Changing supersites: assessing the impact of the southern UK EMEP supersite relocation on measured atmospheric composition

In January 2016 the United Kingdom’s southern European Monitoring and Evaluation Programme (EMEP) level-2 air pollution monitoring ‘supersite’ was relocated from Harwell, Oxfordshire to Chilbolton Observatory, Hampshire. As no co-location study was undertaken, this work retrospectively investigates whether the supersite relocation has led to discontinuities in the time series of concentrations of commonly studied gaseous pollutants (NOx, NH3, SO2 and O3) and particulate matter (PM2.5 and PM10). Two years of measurements pre- and post-relocation (2014–15 and 2016–17 respectively) were analysed in conjunction with meteorological variables and local emission data. The deweather package was applied to the concatenated time series to minimise the influence of meteorology. Similar average concentrations of PM2.5, PM10, SO2 and O3 were observed, but there were substantial differences in that of NOx and NH3 (increase by factors of ∼1.6 and ∼3, respectively). The considerably higher NH3 concentrations at Chilbolton are attributed to the close proximity of mixed farmland, in particular to a strong south-westerly source contributing to ∼50% of the annual average. NOx and PM concentrations in easterly winds arriving at Chilbolton are ∼2.7 and ∼1.5 times larger than at Harwell, from sources including the M3 motorway and Greater London. Westerly concentrations of NOx remain similar, therefore despite a higher frequency of westerly wind, annual mean concentrations are larger. Lower concentrations of PM arriving from the west result in similar annual averages. The secondary inorganic and black carbon components of PM were broadly similar between the sites. The differences in average NOx and NH3 at Chilbolton must be taken into account when considering long-term regional trends based on the southern UK supersite data.


Introduction
Atmospheric pollution has a significant influence on human and ecosystem health. Inhalation of ozone (O 3 ) and particulate matter has been linked to cardiovascular and respiratory diseases (WHO 2006(WHO , 2013. Deposition of acidic gases causes acidification of terrestrial and aquatic ecosystems, and nitrogen deposition (e.g. from nitrogen oxides, NO x , or ammonia, NH 3 ) leads to eutrophication (Sutton et al 2011, RoTAP 2012. Monitoring of air pollutant concentrations is important for quantifying these effects and their spatiotemporal trends (Fagerli andAas 2008, Malley et al 2016). In Europe, the Cooperative Programme for Monitoring and Evaluation of the Long-range Transmission of Air Pollutants in Europe (EMEP, www.emep.int) has the aim of providing member states with this quantitative information (Tørseth et al 2012). Measurements within EMEP are made at rural sites representative of the surrounding area (Spangl et al 2007, Joly andPeuch 2012) and adhere to prescribed sampling methods and siting criteria detailed by the Chemical The Chilbolton Observatory site is located ∼50 km south of Harwell in an agricultural (mainly arable) landscape, ∼200 m south-east of the edge of Chilbolton village, Hampshire (lat: 51.150°, lon: −1.438°, altitude: 78 m) and 100 km south-west of London (Defra 2018b). Two single-carriageway main roads run ∼1 km to the west (A3057), and ∼3 km to the south (A30) (figure 1). The site began monitoring as an EMEP level-2 supersite on 11th January 2016, after instrument relocation from Harwell.

Measurement data
The full suite of measurements at the two sites is summarised in UK Eutrophying and Acidifying Pollutants (UKEAP) network reports (UK-AIR Library 2018), a subset of which are investigated in this work (table 1, and supplementary information figure S1 is available online at stacks.iop.org/ERC/1/041001/mmedia). Concentration data were downloaded from the UK Department for Environment, Food and Rural Affairs (Defra) online data repository (UK-AIR Data Selector 2018).
Data are evaluated here for the two years either side of site relocation (Harwell 2014-15 and Chilbolton 2016-17, respectively). Data capture statistics are given in table 1. O 3 , NO x , SO 2 , PM 10 and PM 2.5 measurements derive from the UK Automatic Urban and Rural Network (AURN). Instrumentation selection, calibration and data ratification follow EU Air Quality Directives (2008/50/EC), and data are archived as hourly averages (UK-AIR Library 2018). Hourly measurements of NH 3 and other trace gases, together with water-soluble ions within PM 2.5 and PM 10 , are provided by the Monitor for AeRosols and Gases in Air (MARGA) instrument (EMEP 2007, Stieger et al 2018 with data quality assurance processes as described in Twigg et al (2015).
Hourly meteorological data comprise on-site measurements at Harwell and Chilbolton (Wrench). Data were also downloaded for the meteorological station at Benson, Oxfordshire (lat: 51.616°, lon: −1.096°, altitude: 57 m), marked on figure 1(a), from the NOAA Integrated Surface Database using the worldmet package (Carslaw 2018, NOAA 2018 for the full time period being considered. These data were used to validate the use of concatenated meteorological time series from the two supersites (see supplementary information).

Meteorological detrending
To examine for evidence of a step-change in concentration coincident with site relocation, the deweather function (Carslaw 2017) was applied to concatenated Harwell and Chilbolton datasets (2014-17) in a technique Table 1. Annual mean concentrations (with 95% confidence intervals) and annual percentage data capture of measurements for the investigated species at Harwell (2014-15), andChilbolton (2016-17 known as 'meteorological normalisation' (Grange et al 2018). This accounts for non-linear and complex relationships between predictors, such as meteorological or temporal variables (Carslaw and Taylor 2009), allowing changes in time series not directly caused by these predictors to be identified. Models are built using a stochastic process that results in reduced variance of the final model (Friedman 2002), but consequently, a slightly different model is produced with each run (Elith et al 2008). A set of 10 identically-built deweather models were performed for each pollutant time series, using meteorological predictor variables measured at the same site as the pollutants. Further details of the model, including comparisons using different meteorological data, are given in the supplementary information (section S2 and figure S3).

Emission inventories
Annual emission estimates for NH 3 , NO x , SO 2 , PM 2.5 and PM 10 were obtained from the UK National Atmospheric Emission Inventory (NAEI, http://naei.beis.gov.uk/) for both locations. Individual species emissions were taken from the 2015 inventory and were aggregated over the 15 km×15 km area surrounding each site shown in figure 1. The areas were defined according to the gridded agricultural sector NH 3 emissions which have a spatial resolution of 5 km×5 km, but for all other pollutants the underlying resolution was 1 km×1 km.

Results and discussion
3.1. Overview Table 1 summarises the annual mean concentration and corresponding 95% confidence interval for each pollutant for each year. Chilbolton MARGA measurements have the lowest data capture (54%-58%) since these measurements did not commence until 11th February 2016, and instrument issues led to missing data between 9th July and 1st September 2016. Between August 2014 and September 2015 a plume from the volcanic eruption at Holuhraun, Iceland, passed over the UK and is observed in the SO 2 time series (figure S1). Similarly in spring 2014, elevated PM concentrations were caused by a combination of Saharan dust and ammonium nitrate formed from European emissions (Vieno et al 2016) (figure S1). Other PM episodes are also apparent between 2014 and 2017 (figure S1), typically accumulating in low wind speeds and lasting no more than a few days (Defra 2015, 2016, 2017).

NO x
The average 2016 and 2017 concentration of NO x at Chilbolton was approximately 1.6 times greater than the average 2014 and 2015 concentration at Harwell (annual means of 18.5 and 13.8 μg m −3 cf 10.5 and 9.20 μg m −3 respectively, table 1). The NO x deweather time series also shows an abrupt increase coincident in timing with the relocation of the measurement site (figure 2(a)).
In contrast to the greater average NO x concentrations at Chilbolton, the total local NO x emissions integrated over the 15 km×15 km area around Harwell are more than 5 times greater than from the same-size area around Chilbolton (figure 3). The NO x emissions close to Harwell are dominated by specific sources including Didcot town, Didcot B power station in the north-east, and high traffic flow on the dual-carriageway (A34) running north-south 2 km to the east. Across the 225 individual 1 km×1 km grid squares within the area surrounding Harwell, 47% of total local emissions are contributed by the grid square with the highest emissions, whilst the 25 highest emission grid squares contribute 84%. Similar analysis at Chilbolton shows less dominance of local emissions from the single highest (6%) and 25 highest (61%, including part of the A303, ∼8 km north) grid squares.
These differences in NO x source configurations are illustrated by pollution wind roses (figure 4). The aforementioned sources north of Harwell generate the most polluted air transported to the site but do not contribute significantly more to the annual average than other directions. Considering NO x to have an atmospheric lifetime of ∼4-6 h in the mid-latitudes (Beirle et al 2011) and an average wind speed of ∼15 km h −1 for southern England (figure S2; Wrench, NOAA 2018), the London pollution plume could plausibly be observed at both sites. It appears Harwell is less influenced by London (∼100°bearing) than Chilbolton (∼70°), for which this wind direction provides a dominant contribution to the annual average. However, the M3 motorway also contributes NO x to this wind direction, stretching for ∼56 km between Chilbolton and London. Traffic along the length of the M3 (∼95 km) and other major roads are the probable reason why a larger average concentration of NO x (>2.2 times higher) is observed in easterly (0°-180°) than in westerly winds (181°-360°, table 2). More frequent westerly winds during 2017 led to a reduced annual average compared with 2016 (13.8 cf 18.5 μg m −3 ).
The importance of wind direction to NO x concentrations at both sites is reflected in the relative importance of variables used in the deweather models (table S2). Wind direction had the largest mean relative importance of 19.1%, closely followed by ambient temperature (18.7%) and (long-term) trend (18.4%). Unsurprisingly, predicted concentrations of NO x are elevated at lower ambient temperatures (10°C) and during colder months (October-February), as these conditions limit the dispersion of emissions (AQEG 2004), and there is a higher demand for domestic heating. The importance of the trend variable is a consequence of the abrupt increase in NO x at the time of site relocation in the concatenated deweather time series ( figure 2(a)). This supports the interpretation of a real difference in annual mean NO x concentration between Harwell and Chilbolton.

NH 3
The annual mean NH 3 concentrations at Chilbolton in 2016 and 2017 (5.88 and 6.23 μg m −3 , respectively, table 1) are about 3 times higher than at Harwell in 2014 and 2015 (1.96 and 2.05 μg m −3 , respectively), the greatest difference of all pollutants investigated. Although data capture at Chilbolton in 2016 was rather low (58%), and most missing data occurred during the summer (figure S1) when NH 3 concentrations are typically larger (Tang et al 2018), therefore the annual average in 2016 is likely biased low by the missing data.
Application of the deweather model to the concatenated time series of NH 3 concentrations ( figure 2(b)) confirms a distinct increase in mid-February 2016, coincident with commissioning of the MARGA at Chilbolton. Trend is the dominant variable in the deweather model (45.5±0.6%, figure S4 and table S2).   3). However, the total at Harwell includes contribution from a disproportionately large 1 km×1 km grid square over the site of Didcot B power station, over 3.4 times greater than any other grid square surrounding either site. When emissions from only the agricultural sector are considered, the greatest emissions are associated with the 5 km×5 km grid square also containing the site, suggesting the presence of a nearby agricultural source or sources.
Polar plots of NH 3 concentration as a function of wind speed and direction for both sites (figure 5) reveal an obvious dominant local source to the south-west of Chilbolton. Figure 6 demonstrates the difference in concentration between this source and background NH 3 at Chilbolton in 2016, by dividing prevailing wind into four direction sectors: east (E, 30°-180°), south-west (SW, 180°-250°), north-west (NW, 250°-350°) and north (N, 350°-30°). The N sector appears to reflect background NH 3 levels, with a mean concentration of 2.2 μg m −3 , comparable to average NH 3 concentrations measured at Harwell across all wind directions (figure 5). Mean concentrations in E, NW and SW sectors are higher (3.1, 3.4 and 7.1 μg m −3 , respectively) and demonstrate an inverse relationship with wind speed, suggesting nearby sources at Chilbolton. Approximately 1.5 km from the    site in the SW sector is the centre of a cattle farm, whilst a further 1.5 km in the same direction is a mushroom farm, likely to release NH 3 at regular intervals coinciding with mushroom growth cycles (Sather et al 2008). For the remaining E/NW sectors, NH 3 concentrations presumably reflect levels associated with the surrounding intensively managed arable land, which in general are larger emitters than agricultural activities near Harwell.

SO 2
Hourly SO 2 concentrations at both sites are determined by two different measurement methodologies (table 1) but there was poor agreement between them ( figure S5). This is anticipated, as the limit of detection ( Following this event there are no discernible changes in SO 2 concentration, including through the site relocation, apart from a singular peak in August 2014 which can be attributed to an Icelandic volcanic eruption (Twigg et al 2016). Whilst the emission inventory estimates greater local SO 2 emissions around Harwell than around Chilbolton (figure 3), these are overwhelmingly dominated at Harwell by Didcot B power station. The relative infrequency of north-easterly wind to the site, coupled with the elevated source of these emissions, resulted in little influence of Didcot B on ambient measured concentrations. Consequently, there is effectively no change in measured SO 2 arising from the site relocation.

Particulate matter
Data capture was high for total PM 10 and PM 2.5 (table 1)  . Data points are medians of 50 hourly measurements sorted by ascending wind direction, following Flechard and Fowler (1998). almost twice any other hourly PM 2.5 concentration. The stochastic nature of the deweather function resulted in some models being built using this data point, while others were not.
The annual NAEI emissions integrated over the 15 km×15 km area surrounding each site are substantially greater around Harwell than Chilbolton for both PM 10 and PM 2.5 (figure 3). Locations of high PM 2.5 emissions within the area surrounding Harwell include those with substantial NO x emissions (Didcot town, Didcot B power station and the A34). The 25 1 km 2 grid squares with highest PM emissions around Harwell contribute 63% to the total, whereas the equivalent at Chilbolton contribute 54%. As with NO x , an increase in PM 2.5 concentrations is observed in the prevailing easterly wind at Chilbolton as compared to Harwell (13.2 in 2016 cf 8.1 μg m −3 in 2015, table S3), however westerly winds have lower concentrations (6.8 cf 9.6 μg m −3 ), which accounts for the comparable annual averages pre-and post-relocation. The same conclusions apply for PM 10 (figure S6).
Data capture for MARGA measurements were poorer than for TEOM-FDMS measurements (table 1). Nevertheless for all secondary inorganic aerosol (SIA) ions analysed, the data capture rate was sufficient for comparisons of annual measurements (all years >58%).  (table 1). However, the deweather model time series for O 3 data (figure 2(e)) does not show evidence of a step-change associated with the site relocation (the modelled decrease in mid-2016 is coincident with the similar observation for PM). The slightly lower annual mean background O 3 at Chilbolton than at Harwell may be due to interannual variability in O 3 , but the lower background concentration is also consistent with an inverse concentration relationship with the unambiguously higher background NO x at Chilbolton.

Conclusions
The relocation of the southern UK EMEP supersite from Harwell to Chilbolton in January 2016 has not resulted in discontinuities in average measured concentrations of PM 2.5 , PM 10 , SO 2 and O 3 (based on two-year pre-and post-relocation time comparisons), but has led to substantial increases in average concentrations of NO x and NH 3 , by a factor of ∼1.6 and ∼3, respectively. Concentrations of NO x and PM in easterly wind arriving at Chilbolton are ∼2.7 and ∼1.5 times larger than at Harwell, from common sources including the M3 motorway and Greater London. Prevailing winds from the west contribute similar NO x concentrations at both sites, therefore despite a higher frequency of westerly wind at Chilbolton, the larger easterly concentrations result in larger annual means. Westerly winds carry lower concentrations of PM to Chilbolton than Harwell, thereby resulting in similar annual averages. Measurements show no substantive difference in the contribution of secondary inorganic aerosols (SIA) and black carbon to the  total PM mass between the two site locations, although more NH 4 + events are observed in the SIA at Chilbolton.
Background concentrations of NH 3 at both sites reflect the presence of mixed farmland; however the contribution of very strong local sources to the south-west of Chilbolton cause the large increase in annual average between sites.
In conclusion, when considering long-term regional trends based on the southern UK supersite data, the increase in NO x and NH 3 at Chilbolton must be taken into account and the Harwell and Chilbolton datasets should be treated separately.