UK surface NO2 levels dropped by 42% during the COVID- 19 lockdown: impact on surface O3

We report changes in surface nitrogen dioxide (NO2) across the UK during the COVID-19 pandemic when large 10 and rapid emission reductions accompanied a nationwide lockdown (23rd March—31st May, 2020, inclusively), and compare them with values from an equivalent period over the previous five years. Data are from the Automatic Urban and Rural Network (AURN) that form the basis of checking nationwide compliance with ambient air quality directives. We calculate that NO2 reduced by 42±9.8% on average across all 126 urban AURN sites, with a slightly larger (48±9.5%) reduction at sites close to the roadside (urban traffic). We also find that ozone (O3) 15 increased by 11% on average across the urban background network during the lockdown period. Total oxidant levels (Ox = NO2 + O3) increased only slightly on average (3.2±0.2%), suggesting the majority of this change can be attributed to photochemical repartitioning due to the reduction in NOx. Generally, we find larger, positive Ox changes in southern UK cities which we attribute to increased UV radiation and temperature in 2020 compared to previous years. The net effect of the NO2 and O3 changes is a sharp decrease in exceedances of the NO2 air quality 20 objective limit for the UK, with only one exceedance in London in 2020 up until the end of May. Concurrent increases in O3 exceedances in London emphasize the potential for O3 to become an air pollutant of concern as NOx emissions are reduced in the next 10-20 years.


Introduction
The current Coronavirus SARS-CoV-2 (COVID-19) outbreak was first identified in Wuhan, China, in December 2019, and was recognised as a pandemic by the World Health Organization (WHO) on 11 March 2020 (WHO, 2020). As of early August 2020, there have been almost 18 million confirmed cases and over 700,000 deaths reported across the world (https://coronavirus.jhu.edu/map.html). Efforts to prevent the virus spreading have 30 included severe travel restrictions and the closure of workplaces, inevitably leading to a significant drop in emissions of primary air pollutants from several important sectors. This has provided a unique opportunity to examine how air pollutant concentrations respond to an abrupt and prolonged perturbation, followed by policyrelatable increases as restrictions are incrementally relaxed.

35
The effects of the change of emissions on NO2 and O3 have been observed using satellite and in-situ measurements in several studies. Table 1 summarizes studies from a growing body of work that report changes in NO2 and other air pollutants in countries across the world that are associated with the global COVID-19 lockdown, including satellite observations (Liu et al., 2020) and in situ measurements. These studies have used various methods to isolate the impact of the COVID-19 lockdown on changes in air pollutants from confounding factors, e.g. meteorology, using atmospheric chemistry transport models and weather normalisation techniques based on machine learning (ML) algorithms. In Europe, reductions of NO2 are typically slightly larger than we have seen in the UK in our study, perhaps reflecting more stringent lockdown policy. In Spain, NO2 was reduced by 50% at both urban traffic and urban background sites (Petetin et al., 2020) and in Rome, Turin and Nice, NO2 was reduced by 46, 30 and 63% respectively (Sicard et al., 2020). In all of these studies similar magnitude increases of O3 were observed, mainly attributed to the decreased NO. Further afield, in India TROPOMI satellite measurements showed that during the COVID-19 lockdown, there was a 18% decrease in NO2 over the whole country, with a 54% decrease over New Delhi compared to the same period in 2015-2019 (Pathakoti et al., 2020). In situ measurements in New Delhi showed a 53% decrease in NO2 and a 0.8% increase in O3 for the lockdown period compared to the 2 weeks immediately preceding it. In Rio de Janeiro, Brazil, there was a 24-33% decrease in NO2 50 during the lockdown compared to the week before (Dantas et al., 2020) and in Sao Paulo data from urban roadside sites showed a 54% decrease in NO2 compared to the previous 5 years (Nakada and Urban, 2020). In China, satellite observations showed a mean NO2 decrease of 21% decrease across the whole country, relative to a similar period in 2015-2019 (Bao and Zhang, 2020). In situ measurements in cities in northern China measurements before and after lockdown showed a 53% decrease in NO2 (Shi and Brasseur, 2020) and in-situ measurements in 55 cities across the whole of China showed a 60% decrease in NO2 comparing 1-24th January 2020 and 26th January -17th February 2020 (Huang et al., 2020). Both these two studies also reported a >100% increase in O3. These studies were both during wintertime so the O3 increase was largely attributed to the reduction on NO emissions reducing titration of O3 to NO2, however the possible effect of reduced particles on UV radiation and hence O3 production was also considered to have led to some of the increased O3. Le et al, 2020 use satellite data to show 60 a 71.9% decrease in NO2 and 93% decrease in Wuhan at the peak of the outbreak. They also report a 25.1% increase in O3 in Wuhan, largely attributed to a reduction in titration with NO. In the USA one study using EPA data showed a mean decrease of 30% of NO2 in urban areas of Seattle, Los Angeles and New York during the lockdown. The study did not show any consistent change in O3 levels (Bekbulat et al., 2020). Here, we report changes in nitrogen dioxide (NO2) across the UK and discuss them in context of observed changes in surface ozone (O3).
In 2018 the road transport sector accounted for 37% of UK NOx (sum of NO and NO2), the largest emission from a single sector, followed by energy industries (21%), non-road transport (mainly rail and aviation) (15%), manufacturing industries and construction (10%) and domestic combustion (9%)

70
(https://www.gov.uk/government/statistical-data-sets/env01-emissions-of-air-pollutants). In major cities, the contribution from road transport is typically much higher. On average across six cities in the UK (London, Bristol, Cardiff, Newcastle, Glasgow and Belfast)e.g. 47±6 % comes from road transport, , with 17±5 % from domestic combustion, 15±6 % from non-road transport sources (mainly rail), and 14±10 % from energy industries and 6±2 % from industrial combustion. In recent years, there has been a pronounced reduction in NOx emission (Defra,  At present the main pollutants of concern are NO2 and particulate matter with diameter smaller than 2.5 microns (PM2.5) in urban centres, and O3 in urban, suburban and rural environments, with exposure to excess levels of these species is known to have a negative effect on human health Kurt et al., 2016;Mannucci et al., 2015). O3 is a secondary air pollutant formed photochemically by the oxidation of volatile organic compounds 85 (VOCs) in the presence of NOx (Monks et al., 2015). It is generally lower in urban areas due to reactions with NOx, but in the past two decades over the UK (Finch and Palmer, 2020), and across the world (Fleming et al., 2018;Lefohn et al., 2018;Ma et al., 2016;Paoletti et al., 2014;Sicard et al., 2013;Sun et al., 2016), there have been large mean surface O3 increases in urban centres driven by reduced NOx emissions. In more rural environments, the opposite has been observed, with O3 decreasing with decreasing NOx emissions (Cooper et al, 90 2012;Cooper et al., 2014;Strode et al., 2015). Air pollution has led to an estimated 29,000 premature deaths/year in the UK, equivalent to 340,000 life years across the population in any one year and costs the UK economy between £10 billion and £20 billion/year (Royal College of Physicians, 2016). To meet the UK Government's clean air strategy (UK Government, 2019) and its commitment to achieve zero carbon emission target by 2050, sales of non-zero emission cars, vans and motorcycles will end by 2035. One of the challenges associated with 95 the progressive move to a low-NOx vehicle fleet in the UK is to understand the impacts on surface air pollution if other emissions are not reduced commensurately.
The widespread and rapid reduction in UK transport activity (and therefore the associated emissions) from the COVID-19 lockdown represents a natural experiment to study air pollution with a greatly reduced volume of NOx-100 emitting vehicles that we use as a proxy for a future low-NOx vehicle fleet. Figure 1 summarises the timeline of events associated with COVID-19 in the UK, including Google mobility data that describe the percentage changes in transport from a pre-lockdown baseline and daily mortality values reported by the UK Office of National Statistics. Google mobility data was used as a proxy for traffic counts as it is readily accessible, however for any quantitative analysis of the effect of reduction in traffic on pollution levels, real traffic counts or flow data would 105 be required. The UK Foreign and Commonwealth Office issued a travel advisory on 28 th January not to travel to mainland China. The first two UK cases of COVID-19 were confirmed on 31 st January, with a third case confirmed on 6 th February. As the number of cases continues to rise, the first UK death from COVID-19 was confirmed on the 5 th March. On that same day, the UK government moved from the "containment" to the "delay" phase of addressing COVID-19, which included, for example, social-distancing. A UK-wide lockdown was announced 110 nearly three weeks later on 23 rd March, with citizens instructed to stay at home with the exception of shopping for basic necessities and one form of exercise per day, medical needs and travel associated with key workers. An immediate effect of these restrictions in movement was a large and progressive drop in transport use, with an associated reduction in motor vehicles throughout the lockdown period. We use in situ measurements collected across the UK to examine how these reductions (and other changes) have affected NO2 in the UK, with a 115 discussion on how this could have, in turn affected O3. We also examine the changes in exceedances of limit values for NO2 and O3 and assess whether the COVID-19 lockdown can provide useful information on how air pollution will respond to future changes in emissions due to the move to a low-carbon economy. In the next section we discuss the data we use, in section 3 we describe our results for NO2 that we put into context in section paper in section 5.

In situ Measurements of NO2 and O3
We use data collected as part of the Defra Automatic Urban and Rural (AURN) network, currently consisting of 125 150 active sites across the UK ( Figure S1 and Tables S1 and S2) and is the main network used for compliance reporting against the Ambient Air Quality Directives. It includes automatic air quality monitoring stations measuring oxides of nitrogen (NOx), sulphur dioxide (SO2), ozone (O3), carbon monoxide (CO) and particles (PM10, PM2.5). Online measurements of VOCs are available at a small number of sites. These sites provide hourly information which is communicated rapidly to the public, using a wide range of electronic media and web 130 platforms. More detail can be found at https://uk-air.defra.gov.uk. Three different site types are used in this analysis. Urban traffic sites are defined as being in continuously built-up urban areas, with pollution levels predominantly influenced by emissions from nearby traffic. Urban background sites are located such that pollution levels are not influenced significantly by any single source or street, but rather by the integrated contribution from all sources upwind of the stations. These can be considered more representative of residential areas. Rural 135 background sites are sited more than 20 km away from agglomerations and more than 5 km away from other builtup areas, industrial installations or motorways or major roads, so that the air sampled is representative of air quality in a surrounding area of at least 1,000 km 2 .
The AURN network uses standardised techniques and operating procedures to ensure data are comparable. Full 140 details can be found at https://uk-air.defra.gov.uk/assets/documents/reports/empire/lsoman/ but a brief description will be given here. Nitric oxide (NO) in the sample air stream reacts with O3 in an evacuated chamber to produce activated nitrogen dioxide (NO2*). This then returns to its ground (un-activated) state, emitting a photon (chemiluminescence). The intensity of the chemiluminescent radiation produced depends upon the amount of NO in the sampled air. This is measured using a photomultiplier tube (PMT) or photodiode detector, so the detector 145 output voltage is proportional to the NO concentration. The ambient air sample is divided into two streams. In one stream, the ambient NO2 is reduced to NO (with at least 95% efficiency) using a molybdenum catalyst converter before reaction. The molybdenum converter should be at least 95% efficient at converting NO2 to NO. External gas cylinders or an internal permeation oven and zero air scrubber are used to provide daily automatic check calibrations for NO. The NO2 conversion efficiency is checked every 6 months using either an NO2 calibration 150 cylinder or gas phase titration of the NO with O3. In recent years it has become well established that NO2 measurements using molybdenum converters can overestimate NO2 due to interferences from other oxidised nitrogen species (e.g. HNO3, PAN, HONO) (Steinbacher et al., 2007). However, in urban environments the interferences are often minimal compared to the levels of NOx (Villena et al., 2012). In addition, as we are looking at a change in NO2, it is likely that any interference that is present will be there in very similar amounts in both the 2020 and 2015-2019 data. Ozone is measured by UV absorption at 254nm, with concentrations calculated using the Beer-Lambert Law (Parrish and Fehsenfeld, 2000). An O3-removing scrubber is used to provide a zeroreference intensity. An internal ozone generator and zero air scrubber are used to provide daily automatic check calibrations and instruments are calibrated with a primary ozone standard every 12 months. Whilst the accuracy of the measurement will vary on a site by site basis, the maximum allowed uncertainty for the AURN network is 15% for NO2 and O3 measurements. To study the effect of the lockdown on NO2 levels in the UK, we use measurements from 66 Urban Traffic and 62 Urban Background sites across the UK, all that have measurements between 2015 to the end of May 2020.

165
Measured meteorological data (wind direction, wind speed and temperature) is not available at most AURN sites so modelled data, based on the position of the site, from the UK Met Office unified model is used. UV-A irradiance data is taken from measurements made by the Public Health England (PHE) solar network.

170
To quantify the impact of the COVID-19 lockdown on atmospheric levels of NO2 and O3, we compare measurements during the lockdown with values corresponding to 'business as usual' (BAU), i.e. what we would have expected in the absence of the pandemic. To determine our BAU scenario, we first linearly detrend and deseasonalise NO2 data at each AURN site. To deseasonalise the data, we determine the climatology based on the mean annual cycle of the previous five years (from January 1st 2015 to December 31st 2019) which is then 175 repeated to match the length of the time series, subtracted from the mean to standardise the data, and then subtracted from the original time series to produce a time series of the residuals. This five-year period is sufficiently long to take into account year to year variations in meteorology but short enough to reduce the impact of any longer-term trends driven by earlier changes in emission standards. We then calculated the difference between a linear regression model of the previous data, projected forward to June 2020 to predict BAU values of 180 NO2 and O3 ( Figure 2) and calculated the difference between this and the measured values. We acknowledge there are uncertainties associated with our approach, but this method offers simplicity and straightforward error propagation. Other more complex methods to determine BAU that, for example, explicitly take into account local changes in meteorology (Grange and Carslaw, 2019) will also be subject to uncertainties, e.g. the extent which regional-scale meteorological fields can describe smaller-scale variations in atmospheric pollutants. We define 185 the start of the UK lockdown period as the 23th March 2020 when the lockdown was advised by the UK government. Figure 1 shows that a decrease in mobility in the transport sector is already evident from the 9 th March, which in the absence of any obvious change in law is perhaps influenced by the emerging crises in nearby European countries. Our analysis concludes on 31 st May 2020 the day before the first phase of lockdown easing in England.

190
We use independent sample Mann-Whitney U-tests to test the significance of changes in mean concentration for each site between the lockdown period and the mean of same period for the past five years, the lockdown period and measurements in 2020 prior to the lockdown and measurements from prior to lockdown in 2020 with the same period for the previous five years. This test indicates how likely the observed changes in mean concentration 195 between the different time periods are due to chance and noise in the data or whether they are statistically significant and be attributed to a real signal, which in our work is the start of the lockdown. We use this test rather than a t-test or z-test due to the large sample size and non-normal distribution of the data.

Meteorological Context
It is well understood that ambient concentrations of air pollutants are greatly affected by meteorology, with low wind speeds causing a build-up of pollutants and over the UK easterly flow is often accompanied by pollution from mainland Europe. Figure  was the fifth wettest on record and the fifth warmest. February 2020 was the wettest ever February recorded in 220 the UK. The wind roses also show that 2020 saw much stronger winds than the mean of the previous five years.
The six cities saw an average wind speed in 2020 of (6.5±1.2) ms -1 , which was 33.5% higher than the average of the previous five years. Since the beginning of the COVID-19 lockdown, meteorological conditions have been much more settled, with high pressure and easterly winds dominating UK weather since mid-March, especially in southern and western UK. Average wind speeds across the six cities was (4.1±0.4) ms -1 , although this is still an 225 increase of 7.5% compared to the previous 5 years. Of the cities analysed, Cardiff saw the largest increase in wind speed for 2020 compared to the previous five years (16.8%), with Bristol showing a 10% increase. The other cities all saw slight (<5%) decreases in wind speed in 2020. Typically, lower wind speed meteorological conditions are associated with higher levels of air pollution due to increased atmospheric stability and transport of pollution from mainland Europe in the UK, respectively, and so care must be taken when comparing pre and post-lockdown 230 levels of air pollution, as described in section 2.3, and comparing to the average of the previous 5 years is a better measure of the changes.

Observed changes in daily mean and diurnal variations of NO2
Measurements in 2020 from 65 urban traffic (figure S2a) and 61 urban background (

255
%. Urban background sites show a smaller mean reduction of (4.9±1.1) μg m -3 , equating to a decrease of (40.6±10.1) % in NO2 levels across the UK. The largest decrease of (25.7±3.3) μg m -3 was observed in London Hillingdon, corresponding to (59.3±9.6) %. Small increases (< 3 μg m -3 (< 10%)) were seen in York Bootham (Yorkshire and Humberside), and Eastbourne (South East). On average across all urban sites (traffic and background), a decrease in NO2 of (42±9.8) % is observed. We see that, whilst NO2 concentrations do tend to be 260 higher at lower wind speed ( Figure S3(b), there is very little correlation between the observed change in NO2 between 2020 and the previous five years and any change in wind speed ( Figure S3(a)).
We perform independent Mann-Whitney U-tests on NO2 measurements during the lockdown period and the mean from the same period from the previous five years, for NO2 measurements during the lockdown period and 265 measurements in 2020 immediately prior to the lockdown, and for NO2 measurements immediately prior to lockdown in 2020 with the same period for the previous five years. We find that using these tests that 115 out of the 128 (89%) urban sites show a statistically significant (p < 0.01) difference in NO2 between the mean observations during lockdown to the mean of the same period during the past five years. We also find 112 sites (from a possible 128 sites, 88%) show a significant difference between NO2 measurements made in 2020 immediately prior to the lockdown to the mean of the same period from the previous five years, with urban background sites showing a -6.3±1.5 μg m -3 change and urban traffic 9.2±1.9 μg m -3 ) We attribute this difference to changes in meteorology during January and February 2020 (Figure 3), in particular wind speed, which was on average 33% higher than the average of for the previous 5 years. . Finally, we find that 94 sites (75%) show a statistically significant difference between mean NO2 observations during lockdown and immediately prior to lockdown in 2020, implying there was a significant drop in NO2 across the UK as a direct result of the lockdown.
We also examine mean changes in the diurnal cycles of NO2 at urban traffic and urban background sites in London, Bristol, Cardiff, Newcastle, Glasgow and Belfast during the lockdown period compared to the same periods from the 2015-2019 mean. Based on the diurnal profile of NO2 levels, we find that ( Figure S4) typical pre-lockdown diurnal cycles are driven by emission peaks in the morning and evening rush hours, with the evening peaks supressed due to the higher mean boundary layer that is grown during the day. In general, we find that the evening rush hour peaks at urban traffic sites across all cities during the lockdown period are suppressed compared to previous years, potentially due to changing working patterns. A notable exception is in Cardiff where the morning rush hour peak is suppressed. In contrast, at urban background sites in Cardiff the diurnal cycle of NO2 is very 285 similar in 2020 to the previous years, with rush hour peaks of similar magnitude in the morning and evening. A reason for these observed diurnal cycles could be domestic combustion, which typically makes up around 17% of urban NOx emissions (compared to 47% for road transport and 15% for other transport e.g. rail). We do not expect domestic combustion to have changed much during the lockdown, therefore its contribution to the total (and the diurnal cycle) will be greater.

Observed changes in daily mean O3 and Ox (= O3 + NO2)
Typically, close to sources of NOx, O3 is suppressed due to the reaction of high levels of NO with O3. Further away from the sources, O3 can reform through the oxidation of NO to NO2 with peroxy radicals (formed from the reaction of VOCs with OH) and subsequent photolysis of NO2 to form O3. To account for this photochemistry,

295
we also report changes in the total oxidant, OX, the sum of O3 and NO2, which should be approximately conserved in the absence of any change in the source strength of NOx or VOCs or a change in OH.  in 2020 compared to the same period averaged across 2015-2019. After we remove site-specific trends, as described in section 2.3, we find that O3 at urban background sites increased by a mean value of (7.2±2.6) μg m -3 during the lockdown period when compared with the previous five years equating to a percentage increase of 305 (11±3.2) %. Leamington Spa (West Midlands) and London Hillingdon observed the largest mean increases of (21.3±2.7) μg m -3 )(35±3.6) %) and (21.6±3.6) μg m -3 ((54±7.5) %) respectively. Three sites observed a decrease during the lockdown with Aberdeen seeing a large decrease in O3 of (24.0±4.5) μg m -3 ((36±8.6) %) even though this site also experienced a substantial decrease in NO2. We do not have a definitive explanation for this result, but it is consistent with a NOx-limited photochemical environment in which a decrease in NO2 would reduce O3 310 production. This could be achieved by possible fugitive emissions from the onshore gas terminals near Aberdeen, although we have no VOC measurements to confirm this so the hypothesis is entirely speculative. Only three urban traffic sites measured O3 during our study period. Of those Marylebone Road (London) saw the largest increase ((32.0±4.8) μg m -3 or (104±10.1) %) followed by Exeter Roadside (South West) with an increase (20.0±2.8) μg m -3 ((47±5.5) %) and Birmingham A4540 Roadside (West Midlands) with an increase of (13.3±2.5) 315 μg m -3 ((25±3.9) %).
A similar statistical analysis has also been carried out for daytime (10:00 -18:00 UTC) Ox (NO2 + O3) and we find that a mean increase of Ox at urban background sites of (3. Following our approach for NO2, we use independent the Mann-Whitney U-test to determine the significance of changes in O3 pre-lockdown and lockdown periods in 2020 and in the previous five years. We find that 36 out of 46 urban sites (78%) show a statistically significant (p < .01) difference between the mean O3 observations during lockdown to the mean of the same period from 2015 to 2019. However, we also find that 41 of those sites (83%) show a statistically significant difference between O3 measurements immediately prior to the lockdown compared to the mean of the same period from 2015 to 2019. Finally, we find that 40 sites (95%) show a statistically 330 significant difference between O3 observations during the lockdown period and values taken from the period immediately prior to the lockdown.

Relationship to emissions
During the lockdown period there has been around a 75% reduction in road traffic across the UK, (using Google 335 activity data as a proxy for traffic) (see Figure 1). According to the NAEI, road transport is estimated to make up 53% of NOx emissions in the 1km x 1km grid square that both urban background and urban traffic sites are situated in (Defra, 2018b). Therefore, we might expect there to be a reduction in NO2 of around 40% across all sites. Mean decreases in NOx are very similar to those for NO2 described above (47% at urban traffic, 40% at urban backgroundsee figure S6), which is in line with the 40% reduction figure. However, it is clear that individual 340 sites have very different behaviour and the 75% traffic reduction may not necessarily equate to 75% reduction in emissions because different types of vehicle were affected differently, with the most reduction in passenger cars and less reduction in high emitters like HGVs. There is a wide range of contributions of NOx emissions from road traffic across the sites and there does not appear to be much correlation between this and the reduction seen during lockdown (see Figure S7), suggesting that the change in traffic flow near to individual sites is variable and will 345 be to largest contributing factor to NO2 reductions.

Surface O3
The COVID-19 lockdown has resulted in a significant decrease in NO2 in cities across the UK, largely caused by 350 the reduction of NOx emissions due to reduced traffic, and a concurrent increase in O3. NOx and O3 are closely linked through their photochemistry and here we examine the reasons for the O3 increase (Lelieveld and Dentener, 2000). Figure 6 shows the relationship between NO2 and O3 during the lockdown period, across all the AURN sites we examined. There is a clear anti-correlation between median NO2 and median O3 for all data, with data from 2020 tending towards lower NO2 and high O3 (Figure 6(b)). We also see that there is a correlation between 355 the change in NO2 and the change in O3 between 2020 and the previous five years (Figure 6(a)). Another key factor that plays a role in O3 concentrations is meteorology (Monks, 2000). High levels of actinic radiation cause the photochemistry involved in O3 formation to happen faster and low wind speed conditions allow precursor species such as NOx and VOCs to build up and react to form O3. Therefore, observed variations of O3 in different UK cities will be influenced by a number of processes to varying degrees. Figure 7 examines NO2, O3 and total daytime (10:00 -18:00) Ox during the lockdown period for urban background sites in six different cities across the UK.
Any change in Ox can be thought of as a change in the abundance of oxidants, taking into account the repartitioning of NO2 and O3 caused by changes in NOx emissions. Whilst all cities have seen an increase in O3 in the urban background compared to previous years, only southern UK cities saw a significant increase in total Ox, with London, Bristol and Cardiff showing increases of 5.1±0.3%, 5.8±0.6% and 5.6±1.2% respectively. In contrast, Ox 365 slightly decreased in Newcastle (-3.2±0.3%), Glasgow (-2.8±0.2%) and Belfast (-1.4±0.2%). To assess if OH is the cause of changes in Ox, we examine six measurements of total UVA at eight sites in the UK and compared data from 2020 to the mean of the previous five years ( Figure S8). We find levels of UV across the UK were higher in 2020 compared to previous years, with the largest increases in southern UK. London, Chilton and Camborne saw increases of around 50% compared to previous years, with Glasgow and Inverness showing smaller 370 increases of around 30%. Figure 8 shows a summary of the Ox change in 2020 compared to 2015-2019 from individual sites across the UK as a function of latitude. We find a positive trend in Ox towards lower latitudes, consistent with the higher excess UV levels further South. Therefore we conclude that in the cities in southern UK, some of the O3 increase is not solely attributable to reduced NOx, but also an increase in photochemistry related to the hot sunny weather experienced in 2020.

375
Observed variations in O3 may also be affected by changes in precursor VOCs. Online measurements of VOCs are only available at a small number of sites and here we consider measurements made at London Marylebone Road (an urban traffic site) and London Eltham (an urban background site). Figure S9 shows measurements of a range of different VOCs for each site during 2020 and mean values for 2017-2019 when data are available at both 380 sites. The data show most VOCs decrease in concentration during the post-lockdown period in 2020 compared to previous years. This is particularly true at the urban traffic site and for species such as benzene and toluene that have a largely traffic source and saw a decrease of 23±5.1% and 29±6.5%, respectively, compared to previous years. At Eltham the decreases were both around 12%. VOCs have a wide range of lifetimes and emissions sources and they can be transported large distances, meaning their concentrations at a given site is much more affected by 385 meteorology and chemistry than NO2. Indeed, in London according to the NAEI (in 2018), road transport only contributes 11% to sources of benzene, with other major sources being domestic / commercial combustion (69%), other transport (10%) and offshore oil and gas production (6%) (Vaughan et al., 2016). Therefore it is not surprising that VOCs show less of a reduction during the lockdown than NO2. When examining O3 it can be useful to look at the total VOC loading and OH reactivity (k'). Figure S10 shows total VOC loading in ppb and total OH 390 reactivity for each day in 2020, with colours showing the percentage change from the previous three years for that day. A full analysis of the behaviour of different VOCs during the lockdown period is beyond the scope of this work, and it is unlikely that the measurements made at the AURN sites cover all VOCs that contribute to OH reactivity (e.g. few oxygenated compounds or larger VOCs are measured). Our focused analysis shows that while the picture is not straightforward, there is an apparent decrease in total VOCs at both sites compared to previous 395 years. Mean values for total VOCs at Marylebone Road were 17% lower and the corresponding k' 15% lower than the 2017-2019 mean. At Eltham total VOCs saw a decrease of 10%, with a slight increase in the total k', largely driven by an increase in biogenically emitted isoprene. Figure 9 shows daytime mean (10:00 -18:00) isoprene data and its contribution to k' at two sites in 2019 and 400 2020 (the only years where reliable isoprene data is available). Observed isoprene was a factor of two higher at both Marylebone Road and London Eltham during April and May 2020 compared to those months in previous years. Isoprene represents only a small contributor to OH reactivity at Marylebone Road, but at Eltham in 2020 it represents around 25% of total k'. Biogenic emissions of isoprene, originating from a variety of trees and shrubs, are driven in part by temperature and so it is perhaps not surprising that isoprene levels at the London sites were 405 higher in 2020 compared to 2019 due to the fact that temperature was approximately 2 o C higher in 2020 compared to 2019 for the lockdown period.. Further detailed chemical modelling studies, beyond the scope of this study, are required to assess in detail the chemistry behind O3 formation and how this has been affected by the lockdown, however we observe that O3 has increased across the UK and see a clear anti-correlation with a decrease in NO2 across the sites. We also see an increase in total Ox at Urban Background sites in the South of England, likely due 410 to increased radiation and biogenically emitted VOCs compared to previous years, things that are unlikely to be linked to the COVID-19 lockdown.

Exceedances.
To put the changes in air pollutants in context with human health effects we have examined the number of 415 exceedances of UK air quality objectives (Defra, 2019) and EU directive limits (EEA, 2016) for both NO2 and O3 in 2020 compared to previous years (see Table 2). For this analysis, we have used data from the London Air Quality Network (LAQN) consisting of 9 kerbside, 31 roadside, and 14 background sites for NO2 and 1 kerbside, 7 roadside and 5 urban background sites for O3 in the Greater London area. London has historically had by far the largest number of air quality exceedances in the UK so this analysis allows us to see the effect of the lockdown 420 on the city with the most acute air pollution problems.
Exceedances were calculated on a per site basis, and then summed across all sites of a given type. For NO2 a simple one-hour mean was calculated and each hour greater than 200 ug m -3 was counted as an exceedance. We calculated a rolling mean value for O3, using a window of eight hours and a step size of one hour. If a given 425 calendar day saw this rolling mean exceed 100 ug m -3 , an exceedance was counted. Using this method multiple exceedances (contiguous or separated in time) were only counted as one to avoid ambiguity in their definition, and therefore can be thought of as "days when an O3 exceedance occurred". for the year or how many exceedances will breach the 18 allowed by the air quality objective. Consequently, further analysis on data collected for the whole of 2020, including the period when lockdown restrictions were relaxed, is required to put 2020 into context of previous year. As an estimate of the effect the lockdown may have on total exceedances in 2020, we replaced the number of exceedances during the lockdown period in 2019 with the number from 2020. This resulted in a 47 % decrease (34 to 18) in total exceedances of NO2 at kerbside sites and a 12 % (76 to 67) at roadside sites. As the effects of lockdown certainly extend beyond the end of the time period explored by this study, we would expect there to be less exceedances still during the remainder of 2020.
When considering any health benefits to this apparent improvement in air quality due to reduced NO2 we should also consider exceedances to O3 limits. The WHO has set a guideline value for ozone levels at 100 µg m -3 for an 8-hour daily mean. Figure  it will be necessary to analyse data from the whole year, alongside NO2, in order to fully assess the importance but it is clear that any perceived benefits of reduced NO2, both during the lockdown and in a lower NOx future, should be considered alongside any concurrent increase in O3.

5 Summary and conclusions
We examined NO2 and O3 measurements from urban traffic and urban background sites across the UK during the in 2020 compared to previous years; we find no evidence to suggest that higher UV levels were due to cleaner 470 skies related to air pollution changes due to the lockdown. Analysis of exceedances of air quality objectives in London for NO2 and O3 show that whilst there has been a decrease in exceedances of the NO2 objective, this has come alongside an increase in O3 exceedances. If we are to take the COVID-19 lockdown as an analogue of how air quality will respond to future reductions in emissions from vehicles (e.g. over the next 10-20 years), then observations show that there could be a corresponding increase in O3 which should be considered in any air quality 475 abatement strategy.. In China, NOx reductions have led to increases in O3 (Li et al., 2019a;Ma et al., 2016;Sun et al., 2016) and therefore air quality abatement strategies are being developed in order to offset this, largely by also controlling VOCs (Li et al., 2019b;Le et al., 2020). These changes are attributable to photochemical processes (e.g. the reduction in particles causing increased radiation and photochemistry), however our study shows that a large reduction in NOx, directly causes an increase in O3 due to a reduction in titration with NO. In addition, a warming climate may lead to increased emissions of biogenic VOCs, further adding to the O3 burden. Therefore it will be vital to control anthropogenic VOCs in the UK to avoid any health gains made by the reduction of NO2 being offset by O3 increases.

485
The AURN data is all available for public download from the UK-AIR website (uk-air.defra.gov.uk). The LAQN data is available from the LondonAir website (londonair.org.uk). UVA data is available on request from Public Health England.            (Defra, 2019). Note that the UK has adopted the EU NO2 limit as a part of its air quality objectives, but improves upon the O3 obligations where O3 must not exceed 120 ug m -3 more than 25 times per year in a given 3 year window (EEA, 2016).