Widespread changes in UK air quality observed from space

Previous studies have used surface observations of pollutants (e.g., nitrogen dioxide [NO2] and aerosols) to evaluate improvements in United Kingdom (UK) air quality over recent decades. However, surface monitoring networks provide limited spatial coverage and are not always representative of air quality over a wider region. Satellite observations, such as tropospheric column NO2 (TCNO2), aerosol optical depth (AOD) and sub‐column (0–6 km) ozone (SCO3), provide widespread monitoring of air quality on a national scale. In this study, we use such observations to analyse trends in UK air quality, between 2005 and 2015, finding significant decreases in TCNO2 and AOD over UK pollution hotspots. The largest changes in short‐lived NO2 are located over populated (i.e., source) regions, associated with large emissions of nitrogen oxides (NO x). AOD changes are more spread out, and less strongly associated with source regions, due to a longer aerosol lifetime, secondary aerosol formation and non‐urban aerosol sources. SCO3 shows no significant trends over England/Wales, but significant positive trends over Scotland, which is consistent with previous studies of changes in background tropospheric ozone in other remote regions of north‐western Europe.

Previous studies have used surface observations of pollutants (e.g., nitrogen dioxide [NO 2 ] and aerosols) to evaluate improvements in United Kingdom (UK) air quality over recent decades. However, surface monitoring networks provide limited spatial coverage and are not always representative of air quality over a wider region. Satellite observations, such as tropospheric column NO 2 (TCNO 2 ), aerosol optical depth (AOD) and sub-column (0-6 km) ozone (SCO 3 ), provide widespread monitoring of air quality on a national scale. In this study, we use such observations to analyse trends in UK air quality, between 2005 and 2015, finding significant decreases in TCNO 2 and AOD over UK pollution hotspots. The largest changes in short-lived NO 2 are located over populated (i.e., source) regions, associated with large emissions of nitrogen oxides (NO x ). AOD changes are more spread out, and less strongly associated with source regions, due to a longer aerosol lifetime, secondary aerosol formation and non-urban aerosol sources. SCO 3 shows no significant trends over England/Wales, but significant positive trends over Scotland, which is consistent with previous studies of changes in background tropospheric ozone in other remote regions of north-western Europe. Degradation in air quality in the United Kingdom (UK) is an important public health concern, estimated to result in approximately 40,000 premature deaths per year (Holgate, 2017) and an average reduction in life expectancy of 7-8 months (House of Commons, 2010). Exposure to pollutants such as ozone (O 3 ), nitrogen dioxide (NO 2 ) and particulate matter (PM 2.5 & 10 -atmospheric particles with diameters less than 2.5 and 10 μm, respectively) leads to health ailments such as lung disease and cancer, cardiovascular problems, asthma and eye irritation (WHO, 2014). In light of these concerns, government policy has led to the monitoring of many pollutants (e.g., Automated Urban and Rural Network [AURN], Department for Environment, Food and Rural Affairs [DEFRA], 2015) and efforts to reduce their emissions (e.g., the Environment Act 1995, the National Emission Ceilings Regulations 2002 and the Air Quality Standards Regulations 2010; DEFRA, 2011). Multiple studies have highlighted the impact of emission control legislation on abundances of pollutants, such as nitrogen oxides (NO x ) and aerosol, over the United Kingdom (e.g., Carslaw et al., 2011;Turnock et al., 2015). Trends between 2000 and 2014 in UK summer-time surface concentrations of the secondary pollutant tropospheric O 3 , show significant variability but an overall increase (Chang et al., 2017). Munir et al. (2013) found positive trends in mean surface O 3 at urban and rural AURN sites, 1993-2011, but a decrease in peak O 3 concentrations. Since surface monitoring sites are point measurements, they are often unrepresentative of pollution over a wider region given inhomogeneities in air pollutant concentrations that are produced by strong gradients in surface emissions and complex meteorology. Satellite observations, given their wide spatial coverage, may offer an opportunity to assess changes in UK air quality on a national scale.
Only a few studies have used satellite data to evaluate UK or north-western Europe air quality. For instance, van der A et al. (2008) and Zhou et al. (2012) used satellite tropospheric column NO 2 (TCNO 2 ) to identify negative trends in NO 2 over the United Kingdom/Europe between 1996-2005and 2004-2009, respectively. Pozzer et al. (2015 detected large negative trends in aerosol optical depth (AOD) over Europe (including United Kingdom), between 2001 and 2010, from the Moderate Resolution Imaging Spectroradiometer (MODIS), but this was much weaker for other instruments. Alpert et al. (2012) also found negative AOD trends over London, between 2002 and 2010, from MODIS-Terra, but MODIS-Aqua and the Multi-angle Imaging Spectroradiometer (MISR) observed positive trends. This study utilises satellite observations of TCNO 2 and AOD to investigate more recent changes in UK pollution, building upon the previous studies mentioned above. However, by using higher resolution satellite data, we are able to investigate UK pollution hotspots, such as Manchester, Birmingham and the Yorkshire power stations, which have not been investigated before in significant detail. We also utilise satellite measurements of subcolumn (0-6 km) O 3 (SCO 3 ) for the first detailed look at tropospheric O 3 trends over the United Kingdom from space. Section 2 discusses the satellite data and gridding methodology used to obtain satellite data at a high resolution, section 3 presents our results and section 4 highlights our conclusions.
2 | SATELLITE DATA AND HIGH RESOLUTION GRIDDING ALGORITHM 2.1 | Satellite data TCNO 2 and SCO 3 data (Miles et al., 2015) come from the Ozone Monitoring Instrument (OMI), on-board NASA's EOS-Aura satellite, which are provided by the Tropospheric Emissions Monitoring Internet Service and the Rutherford Appleton Laboratory (RAL), respectively. AOD data is from MODIS, on-board NASA's EOS-Aqua satellite, which has been obtained directly from NASA. Both Aura and Aqua are polar orbiting with approximate equator crossing times of 13:30 local time (LT). OMI and MODIS have nadir-viewing spectral ranges of 270-500 nm (Boersma et al., 2007) and 0.41-15 μm (Remer et al., 2005), respectively, with pixel sizes in the order of~10−100 km ×~10 −100 km depending on viewing angle. All data sets have been quality controlled for geometric cloud fraction less than 0.2, good quality flags and the OMI row anomalies (Braak, 2010) where applicable.

| Gridding algorithm
Several studies have applied complex methods to map air quality satellite data onto high resolution regular grids (e.g., 0.05 × 0.05 ). This provides much more spatial detail than the resolution of 0.25 × 0.25 used in many studies such as Pope et al. (2014) and Zhou et al. (2012). Russell et al. (2011) use higher resolution inputs than the standard TCNO 2 products to generate the air mass factors (AMFs) in the retrieval process for the Berkeley High Resolution (BEHR) TCNO 2 product. Barkley et al. (2017) used the area-weighting tessellation algorithm of Spurr (2003) to investigate air quality over the Middle East. In this study we use a simpler methodology of slicing up satellite pixels and mapping them onto a higher resolution grid, which is less computationally expensive, while maintaining a robust data product. Often, only the central position of the retrieval is used to map satellite data onto a regular grid. This can result in missing information if the retrieval pixel overlaps several grid boxes on a higher resolution regular grid (e.g., 0.05 × 0.05 instead of 0.25 × 0.25 ). By using the central and corner positions of the satellite pixel, this study slices the pixel into many sub-pixels, which are mapped onto the higher resolution grid retaining much of the information and yielding more detailed spatial patterns. This is shown in Figure 1 for TCNO 2 (a,d), SCO 3 (b,e) and AOD (c,f ), where TCNO 2 and AOD are mapped onto a 0.05 × 0.05 grid and SCO 3 is mapped onto 0.5 × 0.5 grid. Even though TCNO 2 and SCO 3 are both retrieved from OMI with the same horizontal resolution, multiple swath pixels have to be merged together in the RAL O 3 retrieval process to get a robust tropospheric signal-to-noise ratio due to stratospheric O 3 interference. Therefore, the horizontally coarser OMI SCO 3 data is mapped onto a lower resolution grid. This gridding technique is discussed in more detail in File S1, Supporting information.

| RESULTS
We examined the 2005-2015 TCNO 2 , SCO 3 and AOD trends using the methodology of van der A et al. (2006,2008). Here the monthly species' time-series is represented by the function: where Y t is the monthly species column for month t, X t is the number of months after January 2005, C is the January 2005 species concentration, B is the monthly linear trend and Asin(ωX t + ϕ) is the seasonal model component (Weatherhead et al., 1998). A is the amplitude, ω is the frequency (set to 1 year; ω = π/6) and ϕ is the phase shift. C, B, A and ϕ are the fit parameters based on a nonlinear least squares fit. N t represents the model errors/residuals. The linear trend, B, is determined to be significant at the 95% confidence level if |B/σ B | > 2.0 (Weatherhead et al., 1998). σ B represents the precision of the trend. This is calculated by: where n is the number of years, α is the autocorrelation in the residuals (N t ) and σ N is the standard deviation in the residuals. For TCNO 2 and AOD four hotspot regions were investigated defined as London, Manchester, Drax (Yorkshire/Humberside for AOD) and Birmingham from the pollutant spatial distributions in Figure 1d,f, respectively. London, Manchester and Birmingham represent large populous urban regions, while Drax is a large power station in Yorkshire. Pope and Provod (2016) identified Drax, along with the nearby Eggborough and Ferrybridge C power stations, as the source of the North-East Yorkshire TCNO 2 hotspot. In each of the regions, we see significant negative linear trends (blue lines, Figure 2) over the 11-year period in TCNO 2 ; London (−0.23 AE 0.05 × 10 15 molecules cm −2 year −1 ), Manchester year −1 ), Drax (−0.22 AE 0.10 × 10 15 molecules cm −2 year −1 ) and Birmingham (−0.11 AE 0.05 × 10 15 molecules cm −2 year −1 ). We calculate insignificant trends in the stratospheric slant columns and tropospheric AMFs, both used to derive the tropospheric column (Boersma et al., 2004), giving confidence to our interpretation of the negative TCNO 2 trends found here. Similar negative trends were seen in the AOD ( Figure 3); London (−0.0061 AE 0.0026/year), Manchester (−0.0057 AE 0.0017/year), Yorkshire/Humberside (−0.0050 AE 0.0019/ year) and Birmingham (−0.0052 AE 0.0024/year). However, it should be noted that the AOD time-series over Manchester has a large proportion of missing data in the winter-time. The specific longitude-latitude grid boxes (shown in Figure 1e) used to define these locations are described in the captions of Figures 2 and 3. Some of the TCNO 2 and AOD boxes (Figure 1e) are slightly larger than the hotspot domain to obtain sufficient data (grid boxes) to identify significant (non-significant) trends in the TCNO 2 (stratospheric slant column NO 2 and AMFs). Maps showing the spatial distribution of all grid box trends for NO 2 and AOD are shown in File S1.
Our London TCNO 2 trend is nearly four times larger than that found by van der A et al.   (DEFRA, 2011), and the United Kingdom has received a final warning to address unsafe levels of NO 2 or face legal action from the European Commission (2017). Though UK NO x emissions have decreased, longrange transport of pollution from continental Europe, as shown by Pope et al. (2014), will contribute significantly to the total NO 2 concentrations, especially in urban regions. The NAEI NO x emissions also do not account for the under-representation of diesel car sources (Jonson et al., 2017), which potentially weakens the reported decreasing NAEI UK NO x emissions trend. NO 2 has a relatively short lifetime compared with aerosol (Seinfeld and Pandis, 2006), so changes in NO 2 are more concentrated over populated regions (anthropogenic emission sources). Primary aerosol, due to its longer life-time (several weeks), is more susceptible to long-range transport from source regions yielding a more homogeneous spatial distribution than that of NO 2 . Therefore, changes in source region aerosol emissions will result in satellite observed AOD changes over a wider region with a lower average population density. This is highlighted in Figure 4 where the population fraction, subject to different changes in air pollution, is distributed against these trends represented by equally spaced bins during the 2005-2015 period. Year 2010 population data from the Socioeconomic Data and Applications Centre (SEDAC) is used. Ten equally spaced bins are used for both pollutants ranging between the trend's 0.5th and 99.5th percentiles. Approximately, 90% of the population have benefited from negative AOD trends (bottom panel) between −0.0113/year and −0.0013/year (i.e., distributed across four bins with peak population exposure of 32%). However, the TCNO 2 trends (top panel) are distributed more evenly with 15% peak  Figure 1f highlights a larger AOD hotspot over Yorkshire/ Humberside, potentially linked to the accumulation of pollution in the Vale of York, a low lying region flanked by the uplands of the Pennines on the west and the North York Moors to the north. A negative trend here may be explained by reduced in situ aerosol/precursor gas emissions and transport of cleaner air masses accumulating in the region.
As the spatial resolution of SCO 3 is coarse relative to the TCNO 2 and AOD, we focus on regional average trends (Figure 5a   surface O 3 trends in northern Europe, but were insignificant in summer. Ordóñez et al. (2007) associated increases in background tropospheric O 3 , using models and ozonesondes, with downwelling of lower stratospheric O 3 , especially in winter and spring. The England/Wales and Scotland December-January-February (DJF) trends support this where there is an insignificant and significant (95% confidence level) increase in SCO 3 , respectively. However, the spring (March-April-May, MAM) trends show insignificant negative trends (potentially linked to different conflicting processes and large meteorological inter-annual variability). The England/ Wales DJF trend is possibly insignificant as the region is more polluted (Figure 1d) with larger emissions of nitric oxide (NO), which can act as a sink of O 3 weakening the wintertime trend. Over Scotland, there is also an increasing June-July-August (JJA) trend (90% confidence level) which supports the results of Chang et al. (2017). Both of DJF and JJA seasons appear to be driving the positive annual SCO 3 trend over Scotland, despite the large SCO 3 inter-annual variability. The investigation of upper tropospheric O 3 trends (450-100 hPa) in the OMI data set highlighted increases in tropospheric O 3 over the whole UK, but these annual trends are not significant (not shown).

| CONCLUSIONS
By using high resolution data sets of satellite-observed air quality, this study has identified recent (2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015) widespread changes in UK tropospheric column NO 2 (TCNO 2 ), sub-column ozone (SCO 3 ) and aerosol optical depth (AOD). Over the UK pollution hotspots (e.g., London), there have been significant decreases in TCNO 2 and AOD resulting from . When pollution trends, both significant and non-significant, across the United Kingdom are weighted by population fraction exposed, it suggests that the general public are benefiting more from larger changes in NO 2 than aerosols. The lifetime of NO 2 is shorter than that of aerosol, so any changes in emissions (i.e., anthropogenic activities) are more strongly co-located with populated regions. Finally, trends in SCO 3 are significant over Scotland, a more remote and cleaner region of the United Kingdom, which are consistent with other studies showing increasing background tropospheric O 3 over north-western Europe. and September-October-November (SON, yellow) have been calculated for England/Wales and Scotland in the bottom two panels. * and ** represent significant trends at the 90% and 95% confidence levels. These trends are calculated using a linear version (i.e., no seasonal term) of Equation 1