How have surface NO₂ concentrations changed as a result of the UK's COVID-19 travel restrictions?

The hourly air quality data used in this study comes from the Automatic Urban and Rural Network (AURN). The AURN includes automatic air quality monitoring stations measuring a range of atmospheric pollutants which undergo routine quality control. These sites provide hourly information which is communicated to the public via the Department for Environment, Food and Rural Affairs (DEFRA) website (DEFRA 2020). In this study we use hourly NO₂ concentrations which are measured using chemiluminescence detectors. Currently there are 142 AURN sites measuring NO₂ with the majority of sites classified as urban traffic (70), urban background (65) or rural background (21) (figure 1). Urban sites are located in built-up areas containing buildings with at least 2 floors. As a result, urban sites typically measure air quality which is representative of only a few km². Rural sites are located more than 5 km from built-up areas, industrial installations or major roads. The air sampled by rural sites is therefore representative of air quality in a surrounding area of 1000 km² (AURN 2020). The pollution levels measured at traffic sites are determined predominantly by the emissions from nearby traffic (Carslaw and Beevers 2004) whereas the pollution levels measured at background sites are not influenced significantly by any single source or street, but rather by the integrated contribution from all sources upwind of the station (Donnelly et al 2015).

Zoom In Zoom Out Reset image size

The hourly meteorological data used to build the statistical models in this study comes from the UK Met Office UKV model for the period 01/01/15 - 30/04/20. The UKV is a variable resolution UK model which has a high resolution inner domain (1.5 km grid boxes) over the area of forecast interest, separated from a coarser grid (4 km) near the boundaries by a variable resolution transition zone (Tang et al 2013). There are 70 vertical levels on a terrain-following hybrid-height vertical co-ordinate with Charney-Philips staggering. Sub-grid scale processes such as, boundary layer turbulence, radiation, cloud, microphysics and orographic drag are represented by parameterizations. The UKV has an advantage in that it contains a better representation of orography and other land surface parameters than coarser resolution models. Hourly 4D-Var assimilation allows the model to be initialised every hour to produce state-of-the-art weather forecasts for the UK (Ballard et al 2016). Meteorological data for the 5-year period (01/01/15 - 31/12/19) is extracted from the UKV and interpolated to the location of the AURN sites to build the statistical models described in section 4.

The UK Met Office's Air Quality model (AQUM) is used in this study to simulate the regional transport of NO₂. AQUM is the operational air quality forecast model for the UK (Savage et al 2013). The model has a 12 km horizontal grid spacing covering much of Western Europe. There are 70 vertical levels up to a model top height of 40 km. The model's meterological fields are intialised with data from the Met Office global model. The chemistry fields are initialised with data based on the 24-hour forecast from the previous day (Davis et al 2014). Boundary conditions for meteorology and chemistry come from the Met Office global model forecast and Copernicus Atmosphere Monitoring Service global forecasts respectively (CAMS 2020). Emissions of NO₂ within the AQUM domain are generated by merging two data sets: the 2017 National Atmospheric Emissions Inventory (NAEI) (1 × 1 km) for the UK and 2017 European Monitoring and Evaluation Programme (EMEP) (0.1 × 0.1 degrees) for the rest of the model domain.

A quantitative evaluation of the ability of AQUM to capture the regional spatial distribution of NO₂ when sampled under the Lamb weather types, was performed by Pope et al (2015). Lamb weather types are an objective classification of midday UK circulation patterns (Lamb 1972). Pope et al (2015) found that AQUM could reproduce the large-scale accumulation of tropospheric column NO₂ over the UK under anticyclonic conditions and that transport is an important factor in governing the variability of UK air quality on synoptic timescales. AQUM is used in this study to qualitatively visualise the regional transport of NO₂, therefore emissions have not be altered to reflect post-lockdown emissions changes.

In order to predict hourly NO₂ concentrations during the ongoing COVID-19 travel restrictions multiple linear regression models are built using up to 5-years of AURN and UKV data (2015–2019) similar to the method of Shi and Harrison (1997). The models predict the hourly NO₂ concentrations we would expect, during the COVID-19 period, given no change in NO₂ emissions. Several meteorological and temporal explanatory variables are used to predict NO₂ concentrations at each AURN station.

The Akaike Information Criterion (AIC) is used to determine which explanatory variables to include in our model. The model with the lowest AIC score is expected to have the best balance between its ability to fit the data set and its ability to avoid over-fitting the data set. The explanatory variables used in this study are shown in table 1. Some explanatory variables are cyclic not linear, such as wind direction (i.e. 0 and 360 degrees have the same direction). To account for this, wind direction is partitioned into its northerly and easterly components. Similarly, several of the temporal variables are cyclic. For day of year, day of week and hour of day a cyclic relationship is assumed with fixed periods of 365 days, 7 days and 24 hours respectively.

Changes in NO₂ concentrations between the pre- and post-lockdown period may be due to changes in the primary emission of NO₂, changes in the concentration of chemical precursors which can transform to produce secondary NO₂ (i.e. NO and O₃), changes in the meteorology which affects the transport and mixing of NO₂ and its precursors, or a combination of all three. Differentiating changes in primary from secondary NO₂ is not possible using multiple linear regression modelling. Instead we focus on partitioning the differences attributed to changes in emissions (Δ NO₂ Emissions) from changes in the meteorology (Δ NO₂ Meteorology).

The statistical models described in section 4 are used to predict hourly NO₂ concentrations for the period 01/02/20–30/04/20. The model predictions assume no change in the emission of primary NO₂ or the emission of chemical precursors which transform to produce secondary NO₂. We partition the contribution to change in NO₂ between the post- and pre-lockdown period as follows:

• (a)  
  Δ NO₂ Total: The difference between mean measured post-lockdown NO₂ concentrations ($\mathrm{NO}_{2}^{post}$) and mean measured pre-lockdown concentrations ($\mathrm{NO}_{2}^{pre}$) are compared to the 5-year averaged NO₂ concentrations for the period 1 February - 30 April ($\overline{NO_{2}}$) at each AURN site.

  $$\begin{equation} \Delta \mathrm{NO}_{2}\;\mathrm{Total} = \frac{\mathrm{NO}_{2}^{post}-\mathrm{NO}_{2}^{pre}}{\overline{\mathrm{NO}_{2}}} \end{equation} \tag{ 1 }$$
• (b)  
  Δ NO₂ Meteorology: The difference between mean predicted post-lockdown NO₂ concentrations ($\widehat{\mathrm{NO}}_{2}^{post}$) and mean measured pre-lockdown concentrations are compared to 5-year averaged NO₂ concentrations. The percentage change is attributed to changes in the meteorology.

  $$\begin{equation} \Delta \mathrm{NO}_{2}\;\mathrm{Meteorology} = \frac{\widehat{\mathrm{NO}}_{2}^{post}-{\mathrm{NO}}_{2}^{pre}}{\overline{\mathrm{NO}_{2}}} \end{equation} \tag{ 2 }$$
• (c)  
  Δ NO₂ Emissions: The difference between mean predicted post-lockdown NO₂ concentrations and mean measured post-lockdown NO₂ concentrations are compared to the 5-year averaged NO₂ concentrations. The percentage change is attributed to the changes in the emission of primary NO₂ or the emission of chemical precursors which transform to produce secondary NO₂.

  $$\begin{equation} \Delta \mathrm{NO}_{2}\;\mathrm{Emissions } = \frac{\mathrm{NO}_{2}^{post}-\widehat{\mathrm{NO}}_{2}^{post}}{\overline{\mathrm{NO}_{2}}} \end{equation} \tag{ 3 }$$

The mean NO₂ concentrations at 142 AURN sites are analysed for the pre-lockdown (01/02/20 - 16/03/20, figure 3(a)) and post-lockdown (17/03/20 - 30/04/20, figure 3(b)) periods. It is clear that since lockdown, NO₂ concentrations have reduced over much of the UK. Figure 3(c) shows the percentage change between the post- and pre-lockdown periods (Δ NO₂ Total). In the north of England NO₂ concentrations have reduced by an average of 20%, but in the south of England and parts of Wales, several AURN sites show an increase in NO₂ concentrations after lockdown of over 20%.

Zoom In Zoom Out Reset image size

Figure 4 shows the time series of measured NO₂ concentrations at 6 sites in the UK. At all sites the NO₂ concentrations are below the 5-year average for much of the period studied. Figures 4(a) and (b) are for the London Marylebone Road and Glasgow Kerbside urban traffic AURN sites respectively. At London Marylebone Road the change in NO₂ concentrations before and after lockdown was -36$\mu\mathrm{gm}^{-3}$ and at Glasgow Kerbside NO₂ concentrations changed by −20$\mu\mathrm{gm}^{-3}$. This corresponds to a percentage change relative to the 5-year average of -42% and -30% respectively. Thus the anticipated decrease in NO₂ concentrations due to reduced emissions from traffic is clear at these AURN sites. However, at Birmingham Adcocks Green (figure 4(c)) and Exeter Roadside (figure 4(d)) there is no change in NO₂ concentrations after lockdown. Furthermore, at Chilbolton Observatory (figure 4(e)) and Eastbourne (figure 4(f)) there has been an increase in NO₂ concentrations after lockdown, corresponding to Δ NO₂ Total of +36% and +46% respectively.

Zoom In Zoom Out Reset image size

Figure 5 shows the measured NO₂ concentrations, with a 24-hour moving average applied (blue) and model predictions for the same period (red) for 4 AURN sites. At all sites the models capture the NO₂ variability in the pre-lockdown period well (R² > 0.5). In the post-lockdown period, the measured NO₂ concentrations are generally lower than the model predictions. Since the model predicts the NO₂ concentrations expected given no change in NO₂ emissions, this suggests that emissions of NO₂ have reduced at all 4 sites. This is particularly evident at London Marylebone Road (figure 5(a)) and Glasgow Kerbside (figure 5(b)) where NO₂ emissions reduce by 34% and 49% respectively. At Birmingham Adcocks Green (figure 5(c)) the mean concentration did not change significantly from the pre- and post-lockdown period ($\lt 1\mu\mathrm{gm}^{-3}$). However, given no change in emissions, changes in the meteorology would have resulted in a change to NO₂ concentrations of +18%. Changes in the emissions caused a change in NO₂ concentration of −20%, therefore resulting in a negligible total change. Thus, despite the fact that there was little change in total NO₂ concentrations, emissions of primary NO₂ and secondary NO₂ precursors were still found to reduce as a result of the COVID-19 restrictions. Similarly, at Exeter Roadside (figure 5(d)) there is no change in the mean concentration between the pre-lockdown period and the post-lockdown period. However, changes in the meteorology would have resulted in an increase in NO₂ concentrations of +33%. This was offset by a reduction in emissions of −33%, resulting in no total change in NO₂ concentrations. Performing similar analysis at all 142 AURN sites in England shows that, given no change in emissions, the changes in meteorology between the pre- and post-lockdown period would have led to a mean increase in NO₂ concentrations of +6% (table 2). Conversely, changes in emissions would have led to a mean reduction in NO₂ concentrations of −18%, resulting in a mean total change of −12% (table 2).

Zoom In Zoom Out Reset image size

In this section the spatial variability in the total NO₂ concentration changes shown in figure 3(c) is discussed. Firstly, the north/south pattern in total NO₂ concentration changes is investigated. Changes in the meteorology are found to be responsible for larger increases in NO₂ concentrations in the south of England and Wales than in central and northern parts of the UK (figure 6(a)). Changes in emissions are however more uniform over the UK (figure 6(b)). The meridional gradient in meteorological attributed changes in NO₂ is likely to be due to the advection of secondary NO₂ over the south of the UK by the predominantly south-easterly winds. This secondary NO₂ is produced over the English Channel and BeNeLux regions and transported towards the UK during periods dominated by high-pressure and easterly winds. Two examples of anticyclonic NO₂ transport during the post-lockdown period are shown in figures 7(a) and (b) with plumes of NO₂ being transported over the UK, particularly the south and east coast of England. According to Lamb weather type, between 17/03/20 and 30/04/20, 48% of the days were classed as anticyclonic, resulting in frequent regional transport of NO₂ from Europe to the UK in the post-lockdown period.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Secondly, within the north/south pattern in total NO₂ concentration changes there is some spatial variability. The largest increases associated with changes in the meteorology are seen at rural sites (20 ± 2%, table 2) at which the NO₂ measurements are more representative of large areas (1000 km²) and thus NO₂ concentrations are dominated by the regional advection of secondary NO₂ (Harrison and McCartney 1980). Conversely, mean changes in NO₂ concentrations at urban traffic and urban background sites (−27% and −14% respectively, table 2) are representative of more local areas (few km²) and are more influenced by local reductions in emissions of primary NO₂ and secondary NO₂ precursors (i.e. NO). The largest reduction in emissions are seen at urban traffic sites such as Bristol Temple Way (−59%), Oxford Centre (−45%) and Dumbarton (−43%). A few of the sites ($\lt 10\%$) have small increases in NO₂ emissions, however these are largely background sites where the absolute change in NO₂ concentrations is small ($\lt 5\mu\mathrm{gm}^{-3}$).