissions (∼22%). This is   Carslaw et al (2016) found the NO 2 /NO X ratio for diesel cars EURO 4, 5 and 6 to range from 0.25 to 0.34 whereas for EURO-equivalent petrol cars it ranged from 0.05 to 0.12. Table 5 shows the intra-urban variability of NO X emissions per km 2 at road-link level for both scenarios and the relative difference of these scenarios. The impact upon NO 2 emissions, with relative reductions ranging from −11.4% to −58.1%, is stronger than upon NO (−2.2% to −36.4%). Figure 4 shows the spatial distribution of NO X traffic emissions in Barcelona city at road-link level for both scenarios and the relative difference thereof. The relative reductions in NO X emissions (figure 4(i) range only from 2.2% to 10% in the harbour area (figure 1(d) because the fleet is dominated by heavy duty vehicles, from −20% to −30 % in the very builtup Eixample city center district, and from −30% to −40 % in most of the other districts in the city and the ring roads ( figure 1(d)). The lower difference in the centre district may be caused by the greater share of mopeds and motorbikes in that district (25% according to Barcelona City Council 2017a). The impact on NO 2 emissions (figure 4(f)), with a relative difference ranging from 40% to 60%, is greater than it is on NO (figure 4(c)), ranging from 20% to 40%. Figure 5 shows the average NO 2 concentration maps for BCN-20 m, IP-4 km and CAT-1 km. The two mesoscale outputs were interpolated using bilinear interpolation to the 20 m resolution grid for comparison purposes. For the three cases, we show the BAU and the SL scenarios along with the absolute and relative differences thereof. The spatial detail of BCN-20 m enables characterising the impact of excess diesel LDV NO X emissions at street-level; the median value of the average BAU concentrations across the 20 m receptors is 58.9 µg m −3 (interquartile ranges (IQR) from 46.0 to 75.8 µg m −3 ); the median value of the absolute difference between average BAU and SL concentrations is −10.5 µg m −3 (IQR from −6.7 to −14.7 µg m −3 ); and the median value of the relative difference  is −20% (IQR from −15% to −21%). These differences are consistent with the daily results as seen in section S2 figure S2 in the supplementary material.

Impact of excess diesel LDV emissions upon NO 2 concentration
The largest differences are generally found in the areas with highest NO X traffic emissions (figure 4). In fact, the relative differences in NO 2 levels between  scenarios broadly scale with the average NO 2 concentration (figure 6), a feature that is not reproduced in IP-4 km and CAT-1 km (not shown). Diesel LDV NO X emissions strongly contribute to NO 2 concentrations in the ring roads and some major streets that act as the main entrance routes towards the city center. Additionally, some districts such as the Eixample are more affected to excess diesel emissions than others mainly due to the combination of the aspect ratio of the streets and the traffic intensity. In contrast, in areas where NO 2 concentrations are predominantly affected by other sources, the difference between scenarios is comparably low. For instance, in the area surrounding the harbour, the impact of LDV is lower than in other parts of the city due to the high share of diesel high-duty-vehicles operating there (around 45% of the total circulating vehicles). This area is represented by the points showing relative differences below 10% in figure 6. IP-4 km and CAT-1 km show very different spatial patterns; the steep spatial gradients appearing in BCN-20 m are smoothed out and concentrations in the BAU scenario only reach a median value of 41.7 µg m −3 (IQR from 39.5 to 44.7 µg m −3 ) and 38.6 µg m −3 (IQR from 35.0 to 48.4 µg m −3 ), respectively, a 29% (IQR from 14% to 41%) and 34% (IQR from 24% to 36%) less compared to BCN-20 m.

Discussion and conclusions
To quantify the impact of excess NO X diesel LDV emissions upon NO 2 concentrations over Barcelona city we compared simulations with two different emission scenarios: a business-as-usual (BAU) scenario representing NO X diesel LDV emissions under real-world driving conditions and a counterfactual standard limits (SL) scenario, which represents NO X diesel LDV emissions compliant with the Euro emission standards.
We first showed that, in contrast to the mesoscale model, the street-scale model is able to reproduce the NO 2 concentration gradients observed in the city between open areas and trafficked zones. We found a decrease on the order of −30% in total NO X and −50% in total primary NO 2 emissions, consistent with other studies (e.g. Von Schneidemesser et al 2017). The differences are not homogeneous across the city; we estimated higher NO X emissions in the very built-up city center and near the ring roads. Overall this translated into a reduction of median NO 2 concentrations and interquartile ranges in Barcelona municipality of −10.5 µg m −3 (−6.7 to −14.7 µg m −3 ), −6.5 µg m −3 (−5.6 to −8.1 µg m −3 ), and −5.5 µg m −3 (−5.0 to −6.0 µg m −3 ) in absolute terms and −20% ( from −15% to −21%), −18% (from −15% to −19%), −13% (from −13% to −14%) in relative terms using BCN-20 m, CAT-1 km and IP-4 km, respectively. In other words, the mesoscale simulations underestimated the absolute reductions in NO 2 concentrations by ∼38% and ∼48% and the relative reductions by 10% and 35% compared to the street-scale model when NO X diesel LDV emissions were assumed to comply with Euro emission standards.
Other prior studies (e.g. Anenberg et al 2017, Jonson et al 2017) have used low resolution models to characterise the impacts upon PM 2.5 and O 3 . While our study focuses only on NO 2 , it is likely that PM 2.5 or at least a fraction of it is largely underestimated in compact cities at mesoscale resolutions. Stronger underestimations on both the absolute and relative reductions are to be expected as model resolution decreases. Thus, street models are strongly recommended to avoid underestimations in impact assessments of alternative emission scenarios in compact cities.
The largest differences in NO 2 between scenarios were found in the areas with the highest NO X traffic emissions. Using observations, Von Schneidemesser et al (2017) estimated a potential NO 2 reduction in traffic sites across Berlin ranging from −9 to −23 µg m −3 if diesel LDV would comply with the EURO standards. Similarly, we found an average reduction ranging from −13 to −18 µg m −3 at traffic sites during the period of study using the street-scale model. Despite the large improvements, these sites would still not meet the EU and WHO annual limit values assuming that the relative differences between scenarios during our study are maintained during the whole year (40.6 µg m −3 in Gràcia and 45.6 µg m −3 in Eixample sites).
Our results imply an increase in the number of citizens exposed to unhealthy NO 2 levels when comparing the BAU with the SL scenario. We estimate that 90.2% of citizens would be exposed to NO 2 levels above 40 µg m −3 under the BAU scenario, a percentage that would decrease to 76.6% within the SL scenario during the study period (see section S3 and figure S3 in supplementary material). Assuming that the relative difference is maintained throughout the year 2017, the percentage of Barcelona citizens exposed to higher concentrations than the annual limit would have been reduced from 70% (ASPB 2018) to 59%. Future studies should further investigate the associated health implications in a city where exposure to NO 2 was associated with about 929 premature deaths in 2017 (ASPB 2018).
Based on our findings, diesel LDV should actually meet the EURO standards under real-world driving conditions to substantially reduce NO 2 concentration levels in cities with a high proportion of diesel LDV and chronic NO 2 exceedances. Moving towards cleaner LDV in these cities is likely not enough to meet the EU and WHO annual limit values for NO 2 in trafficked streets and other policies targeting the overall reduction of LDV circulating should be encouraged.

Data availability statement
The data that support the findings of this study are available upon reasonable request from the authors. from the urban planning studio 300 000 km s −1 for providing the census-level population dataset disaggregated at the address level. The authors acknowledge PRACE and RES for awarding access to Mar-enostrum4 based in Spain at the Barcelona Supercomputing Center through the eFRAGMENT2 and AECT-2020-1-0007 projects. Jaime Benavides was funded by grant BES-2014-070637 from the FPI programme by the Spanish Ministry of the Economy and Competitiveness. The authors also acknowledge the support from the Ministerio de Ciencia, Innovación y Universidades (MICINN) as part of the BROWNING project RTI2018-099894-B-I00 and NUTRIENT project CGL2017-88911-R and the support from the Agencia Estatal de Investigacion (AEI) as part of the VITALISE project (PID2019-108086RA-I00/AEI/10.13039/501100011033). Carlos Pérez García-Pando acknowledges the longterm support from the AXA Research Fund, as well as the support received through the Ramón y Cajal programme (Grant RYC-2015-18690) of the Spanish Ministry of Economy and Competitiveness.