Drivers for spatial, temporal and long-term trends in atmospheric ammonia and ammonium in the UK

. A unique long-term dataset from the UK National Ammonia Monitoring Network (NAMN) is used here to assess spatial, seasonal and long-term variability in atmospheric ammonia (NH 3 : 1998–2014) and particulate ammonium (NH + 4 : 1999–2014) across the UK. Extensive spatial heterogeneity in NH 3 concentrations is observed, with lowest annual mean concentrations at remote sites (< 0.2 µg m − 3 ) and highest in the areas with intensive agriculture (up to 22 µg m − 3 ), while NH + 4 concentrations show less spatial variability (e.g. range of 0.14 to 1.8 µg m − 3 annual mean in 2005). Temporally, NH 3 concentrations are inﬂuenced by environmental


Introduction
Atmospheric ammonia (NH3) gas is assuming increasing importance in the global pollution climate, with effects on local to international (transboundary) scales (Fowler et al., 2016). While substantial reductions in SO2 emissions and limited reductions in NOx emissions have been achieved in Europe and North America following legislation designed to improve air quality, NH3 emissions, primarily from the agricultural sectors (94 % of total NH3 emissions in Europe in 2014) have seen much smaller 5 reductions (EEA, 2016). In the period 2000-2014, NH3 emissions are estimated to have decreased in the EU-28 (28 member states of the European Union) by only 8 % from 4.3 to 3.9 million tonnes, with the UK contributing 7.2 % in 2014 (EEA, 2016). SO2 emission are estimated to have declined by 69 % and NOx by 39 % across the EU-28 over the same period.
NH3 is known to contribute significantly to total nitrogen (N) deposition to the environment, and causes harmful effects through 10 eutrophication and acidification of land and freshwaters. This can lead to a reduction in both soil and water quality, loss of biodiversity and ecosystem change (e.g. Pitcairn et al., 1998;Sheppard et al., 2011). In the atmosphere, NH3 is the major base for neutralization of atmospheric acid gases, such as SO2 and NOx emitted from combustion processes (vehicular and industrial) and from natural sources, to form ammonium-containing particulate matter (PM): primarily ammonium sulphate ((NH4)2SO4) and ammonium nitrate (NH4NO3). This secondary PM is mainly in the 'fine' mode with diameters of less than 15 2.5 µm (i.e. PM2.5 fraction) (Vieno et al., 2014). The effects of PM on atmospheric visibility, radiative scattering, cloud formation (and resultant climate effects) and on human health (bronchitis, asthma, coughing) are well documented (e.g. Kim et al., 2015;Brunekreef et al., 2015). Inputs of NH3 and NH4 + (collectively termed NHx) are the dominant drivers of ecological effects of deposited N, compared with wet deposited NH4 + in rain (UNECE, 2016) and the importance of NHx can be expected to increase further, relative to oxidised N, as NOx emissions have been decreasing faster than NH3 emissions (Reis et al., 2012;20 EEA, 2016;EU, 2016).
In gaseous form, NH3 has a short atmospheric lifetime of about 24 hours (Wichink Kruit et al., 2012). It is primarily emitted at ground level in the rural environment, and is associated with large dry deposition velocities to vegetation (Sutton and Fowler, 2002). High NH3 concentrations can lead to acute problems at a local scale to, for example, nature reserves located in intensive 25 agricultural landscapes Cape et al., 2009a;Hallsworth et al., 2010;Vogt et al., 2013). The NH3 remaining in the atmosphere generally partitions to PM where the NH4 + can have a lifetime of several days (Vieno et al., 2014). Although NH4 + dry deposits at the surface, the primary removal mechanism for NH4 + is thought to be through scavenging of PM by cloud and rain, leading to wet deposition of NH4 + (Smith et al., 2000). Characterising the relationship between NH3 emissions and the formation of PM is, however, not straight forward; an increase in NH3 emissions does not automatically translate to a 30 proportionate increase in NH4 + (Bleeker et al., 2009). The relationship depends on climate and meteorology as well as the concentration of other precursors to PM formation such as SO2 and NOx (Fowler et al., 2009). Since UK particulate NH4+ is generally dominated by NH4NO3 and (NH4)2SO4 (see e,g. Twigg et al., 2016 andMalley et al., 2016) and NH3 gas is present in excess, then gas-particle transfer of NH3 to NH4 + is the dominant pathway for forming NH4+ in PM. While it is clear that reductions in NH3 emissions will lead to reductions in overall NH4 + concentrations (Vieno et al., 2016), the relative changes in 35 gaseous NH3 and NH4 + particles remains poorly quantified.
International targets have been agreed to reduce NH3 emissions to move towards protection against its harmful effects. These include the UNECE Convention on Long-Range Transboundary Air Pollution (CLRTAP) Gothenburg Protocol and the recently revised EU National Emission Ceilings Directive (NECD 2016(NECD /2284 (EU, 2016). The 1999 UNECE Gothenburg 40 Protocol is a multi-pollutant protocol to reduce acidification, eutrophication and ground-level ozone by setting emissions ceilings for sulphur dioxide, nitrogen oxides, volatile organic compounds and ammonia, which are to be met by 2020. Revised in 2012, the protocol requires national parties to jointly reduce emissions of NH3, in the case of the EU-28 by 6 % between 2005 and 2020 (Reis et al., 2012). Under the revised NECD (EU, 2016), the EU is also committed to reductions of 6% for NH3, but by a later date of 2029, as well as an additional 13% reduction in NH3 emission beyond 2030 compared with a 2005 baseline.
Although this demonstrates that there is currently no strong commitment to reduce NH3 emissions compared with SO2 and 5 NOx, other supporting measures should also be noted including the Industrial Emissions Directive 2010/75/EU (IED), which requires pig and poultry farms (above stated size thresholds) to reduce emissions using Best Available Techniques. The IED applies to around 70 % of the European poultry industry and around 25 % of the pigs industry (UNECE, 2010). In tandem, revised UNECE 'Critical Levels' (CLe) of NH3 concentrations to protect sensitive vegetation and ecosystems were adopted in (UNECE, 2007. These set limits of NH3 concentrations to 1 µg NH3 m -3 and 3 µg NH3 m -3 annual mean for the protection 10 of lichens-bryophytes and other vegetation, respectively (Cape et al., 2009b). The new CLes replaced the previous single value of 8 µg NH3 m -3 (annual mean) and have since been adopted as part of the revised Gothenburg Protocol. Such CLes for NH3 are widely exceeded, including over the areas designated as Special Areas of Conservation (SAC) under the Habitats Directive, and indicates a significant threat to the Natura 2000 network established by that directive (Bleeker et al., 2009;Hallsworth et al., 2010;van Zanten et al., 2017). 15 Few countries have established systematic networks to measure NH3 across their domains. In the Netherlands, a continuous wet annular denuder method (AMOR, replaced by the DOAS (Differential Optical Absorption Spectroscopy) device in 2015) has been used at 8 stations in the Dutch National Air Quality Monitoring Network (Van Pul et al., 2004;van Zanten et al., 2017). The Ammonia in Nature (MAN) network established in 2005 in the Netherlands monitors NH3 with passive diffusion 20 tubes in Natura 2000 areas (Lolkema et al., 2015). In the USA, the Ambient Ammonia Monitoring Network (AMoN) uses passive (Radiello) samplers at 50 sites since Oct 2010 (Puchalski et al., 2011). Hungary ), Belgium (den Bril et al., 2011), Switzerland (Thöni et al., 2004, West Africa (Senegal and Mali under the Pollution of African Capitals program; Adon et al., 2016) and China (Xu et al., 2016) also have long-term NH3 measurements (see review by Bleeker et al., 2009) . 25 In the UK, the National Ammonia Monitoring Network (NAMN) was established in September 1996 with the aim of establishing long-term continuous monthly measurements of atmospheric NH3 gas . Particulate NH4 + measurements were added in 1999, since this was expected to exhibit different spatial patterns and temporal trends to gaseous NH3 (Sutton et al., 2001b). The NAMN thus provides a unique and important long-term record for examining responses to 30 changing agricultural practice and allows assessment of the compliance of NH3 emissions with targets established by international policies on emissions abatement. Measurements of NH3 and NH4 + in the NAMN also address spatial patterns, covering both source and sink areas to test performance of atmospheric transport models, to support estimation of dry deposition of NHx, to improve estimation of the UK NHx budget (Fowler et al., 1998;Smith et al., 2000;Sutton et al., 2001b) and to assist with the assessment of exceedance of critical loads and critical levels (UNECE, 2007). 35 This paper provides an analysis on the state of atmospheric concentrations of NH3 and NH4 + in the UK from 1998 to 2014 and their spatial and temporal trends. Overall, 17 years of continuous long-term NH3 measurement data and 16 years of continuous long-term NH4 + measurement data from the NAMN are analysed to assess trends in concentrations in relation to estimated changes in emissions. The long-term measurement dataset is also used to explore spatial and temporal patterns in NH3 and 40 NH4 + across the UK in relation to regional variability in emission source sectors.

2
Material and Methods

Network structure and site requirements
The design strategy for NAMN was to sample at a large number of sites (>70) using low-frequency (monthly) sampling for cost-efficient assessment of temporal patterns and long-term trends. The network covers a wide distribution of monitoring sites with measurements in both agricultural and semi-natural areas. Monitoring locations are sited away from point sources (> 150 5 m) such as farm buildings, which avoids overestimating NH3 concentrations compared with the grid square, since the aim is to provide meso-scale and regional patterns. In addition, where sampling is carried out in woodland areas, it is made in clearings. It was also recognised that the location of the network sites needed to consider the extent of sub-grid variability and the representativeness of sampling points. Spatially detailed local-scale NH3 monitoring was therefore also carried out at a sub-1 km level to assess the extent to which a monitoring location is representative (Tang et al., 2001b). The NAMN started 10 with 70 sites. Over time, new sites were added to fill gaps in the map, some sites were closed following reviews and some sites had to be relocated due to local reasons, for example land ownership changes or site re-development. The number of sites peaked at 93 in 2000, but since 2009 has been stable at 85 sites. The locations of the NAMN sites for NH3 and NH4 + in 2012 are shown in Figure 1a & b.

<INSERT FIGURE 1>
The selection of NAMN sites to provide a representative concentration field across the UK was aided by the availability of an estimated UK NH3 concentration field at a 5 km by 5 km grid resolution provided by the Fine Resolution Atmospheric Multipollutant Exchange (FRAME) model (Singles et al., 1998;Fournier et al., 2002). A comparison of FRAME modelled NH3 20 concentrations for NAMN sites with FRAME modelled concentrations for the whole of the UK shows that the network has a good representation in the middle air concentration classes of 0.5 -1.5 µg m -3 (33 % of NAMN sites, compared with 29 % of all FRAME 5 km x 5 km grid squares) and 1.5 -3 µg m -3 (32 % of NAMN sites, compared with 39 % of all FRAME 5km x 5 km grid squares), but with an over-representation at high concentrations and under-representation at low concentrations ( Figure   1c). Since air concentrations are more variable in high concentration areas, a larger number of monitoring sites were located 25 in these areas than in remote low concentration areas where air concentrations are more homogeneous. Similarly, the monitoring sites were strategically selected to cover source areas of expected high concentrations and variability on the basis of the FRAME model NH3 concentration estimates (Figure 1a & b), and this approach was expected to provide additional evidence to test the performance of atmospheric dispersion models (Fournier et al., 2005;Dore et al., 2015). When compared with other atmospheric chemistry transport models, FRAME was found to correlate well with measured NH3 concentrations 30 (Dore et al. 2015). The NAMN sites were also similarly checked for representativeness of particulate NH4 + by comparing FRAME modelled NH4 + concentrations at NAMN sites with modelled concentrations for the whole of the UK, which demonstrates a good representation across the range of expected concentrations (Figure 1d).

Atmospheric NH3 and NH4 + measurements
Monthly time-integrated measurements of atmospheric NH3 are made in the NAMN using a combination of passive samplers 35 Tang et al., 2001a) and an active diffusion denuder method referred to as the DEnuder for Long Term Atmospheric (DELTA) sampler (Sutton et al., 2001a &c). In terms of passive samplers, membrane diffusion tubes (3.5 cm long) with a limit of detection (LOD) around 1 µg NH3 m -3  were used in the first 4 years (September 1996-April 2000. These were replaced in May 2000 with the more sensitive Adapted Low-cost, Passive High Absorption (ALPHA, LOD = 0.03 µg NH3 m -3 ) diffusive samplers Tang and Sutton, 2003), following a period of 40 parallel testing (Sutton et al., 2001c). Particulate NH4 + measurement was added to the NAMN in 1999 at all DELTA sites (50) in the first two years (1999 and 2000).
Following this initial period, the sampling density was reduced during early 2001 to 37 sites and has been stable at 30 sites since 2006. Although not presented in this paper, the DELTA samplers additionally provide concentrations of acid gases (HNO3, SO2, HCl) and aerosols (NO3 -, SO4 2-, Cl -, Na + , Ca 2+ , Mg 2+ ) for the UK Acid Gas and Aerosol monitoring network (AGANet) at a subset of NAMN DELTA sites (Tang et al., 2015;Conolly et al., 2016). Measurement data from the AGANet 5 are used to aid interpretation of NH3 and NH4 + results in Sect. 3.5.6.

DELTA method
The DELTA method uses a small pump to sample air (0.2 to 0.4 L min -1 ) in combination with a high-sensitivity gas meter to record sampled volume (Sutton et al., 2001c). Two citric acid coated denuders (10 cm long borosilicate glass tubes) in series 10 are used to collect NH3 gas and to check the collection efficiency. A collection efficiency correction is applied to the measurement (Sutton et al., 2001d). The corrected air concentration (a (corrected)) is determined as in Equation 1: Typically, denuder collection efficiency is better than 90% (Conolly et al., 2016). At 90 % collection efficiency, the correction represents < 1 % of the corrected air concentration. Individual measurements with collection efficiency < 75 % (correction 15 amounts to 11 % of the total at 75%) are flagged as valid, but less certain (Tang and Sutton, 2003). Where less than 60 % of the total capture is recorded in the first denuder, the correction factor amounts to greater than 50 % and is not applied. The air concentration of (a) of NH3 is then determined as the sum of NH3 in denuders 1 and 2 (Equation 2): (2) 20 At sites where particulate NH4 + is also sampled, a 25 mm filter pack with a citric acid impregnated cellulose filter is added after the denuders to capture the NH4 + . The calculated air concentrations (a) of NH4 + is corrected for incomplete capture of NH3 by the double denuder. The corrected air concentrations (a (corrected)) of NH4 + is determined as in Equation 3: 25 a (corrected NH4 + ) = a (NH4 + ) -[((a (corrected NH3) -[(a (Denuder 1 NH3) + a (Denuder 2 NH3)])* (18/17)] (3) For NH4 + sampling, loss of NH3 due to volatilisation of NH4 + from the acid impregnated filter has been investigated, by adding a third citric acid coated denuder after the filter pack which was found to be negligible. At DELTA sites where additional simultaneous sampling of acid gases and particulate phase components are made for AGANet, ion balance checks between 30 anions and cations in the particulate phase are performed to provide an indication of the quality of the particulate measurements. For the acid and base particulate components, close coupling is expected between NH4 + and the sum of NO3and SO4 2-, as NH3 is neutralised by HNO3 and H2SO4 to form NH4NO3 and (NH4)2SO4, respectively (Conolly et al., 2016).
At the Bush OTC site in Scotland (UK-AIR ID = UKA00128), duplicate DELTA measurements are made to assess the 35 reproducibility of the method. For continuous monthly measurements between 1999 and 2014, the R 2 between the duplicate systems was 0.96 for both NH3 and NH4 + (supp. Figure S1).

Passive methods
The NH3 membrane diffusion tubes deployed in the NAMN from 1996 to 2000 are hollow cylindrical tubes (FEP, 3.5 cm long). A cap at the top end holds in place two stainless steel grids coated with sulphuric acid. The lower air-inlet end of the tube is capped with a gas-permeable membrane Tang et al., 2001a;Thijsse, 1996). In comparison, the ALPHA passive sampler is a badge-type high sensitivity sampler with an uptake rate that is ~20 times faster than the diffusion 5 tube. It consists of a cylindrical low-density polyethylene body. An internal ridge supports a cellulose filter coated with citric acid, which is held in place with a polyethylene ring. The open end is capped with a PTFE membrane, providing a diffusion path length of 6 mm between the membrane and absorbent surface .
Triplicate passive samplers are deployed for every measurement in the NAMN. Where the % coefficient of variation (CV) of 10 the triplicate samplers is greater than 30% for the diffusion tubes or greater than 15% for the ALPHA samplers, the sample run is classed as failing the quality control test. Large discrepancies are most likely due to contamination of samples and data from contaminated samples are excluded from the assessment in this paper. and Stoke Ferry (UKA00317). The inter-comparison is used to establish a regression between the active and passive methods, with the DELTA samplers as the reference system, since the air volume sampled is accurately measured with high sensitivity 20 gas meters. The calibration is necessary to account for the fact that the sampling path length in the passive samplers is longer than the distance between the membrane and adsorbent, due to the additional resistance to molecular diffusion imposed by the turbulence damping membrane at the inlet and the presence of a laminar boundary layer of air on the outside of the sampler . In addition, parallel measurements were made at a high NH3 concentration farm site (1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007) to extend the calibration range, and to ascertain linearity of response to high concentrations. To ensure that no bias is introduced in the 25 sampling and to maintain the validity of long-term trends, the calibration is evaluated on an annual basis (Tang and Sutton, 2003;Conolly et al., 2016).
For the period up to 2000 when the diffusion tubes were implemented in the NAMN, their calibration (at 10 µg m -3 ) amounts to an average of 1.5 % compared with the DELTA system. The mean ALPHA sampler calibration (at 10 µg m -3 ), compared 30 with the DELTA system, amounts to a correction of 10 % ( membrane (5 µm pore size) is supported on a regular polypropylene grid and is thicker (305 µm) than the earlier PTFE membrane (also 5 µm pore size, but 265 µm thickness) used which was supported instead on a randomly arranged 35 polypropylene support material. The difference in calibration was therefore due to the extra resistance to gas diffusion imposed by the new thicker membrane. The annual calibration of the methods shows both high precision and constancy between years ( Figure 2), which is important to support the detection of temporal trends in NH3 concentrations. There is no systematic trend over time in either of the passive method calibrations.

<INSERT FIGURE 2>
The comparison of monthly measurement data between the DELTA and calibrated passive measurements demonstrated a close agreement ( Figure 3). The correlation (R 2 ) between DELTA and calibrated diffusion tubes was 0.91 (Figure 3a), while the correlation between DELTA and calibrated ALPHA samplers was 0.92 ( Figure 3b). From the calibrated results, the intercept for the diffusion tubes was 0.10 µg NH3 m -3 , while that for the ALPHA samplers was 0.03 µg NH3 m -3 , demonstrating the improvement in sensitivity with the ALPHA samplers compared with the diffusion tubes ( Figure 3). In the present case the 5 value of the intercepts, even for diffusion tubes, is much less than typical NH3 air concentrations (see Sect. 3). However, this cannot be assumed to be the case in other implementations of the same methods. Experience from other studies using the lower sensitivity diffusion tubes indicates a tendency to overestimate NH3 concentrations under clean conditions (RGAR, 1990;Thijsse et al., 1996;Tang et al., 2001a;Lolkema et al., 2015). This observation points to the need for any application of NH3 passive sampling for ambient monitoring to be accompanied by testing and calibration against a verified active sampling 10 method. In independent assessments, for example in the USA (Puchalski et al., 2011), the ALPHA samplers performed well against a reference annular denuder method with a median relative percent difference of −2.4%.

2.2.3
Chemical analysis NH3 gas captured on the acid coating of the denuder (DELTA), grid (diffusion tubes) or filter paper (ALPHA), and particulate NH4 + captured on the DELTA aerosol filter, are extracted into deionised water and analysed for NH4 + on an ammonia flow injection analysis system. The analytical instrument has changed over the network's operational period from the AMFIA (ECN, NL) to the FloRRIA (Mechatronics, NL), an updated model based on AMFIA (Conolly et al., 2016). The principles of 20 operation of both instruments are the same and are based on selective diffusion of NH4 + across a PTFE membrane at c. pH 13 into a counter-flow of deionized water, allowing selective detection of NH4 + by conductivity (Wyers et al., 1993). The extracted samples were analysed for NH4 + against a series of NH4 + standards and quality controls. Parallel analysis of laboratory and field blank (unexposed) samples were used to determine the amounts of NH4 + derived from NH3 and NH4 + in the atmosphere during transport and storage. The limit of detection (LOD) calculation of the ALPHA and DELTA methodologies are 25 determined as three times the standard deviations of the laboratory blanks. For the DELTA method, the LODs were 0.01 μg m -3 for gaseous NH3 and 0.02 μg m -3 for particulate NH4 + . For the ALPHA method, the LOD was determined as 0.03 μg m -3 .

Data Quality Control
Measurement data are checked and screened, based on the quality management system applied in the UK air monitoring networks (Tang and Sutton, 2003). Data quality is assessed against the following set quality control criteria: a) DELTA system: 30 monitoring of the air flow rate and the use of two denuders in every sample to assess capture efficiency for NH3, and b) passive samplers: use of triplicate samplers for monitoring NH3 concentrations at every site, to allow an assessment of sampling precision, and c) ongoing calibration of passive samplers against the DELTA. Data flags are applied to the dataset; a full list of these is available from the EMEP website (http://www.nilu.no/projects/ccc/flags/index.html). Following the quality control checks and data flagging on the collected dataset, the annually ratified data from the NAMN are made publically available on 35 the Department for Environment, Food & Rural Affairs (Defra) UK-AIR website (https://uk-air.defra.gov.uk/) and are also in the process of being made available on the EMEP website (http://ebas.nilu.no/). An intercomparison of NH3 measurements by the RIVM AMOR system (hourly, Wyers et al., 1993)  results, monthly mean concentrations were derived from the average of hourly AMOR data for the corresponding DELTA and ALPHA monthly sampling periods with good agreement (supp. Figure S6).

Trend Analyses
Statistical trend analysis was conducted on the long-term dataset from the UK NAMN to identify trends in the long-term 5 datasets (univariate monotonic, see e.g. Hirsch et al., 1991), estimate the rate of change and to address the question of whether trends in NH3 and NH4+ concentrations (if any) are consistent with the changes in estimated UK annual NH3 emissions (data downloaded from: http://naei.beis.gov.uk/data/data-selector-results?q=101505)? The dataset is sufficiently long-term (i.e. gaseous NH3: 17 years and particulate NH4+: 16 years) and collected by consistent methods, to allow for effective statistical trend analyses to be carried out. Trend analyses were carried out using (i) linear regression (LR), (ii) Mann-Kendall (MK) test 10 (Gilbert, 1987) on annually averaged and monthly mean data, and (iii) Seasonal Mann-Kendall (SMK) test (Hirsch et al., 1982) on monthly data only. Mann-Kendall tests were performed using the 'Kendall' package (McLeod, 2015) in the R software.
Computation of the Sen's slope and confidence interval (for non-seasonal Sen's slope only) of the linear trend were performed using the R 'Trend' package (Pohlert, 2016). Since concentrations of NH3 show strong seasonality, the SMK test was applied to identify the months that are driving the long-term trends in data. The SMK test (Hirsch et al., 1982) takes into account a 12 15 month seasonality in the time series data by computing the MK test on each of monthly 'seasons' separately, and then combining the results. So for monthly 'seasons', January data are compared only with January, February only with February, etc. No comparisons are made across season boundaries.
The Sen's slope is the fitted median slope of a linear regression joining all pairs of observations. For the SMK, an estimate of 20 the seasonal Sen's trend slope over time is computed as the median of all slopes between data pairs within the same season (i.e. January compared only with January etc.). Therefore no cross-season slopes contribute to the overall estimate of the SMK trend slope. Parametric LR analysis are simple and straightforward to use and interpret monotonic trend assessment in environmental data (e.g. Kindzierski et al., 2009;Meals et al., 2011), but they require assumptions about normality of data and homogeneity of variance of data. The MK approach on the other hand are widely used in environmental time series 25 assessments, e.g. long-term trends in precipitation (Serrano et al. 1999) and long-term trends in European air quality (Colette et al., 2016;Torseth et al., 2012). The main advantages, as discussed in the literature of the MK approach over linear regression for trend assessments are that (i) it does not require normally distributed data, (ii) it is not affected by outliers, and (iii) it removes the effect of temporal auto-correlation in the data. The MK approach are widely used in environmental time series assessments, e.g. long-term trends in precipitation (Serrano et al. 1999) and long-term trends in European air quality (Colette 30 et al., 2016;Torseth et al., 2012). However linear trend assessment have been used in UK air quality monitoring network reports (e.g. Conolly et al., 2016), therefore both approaches were used in this paper primarily as a quality assurance check.

Results and discussion
In order to summarise and discuss the NAMN dataset, the spatial patterns in the measurements of NH3 and NH4 + are considered 35 in Sect. 3.1 (comparison with emission estimates) and Sect. 3.2 (comparison with modelled concentration estimates), seasonal patterns are discussed in Sect. 3.3, and long-term trends across the UK in Sect. 3.4.

Spatial variability in NH3 and NH4 + concentrations in relation to estimated emissions
As a primary pollutant emitted from ground-level sources, NH3 exhibits high spatial variability in concentrations (Sutton et al., 2001b;Hellsten et al., 2008;Vogt et al., 2013), confirmed by NH3 data from the NAMN (e.g. range of 0.06 -8.8 µg m -3 annual mean in 2005) ( Figure 4a). The observed variability is consistent with the large regional variability in NH3 emissions and sources (Figure 4c & d). With agriculture being the main source of NH3 emissions, Figure 4a shows the largest concentrations of measured NH3 in parts of the UK with the highest livestock emissions, such as eastern England (East Anglia), north-west England (Eden Valley, Cumbria) and the border area between England and Wales (Shropshire) (Figure 4d). By contrast, the lowest NH3 measured concentrations are found in the north-west Scottish Highlands (< 0.2 µg m -3 ), which is 5 consistent with the emissions map ( Figure 4c). The 2005 data show exceedance of the Critical Levels for annual mean NH3 concentrations of 1 and 3 µg NH3 m -3 for the protection of lichens-bryophytes and vegetation, respectively (UNECE, 2007) at many of the sites (53 % > 1 µg NH3 m -3 and 13 % > 3 µg NH3 m -3 ). In 2014, exceedance of the 1 and 3 µg NH3 m -3 CLe increased to 60 % and 16 %, respectively. The widespread exceedance of the CLe for NH3 concentrations across the UK thus represents an ongoing threat to the integrity of sites designated under the Habitats Directive, as well as nationally designated 10 Sites of Special Scientific Interest (SSSI) and other sensitive habitats.
Concentrations of NH4 + are less spatially heterogeneous than those of NH3, based on data from 30 sites (e.g. range of 0.14 to 1.8 g m -3 annual mean in 2005) with a more coherent pattern of variation across the country, reflecting regional differences in NH3 concentrations ( Figure 4b). Thus there is a general decreasing gradient from the south-east to the north-west of the UK, 15 due to both NH3 sources in England and import of particulate matter from Europe (Vieno et al., 2014;Dore et al., 2015). The limited variation across the UK for the annual average NH4 + concentrations can be attributed to the atmospheric formation process (providing a diffuse source) and its longer atmospheric lifetime. density, but the correlation was smaller for NH4 + than for NH3 because of the larger contribution to NH4 + concentrations from long-range transport in the Netherlands.

<INSERT FIGURE 4>
The UK NH3 emissions inventory is calculated and spatially distributed annually. Agricultural sources at a 5 km by 5 km grid resolution are combined with a large number of non-agricultural sources Tsagatakis et al., 2016) at a 1 or 30 5 km resolution to produce the annual NH3 emissions data, and maps at a 1 km by 1 km grid resolution are reported by the official UK National Atmospheric Emissions Inventory (NAEI; http://naei.defra.gov.uk/data/mapping). In the UK, agriculture accounts for > 80% of total NH3 emissions and is estimated by the National Ammonia Reduction Strategy Evaluation System (NARSES) model (Webb & Misselbrook 2004;Misselbrook et al., 2015). For the agricultural NH3 emission maps, parish statistics on livestock numbers and crop areas are combined with satellite-based land cover data to model emissions at a 1 km 35 resolution, using the AENEID model (Dragosits et al., 1998;Hellsten et al., 2007). For reasons of data confidentiality, the 1 km data need to be aggregated to produce annual agricultural NH3 emissions maps at a 5 km by 5 km grid resolution. National The AENEID approach (Dragosits et al. 1998) can further be used to classify each 5 km by 5 km grid square in the UK into dominant NH3 emission source categories (Figure 4d), following the method of Hellsten et al. (2008), where grid squares with >45% from a given category are referred to as dominated by that source. The seven categories are: cattle, pigs & poultry (combined for data disclosivity reasons), sheep, fertilizer application to crops and grassland, non-agricultural sources, as well as a mixed category where no single source dominates, and background. Background grid squares are defined by very low 5 NH3 emissions of <1 kg N ha -1 y -1 .
Using the dominant emission sources map, each site in the NAMN is classified to one of the seven categories just described.
This provides information of the main emission source type expected in the 5 km by 5 km grid square containing the monitoring site and is useful for assessing whether the network has a good representation of key emission source categories (Supp. Figure  10 S2a & b). Over the period since the NAMN was established, from 1996 to present, there have been substantial changes in emissions estimated for the different source sectors. For analysis in this paper, the dominant sources map for 2005 emission year was used as representing the mid-point of the data series (1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014) and compared with the classification from other years for consistency. This categorization of sites is used further in the interpretation of the monitored NH3 and NH4 + concentrations and their long-term trends in the next sections. 15

Spatial variability in NH3 and NH4 + concentrations in relation to modelled concentrations
Comparison of measurements with modelled NH3 concentrations from the FRAME model for an example year of 2012 showed 20 significant scatter when considering the full network of sites (n = 85, R 2 = 0.62) (Figure 5a). In this graph, each point is colourcoded according to the estimated dominant NH3 emission source category for the 5 km by 5 km grid square. This updates a similar comparison from Sutton et al. (2001b) for the year 2000. The scatter may be explained by the large local spatial variability of NH3, related primarily to rapid decreases of NH3 concentrations with distance from a source (see e.g. Pitcairn et al., 1998;Dragosits et al., 2002), with the result that a single site measurement only gives an approximate indication of 25 concentrations across the model grid square it is located in. At many of the sites where the model overestimates concentrations, the measurements are in fact carried out in nature reserves, or in clearings inside forests. The monitoring sites in these sink areas are typically well away from local sources. Conversely, some of the outliers where measurements are larger than the model predictions show indications of being affected by nearby emission sources, as was established by investigations during site visits. 30

<INSERT FIGURE 6>
Figure 6 considers measured NH3 concentrations at a subset of sites (44 out of the full 85 sites) that are located away from nearby local sources, in forest or semi-natural areas, following the site classification and assessment by Hallsworth et al. 35 (2010). For this restricted set of sites, R 2 = 0.76 for 2012 which is higher than the correlation for the overall UK network. The improvement in correlation between measured and modelled NH3 concentrations for this subset of sites can be explained by the monitoring locations typically being further away from sources, so that uncertainties in local emissions estimates are to some extent averaged out. This observation is also consistent with the findings of Vieno et al. (2009).

40
In contrast to NH3, the correlation between NAMN measurements and FRAME model output is stronger for particulate NH4 + concentrations (R 2 = 0.87). However, measured concentrations are generally larger than the modelled ones (slope 1.1, intercept 0.16 µg m -3 (Figure 5b). One reason for the better agreement for NH4 + is the more slowly changing spatial patterns in concentrations, which are not expected to vary on a finer scale than the model's 5 km by km grid, improving the representativeness of site-based measurements. The 2012 comparison shown here updates an earlier inter-comparison assessment carried out by Dore et al. (2007) for the year 2002 and demonstrates that the FRAME model is performing well in describing the spatial distribution of NH4 + . However, for the 2012 inter-comparison, the FRAME model appears to 5 underestimate NH4 + at sites with concentrations < 0.6 µg NH4 + m -3 , with better agreement at concentrations above 0.6 µg NH4 + m -3 . This suggests either too low a formation rate for NH4 + in the model at cleaner sites, or too high a removal rate for NH4 + , or a combination of both. The presence of higher measured NH4 + concentrations in remote areas than shown by the model may also indicate that NH4 + has a longer residence time than treated in the model. Similar regressions between NAMN and FRAME NH4 + aerosol concentrations were observed for other years. For example, for 2008 the FRAME model underestimated NH4 + 10 at concentrations < 0.7 µg NH4 + m -3 (slope 1.2, intercept 0.26 µg -3 ; R 2 = 0.89, range = 0.2 -1.4 µg m -3 ). Changes in the chemical climate, such as reduced emissions of SO2 in the UK, are postulated to affect conversion rates of NH3 into NH4 + , as well as the dry deposition rates, leading to more NH3 remaining in the atmosphere (van Zanten et al., 2017). This is discussed further in Sect. 3.5.6.

Seasonal variability in measured UK NH3 and NH4 + concentrations 15
A comprehensive account of the seasonal variability of NH3 and NH4 + for different regions across the UK is provided by the NAMN. In Figure 7, the average seasonal cycles of grouped sites from four different emission source categories are compared for NH3 and NH4 + .

<INSERT FIGURE 7>
20 In addition to substantial differences in the overall magnitude of NH3 concentrations, where the largest concentrations in the network are found at sites dominated by pig and poultry farming, followed by areas where cattle farming predominates, it is clear that the seasonal patterns of NH3 also vary depending on the dominant source type (Figure 7a). For background sites (defined as located in grid squares with NH3 emissions <1 kg N ha -1 y -1 ), a clear summer maximum in NH3 concentrations can 25 be observed, with minimum concentrations occurring in winter. The summer peak is probably related to increased land surface NH3 emissions in warm, dry summer conditions, both from the presence of low-density grazing livestock and wildlife. It is also related to surface factors such as the compensation point for vegetation, which is defined as the concentration below which growing plants start to emit NH3 into the atmosphere (Sutton et al., 1995). The interaction between atmospheric NH3 concentrations and vegetation is complex, leading to both emission and deposition fluxes, depending on relative differences 30 in concentrations. However, it is well established that warm, dry conditions promote NH3 emission from vegetation (e.g. Massad et al., 2010;Flechard et al., 2013). It is therefore possible that bi-directional exchange with vegetation is at least partly controlling NH3 concentrations at remote sites distant from intensive livestock farming.
The possibility for such interactions can be considered further using the example of Inverpolly (UKA00457), a remote 35 background site in the NW Scottish Highlands. This site shows a very clear seasonal cycle with peak concentrations in July when warmer, drier conditions prevail, while lowest concentrations occur during the cooler and wetter winter months ( Figure   8a & b). A smaller peak in NH3 can also be seen annually in April, which indicates potential longer range influences of manure spreading in spring, even at this remote location (Figure 8b). Although there is substantial scatter, Figure 9 shows that there is significant correlation between monthly NH3 concentrations and both temperature (R 2 = 0.33, n = 231, p < 0.05) and 40 precipitation (R 2 = 0.19, n = 231, p < 0.05). The influence of temperature and rainfall on NH3 emission and concentrations is well characterised (e.g. see Sutton et al., 2013;van Zanten et al., 2017).

<INSERT FIGURE 8>
<INSERT FIGURE 9> 5 For sites dominated by emissions from sheep farming, the seasonal profile in NH3 concentrations is similar to that for background sites, although the summer maximum in NH3 is larger than background sites, because grazing emissions are larger (Hellsten et al., 2008). It is notable that the peak NH3 concentration occurs later in the year for background areas (July-September) than for sheep areas (June-August). This may be related to the seasonal presence of lambs, which are often only present for the first part of the summer. In areas with more intensive livestock farming, where emissions comes from either 10 cattle or from pig & poultry farming, the largest concentrations are observed in spring and autumn, corresponding to periods of manure application to land. The spring peak in March is larger than the autumn peak in September, which coincides with the main period for manure application being in spring, before the sowing of arable crops or early on in the grass-growing period (Hellsten et al., 2007). Ammonia concentrations in these areas are also larger in summer than winter, due to warmer conditions promoting volatilization. Interestingly, the dip in concentrations in June matches a period when crops will be 15 actively growing with possible uptake and removal of NH3 from the atmosphere. Vegetation can be a source or a sink of atmospheric NH3 and uptake of NH3 can occur when the relative concentration of NH3 in the atmosphere is higher than inside the plant stoma (e.g. Sutton et al.,1995, Massad et al., 2010Flechard et al., 2013).

20
For particulate NH4 + , as expected for a secondary pollutant, concentrations are more decoupled from the dominant NH3 source sectors in the vicinity of a site. Although the formation of particulate NH4 + primarily depends on the occurrence of NH3 in the atmosphere, synoptic meteorology and long range transboundary transport from continental Europe are important drivers influencing the seasonal variations of NH4 + across the UK, due to its' longer lifetime. (Vieno et al., 2014(Vieno et al., , 2016. The seasonal trends in particulate NH4 + are seen to be broadly similar for the four different emission source sectors (Figure 7b), with the 25 magnitude of the NH4 + concentrations reflecting NH3 concentrations at a regional level. In the atmosphere, particulate NH4 + are primarily in the form of (NH4)2SO4 and NH4NO3, formed when the acid gases HNO3 and H2SO4.in the atmosphere are neutralised by NH3 (Putaud et al., 2010). NH3 preferentially neutralizes H2SO4 due to its low saturation vapour pressure (forming NH4HSO4 then (NH4)2SO4), while NH4NO3 is formed when abundant NH3 is available, In contrast to (NH4)2SO4, NH4NO3 is a semi-volatile component (Stelson & Seinfeid, 1982). Long-term data from the UK Acid Gas and Aerosol Network 30 (AGANet, Conolly et al., 2016) shows a change in the particulate phase of NH4 + from (NH4)2SO4 to NH4NO3, with particulate nitrate concentrations exceeding that of particulate sulphate approximately three-fold (on a molar basis) (Fig. 18a). This suggests that the thermodynamic equilibrium between the gas phase NH3 and HNO3 and the aerosol phase NH4NO3 will have a much greater effect on the seasonal concentrations of NH4 + than (NH4)2SO4. The formation and dissociation of NH4NO3 depend strongly on ambient temperature and humidity (Stelson and Seinfeld, 1982). Warm, dry weather in summer promotes 35 dissociation, decreasing particulate phase NH4NO3 relative to gas phase NH3 and HNO3. During the winter months, low temperature and high humidity favour the formation of NH4NO3 from the gas phase NH3 and HNO3. By contrast, the spring peak in NH4 + concentrations may be attributed to photochemical processes (elevated ozone) leading to enhanced formation of HNO3 during this period (Pope et al., 2016) and also to import of particulate NO3through long-range transboundary transport, e.g. from continental Europe, as discussed in Vieno et al. (2014). Nevertheless, it is notable that the winter minima for NH4 + 40 aerosol concentrations at sheep and background sites are more pronounced than that for pig, poultry and cattle dominated sites.
This may be a result of a combination of smaller NH3 emissions in winter in these areas (as indicated by Figure 7a) and differences in long-range transport to the more remote areas in winter conditions. Overall, the seasonal distributions show that NH3 concentrations are mostly governed by local emission sources and by changes in environmental conditions, with warm, dry weather favouring increased volatilisation. By contrast, particulate NH4 + concentrations are largely determined by more distant sources through long-range transport and synoptic meteorology.

3.4
Long-term trends in estimated UK NH3 emissions 5 UK NH3 emissions are estimated to have fallen by 16 % between 1998 and 2014, from 336 to 281 kt (Figure 10a) (http://naei.defra.gov.uk/). The most significant cause of the estimated reductions has been decreasing cattle, pig and poultry numbers in the UK over this period. Between 2013 and 2014, the decreasing trend in UK NH3 emissions was however reversed with an increase of 3.3 % from 272 to 281 kt NH3 due to an increase in emissions from the agricultural sector from 224 kt in 2013 to 234 kt in 2014. This is attributed to an increase in dairy cow numbers (and dairy cow N excretion) and increase in 10 fertiliser N use (particularly urea, which is associated with a higher emission factor than other fertilisers types used in the UK) (Misselbrook et al. 2015; http://naei.defra.gov.uk/).

Long-term trends in measured NH3 concentrations
The UK NAMN dataset was analysed to compare levels and trends against the NH3 emission inventory. To avoid bias due to 35 changes in the number and locations of sites over the duration of the network, sites with incomplete data runs over selected periods for analysis are excluded. Based on these exclusion criteria, the number of sites with complete data runs was 59 for  Figure 11. This shows the mean UK monitored annual NH3 concentrations of 59 sites with complete data runs from 1998 (first complete year of monitoring) to 2014, summarised in a boxplot, together with annual mean UK rainfall and temperature data and compared with NH3 emissions trends over the same period. The interquartile ranges and the spread of the NH3 concentrations can be seen to be variable from year to year, demonstrating both substantial inter-and intra-annual variability. 10 <INSERT FIGURE 11>

Mann-Kendall non-parametric time series analysis
To detect trends and to indicate the significance level of the trends in the long-term NAMN data, the non-parametric Mann-15 Kendall (MK) approach was used combined with the Sen's slope method for estimating the trend and confidence interval of the linear trend (see Sect. 2.2.5). The classic MK test was used on the annually averaged data (datasets 1b, 2b, 3b), while both the classic MK and seasonal Mann-Kendall (SMK) tests were applied to the monthly averaged data (datasets 1a, 2a, 3a).
Results of the Mann-Kendall tests are summarised in Table 1. For each time series, the median annual trend (in units of µg 20 NH3 -1 y -1 ) is estimated from the Sen's slope and intercept of the MK linear trend. To assess the relative change over time, the % relative median change was calculated from the estimated NH3 concentration at the start (y0) and at the end (yi) of the selected time period (100*[(yi-y0) /y0]) computed from the Sen's slope and intercept. This approach was adopted instead of a direct comparison of actual observed NH3 concentrations at the start (y0) and at the end (yi) of the time series, since there is substantial inter-annual variability in the data ( Figure 10, Figure 16). Using the estimated concentrations at the start and end 25 from the fitted Sen's slope allows using a reference that is less sensitive to inter-annual variability than the actual observed concentrations.  (Table 1). Results from the analysis of monthly data from all three different data groupings (1b, 2b, 3b) (relative median change = 4.2 to 8.2 %) are similar to results for dataset 1a, based on analysis of annual data (Table 1). In the SMK tests on monthly data, two monthly "seasons" (January and April) in dataset 35 1b (1998-2014, 59 sites) are significant (p < 0.05) with a third monthly "season" (August) near-significant at p = 0.06. For datasets 2b (1999-2014, 66 sites) and 3b (2000-2014, 75 sites), August is the only monthly "season" in either time series to be close to significance at p = 0.06. Trends in individual monthly "seasons" are therefore weak and results between the MK and seasonal MK tests on monthly data are similar (Table 1).  The parametric linear regression time series trend analysis was also performed on the different data groupings. Results of the 5 linear regression tests are summarised in Table 2, and a comparison of trends from the Mann-Kendall with the linear regression approach is provided in Figure 12 for annual datasets 1a, 2a, 3a, and Figure 13 for monthly datasets 1b, 2b, 3b. A similar approach to the Mann-Kendall was taken to assess the relative change, by calculating the % relative change from the estimated NH3 concentration at the start (y0) and at the end (yi) of the time series (100*[(yi-y0) /y0]) computed from the linear regression slope and intercept. The different data groupings all show small, but non-significant decreasing trends (relative change = 2.4 10 % to 5.3 %), similar to the trends and % relative median change from the MK and SMK analysis (Figure 12, Figure 13). This suggests that the the errors in the NAMN data are normally distributed and that no or few outliers are present, since the results from the non-parametric Mann-Kendall are very similar to the parametric least squares linear regression.

Trends in NH3 concentrations vs trends in NH3 emissions
Overall, the long-term NH3 concentration data from the UK NAMN suggests evidence of a small, but non-significant decreasing trend (Figure 12, Figure 13). The level of reduction observed in the datasets is however less than the 16.3 %, 15.6 20 % and 13.1 % reduction in estimated UK NH3 emissions over the periods 1998-2014, 1999-2014 and 2000-2014, respectively (Tables 1,2). Inventories have inherent uncertainties such as uncertainties in activity data and emission factors, or may be missing emission sources. In terms of measurement data, it has already been shown in Sects. 3.1 and 3.3 that the annually averaged data mask considerable spatial and seasonal variability in NH3 concentrations. Drivers contributing to this variability include the influence of climate on emissions, variations in management practice for a particular emission source, and influence 25 of local emission sources and interactions on concentrations at a site. In addition, once emissions have taken place, the resulting atmospheric NH3 concentrations are influenced by local deposition, which is in turn affected by receptor surfaces and by concentrations of interacting chemical species that affect atmospheric lifetime and transport distance of NH3 and physical dispersion (e.g. Bleeker et al., 2009;Sutton et al., 2013). In the following sections, we consider the possibility of interactions with climate, emission source type and chemical interactions as this may affect long term trends in NH3 concentrations.  Figure S3). This analysis for the full network is therefore consistent with the observation at a remote site (Inverpolly, Figure 9). The thermodynamic equilibrium shifts NH3 from the aqueous (or particulate) phase to the gas phase with increased temperature, hence emissions from animal manures, soils and vegetation increase with increasing temperature (Asman et al., 1998;Sutton et al., 1993). Conversely, increases in precipitation decrease NH3 emissions because rain events dilute the available NH3 pool, while having the potential to wash urea and NHx in solution from the surface. As NH3 is soluble and washed out of the atmosphere by rainfall, this should also contribute to reduced NH3 concentrations during wet periods.
An exception to this relationship can occur where N is excreted as uric acid from birds (e.g. poultry). In this case, sufficient water is needed to allow hydrolysis to form NH3 (Riddick et al., 2014). In this situation, the arrival of rain promoted uric acid 5 hydrolysis from seabird guano surfaces, which was limited in the absence of soil moisture. It is possible that this interaction could lead to NH3 emissions from field spreading of poultry litter to be larger in wetter years. In a recent trend analysis of NH3 concentrations from the Dutch Air Quality Monitoring Network, an attempt was also made to correct for meteorological (temperature and rainfall) influences for the eight monitoring stations, which broadly produced similar results with slightly enhanced statistical significance for the trends (van Zanten et al., 2017). 10

3.5.5
Influence of local emission sources <INSERT FIGURE 14>

<INSERT FIGURE 15>
15 The inter-and intra-annual variability is also expected to be linked to influences from local emission source and activities. It  Table 3 and results of linear regression analysis are summarised in Table 4. A comparison of trends in measured NH3 concentrations with trends in NH3 emissions for the different source types then provided indicative evidence to support and inform the national emission inventory compilation. In Figure 16, the relative changes in UK emissions between 1998 and 2014 are compared with relative changes in mean measured NH3 concentrations for all NAMN sites, and for grouped sites 25 classified as dominated by cattle, pigs & poultry, and sheep. 3>   <INSERT TABLE 4>   30 For the 17 sites in cattle dominated areas, there is an increasing, but non-significant trend. Overall, based on MK analysis of annual data, the relative change from 1998 to 2014 is a 12 % increase (Table 3, Figure 14), compared with a smaller increase of 4 % from linear regression (Table 4, Figure 14). With the monthly data, there is no discernible trend (0.9 % (MK); 1.4 % (LR)). In the seasonal MK test on monthly data (% relative median change = 3.9 %), no monthly "seasons" are significant, with only January approaching significance at p = 0.07. The near-significant trend for January is likely to be due to unusually 35 high NH3 concentrations recorded in January at some sites in the first few months of the time series, attributed to manure spreading activities taking place in the winter months when the ground was frozen (confirmed by local observations), in direct contravention of good farming practice.

<INSERT TABLE
In terms of UK cattle NH3 emission, this has a decreasing trend with an estimated 11% decrease since 1998 ( Figure 16, Table  40 5), and is therefore clearly in contrast to the non-discernible or small increasing trend (non-significant) in NH3 concentrations from cattle sites. In principle, a signal related to substantial livestock changes associated with the 2000 outbreak of Foot and Mouth Disease might have been expected. However, this outbreak was actually rather localized in north-west England and south-west England, and was followed by substantial restocking from 2001 (Sutton et al., 2006) and there was no detectable signal of FMD in the average for cattle-dominated areas.
The SMK test also show a significant decreasing trend (11 %, overall p < 0.001), with 6 of the 12 monthly "seasons" showing significant trends (Feb, Jun, Nov, Dec: p <0.05, Oct: p < 0.01, Jan: p < 0.001). A decrease in emissions from pig and poultry of 39 % between 1998 and 2014 ( Figure 16, Table 5) is therefore broadly supported, although not matched by a similar decrease 15 in measured NH3 concentrations.
For sheep dominated sites (4 sites), there is an increasing trend in NH3 (MK: +16 %, p = 0.17, Table 3; LR: 20 %, p = 0.09, Table 4) between 1998 and 2014 in the annual data ( Figure 14). The monthly data also show a similar upward trend ( Figure   14) with relative change in concentrations of +19% based on MK (p = 0.10) (Table 3) and +17% based on LR (p = 0.14) ( Table  20 4). The increasing trend at sheep sites is therefore in contrast to the estimated 24 % decrease in NH3 emissions from this sector since 1998 ( Figure 16, Table 5). For the SMK test, no individual monthly "seasons" were significant, although 3 of the monthly "seasons" approached the significance level (Apr, Dec: p = 0.08, Oct: p = 0.09). Overall, the increasing trend from the SMK test is significant at p < 0.01. While the Sen's trend slope from both MK and SMK tests were comparable, at 0.0036 and 0.0033 µg NH3 y -1 , respectively, the % relative median change results computed from them are very different (MK = 16 % cf SMK 25 = 210 %), because the intercepts of the fitted Sen's trend slopes are different (MK = 0.289 µg NH3 m -3 cf SMK = 0.0267 µg NH3 m -3 ). Caution therefore needs to be exercised when interpreting the % relative change results, especially at sites with low NH3 concentrations, which must be examined together with the fitted trends.
At background sites (5 sites where total NH3 emissions for the respective 5 km grid squares are estimated at <1 kg N ha -1 y -1 ), 30 NH3 concentrations also appear to have increased (non-significant). Based on the MK analysis for the period 1998 to 2014, NH3 concentrations increased overall by 18 % and 13 % from the analysis of annual and monthly data, respectively (Table 3).
Results from linear regression were similar, with an overall increase of 13 % and 12 % from analysis of the annual and monthly data, respectively (Table 4). Similar to sheep sites, the % relative median change estimated from the seasonal MK Sen's slope and intercept (+ 49%) is larger than from the classic MK Sen's slope (+13%) due to differences in the intercepts of the fitted 35 trend lines (MK = 0.1528 µg NH3 m -3 cf SMK = 0.0388 µg NH3 m -3 ) since the trend slopes are the same (0.0012 µg NH3 y -1 ).
Overall, the SMK test show a significant increasing trend in the monthly data (p = 0.05). No individual monthly "seasons" were significant, with March, April and November monthly "seasons" approaching the significance level (p = 0.09).
As with the annual UK-wide long-term datasets (Sect. 3.5), it is useful to consider the significance of the NH3 trends for the 40 groupings of sites according to dominant emission source sectors. Table 3 and Table 4 shows that neither the annual nor the monthly time series showed a significant change in NH3 concentrations for the cattle dominated sites. In the case of pig and poultry dominated sites, the decrease in measured NH3 concentrations was significant for both the annual and monthly datasets.
For sheep dominated and backgrounds sites, the estimated increase in NH3 concentrations was not significant based on the MK and linear regression tests on the annual and monthly data, but was significant based on the SMK test of the monthly data.
Overall, these statistics confirm significant differences between NH3 trends for sites dominated by different source types, with concentrations decreasing at pig and poultry dominated sites, concentrations increasing at sheep dominated and background 5 sites, and no significant trend at cattle dominated sites (Table 5).

3.5.6
Changing chemical climate and effects on long-term trends in NH3 and NH4 + Other pollutants that affect NH3 concentrations in the atmosphere include SO2 and NOx emissions, which determine rates of secondary inorganic aerosol formation and therefore the lifetime of NH3 in the atmosphere. UK emissions of SO2 are estimated 10 to have declined significantly by 81 % from 1.6 million tonnes in 1998 to 0.3 million tonnes in 2014 (Defra, 2015). Similarly, NOx emissions over the same period are estimated to have fallen by 50 % from 2 million tonnes to 1 million tonnes (Defra, 2015). The reaction of NH3 with H2SO4 to form (NH4)2SO4 is effectively irreversible (in the absence of in-cloud reprocessing), whereas an equilibrium exists between gaseous NH3 and particulate NH4NO3 and NH4Cl components which are appreciably volatile at ambient temperatures. A change in the particulate phase from (NH4)2SO4 to NH4NO3 suggests that NH3 will remain 15 longer in the atmosphere, since NH4NO3 is volatile and releases NH3 in warm weather.
Elsewhere, a mismatch between reported trends in emissions and measurement data have similarly been investigated. The question of the 'Ammonia Gap' in the Netherlands was debated over a number of years. There, the estimated reduction in emissions due to mitigation measures was not matched by expected decreases in measured NH3 concentrations in air and/or 20 NH4 + in precipitation (Erisman et al., 2001;Bleeker et al., 2009;van Zanten et al., 2017). Similarly in Hungary, monitored NH3 concentrations from long-term measurements did not match the estimated reduction in NH3 emissions following the decline in agricultural livestock population and fertiliser usage after political changes in 1989 (Horvath and Sutton, 1998).
This was subsequently attributed to a reduction in SO2 emissions over the same period, increasing the atmospheric lifetime of NH3 (Horvath et al., 2009). 25 Dry deposition of SO2 and NH3 are enhanced in the presence of both gases, an interaction referred to as "co-deposition" (Fowler et al., 2001). The acid-base neutralization by each of the gases provides an efficient sink for dry deposition on leaf surfaces and deposition enhancement for each gas depends on the relative air concentrations of NH3 and SO2. For SO2, the dry deposition process has been shown to be strongly influenced by ambient concentrations of NH3 because the surface resistance 30 is regulated mainly by uptake in moisture on foliar surfaces, which, in turn, is strongly influenced by the presence of NH3. The large reduction in SO2 emissions and ambient concentrations, compared with the relative stagnation in NH3 emissions and concentrations over the same period has meant that the SO2/NH3 ratio has decreased dramatically. This has led to a systematic decrease in canopy resistance to uptake of SO2 on surfaces, increasing dry deposition of SO2 in the UK (ROTAP 2012). The underlying cause of the decrease in surface resistance is that the ambient NH3 is sufficient to neutralize acidity from the solution 35 and oxidation of deposited SO2, maintaining large rates of deposition.
Similar interactions are seen to be occurring in the UK based on the NAMN data, where the concurrent reduction in SO2 and NOx emissions over the same period (Figure 18b) should theoretically lead to a longer atmospheric lifetime of NH3, thereby increasing NH3 concentrations in the UK, especially in remote areas. The interpretation of the NH3 and NH4 + measurement data can further be aided by comparison with particulate nitrate (NO3 -) and sulphate (SO4 2-) data from the UK AGANet that 40 are made concurrently with the NAMN NH3 and NH4 + measurements at 30 sites (see Sect. 2.2). There is close agreement between the aerosol components, with a near 1:1 relationship between NH4 + and the sum of NO3and SO4 2-, lending support that particulate NH4 + in the UK is mainly derived from NH3 and acidic gases such as SO2 and NOx to form (NH4)2SO4 and NH4NO3, respectively (Conolly et al., 2016). For particulate NH4 + , it has already been shown in Sect. 3.3 that this regional species has less of a relationship to the dominant NH3 source sectors; trend analysis was therefore undertaken using all NH4 + site data combined. As with the NH3 time series analysis, sites with incomplete data runs for particulate NH4 + due to reduced density of NH4 + measurements and site changes occurring from 2001-2006 were excluded (see Sect. 2.2.1). 5

<INSERT TABLE 6>
Two data series for NAMN NH4 + data were selected for analysis, i) 23 sites with complete NH4 + time series from 1999 to 2014, and ii) 30 sites with complete NH4 + time series from 2006 to 2014. Both time series show a large significant downward 10 trend in NH4 + (p < 0.01) ( Table 6, Supp. Figure S4). Overall, MK and LR tests show a significant decrease in NH4 + concentrations by 47 % and 49 %, respectively, between 1999 and 2014 and by 44 % and 43 %, respectively, between 2006 and 2014 (Table 6, Supp. Figure S4). By contrast, concurrent NH3 data from the same sites over the same time periods showed a much smaller, non-significant downward trend between 1999 and 2014 (17 % (MK); 18 % (LR)), and no discernible trend between 2006 and 2014 (+ 3 % (MK and LR)) ( Table 6). This reduction in particulate NH4 + can be seen to be closely associated 15 with parallel decreases in particulate SO4 2and NO3concentrations from AGANet (Table 7, Figure 18a), which are themselves associated with reductions in SO2 and NOx emissions (Table 7, Figure 18b).

<INSERT TABLE 7>
The comparisons above therefore suggest that reductions in SO2 and NOx emissions over the period have led to a lower formation of particulate NH4 + in the atmosphere. Further evidence in support of this is indicated by plotting the ratio of NH3/NH4 + (Figure 17b), which has increased from 1.8 in 1999 to 2.8 in 2014. This demonstrates how a larger fraction of the 25 reduced N is staying in the gas phase as NH3, increasing its atmospheric residence time and maintaining NH3 concentrations at a higher level than solely based on NH3 emission trends. Although the overall changes in NH3 concentrations in the UK dataset are small and in many cases not significant for particular data groupings, they are consistent with similar phenomena observed in Hungary, the Netherlands and Denmark Erisman et al., 2001;Sutton et al., 2003;Bleeker et al., 2009). 30

CONCLUSIONS
Spatial and temporal trends in NH3 are found to be related to variability in emission source types across the UK and also to be influenced by changes in environmental conditions. Extensive spatial heterogeneity in NH3 concentrations was observed, with 35 lowest annual mean concentrations at remote sites (< 0.2 µg m -3 ) and highest in the areas with intensive agriculture (up to 22 µg m -3 ). NH4 + concentrations show less spatial variability (e.g. range of 0.14 to 1.8 g m -3 annual mean in 2005) with a general decreasing gradient from the south-east to the north-west of the UK, due to both regional differences in NH3 concentrations and import of particulate matter into south-east England from Europe.

40
Peak NH3 concentrations are observed in summer at background sites (defined by 5 km grid average NH3 emissions <1 kg N ha -1 y -1 ) and in areas dominated by sheep farming, driven by increased volatilization of NH3 in warmer summer temperatures.
In areas where cattle, pig and poultry farming is dominant, the largest NH3 concentrations are in spring and autumn, matching periods of manure application to fields. By contrast, peak concentrations of NH4 + aerosol occur in spring from long-range transboundary sources. The spatial and seasonal patterns established for sites influenced by different emission source sectors 5 are important for providing a foundation to understanding NH3 exchange processes, impacts and the UK NH3 budget, and to inform abatement strategies.
Official published estimates of UK NH3 emissions are estimated to have declined by 16.3 % between 1998 and 2014. The long-term NH3 concentration data from the UK NAMN suggests evidence of a smaller, but non-significant decreasing trend 10 (6.3 % (MK); 3.1 % (LR)), based on analysis of annually averaged data (n = 59) over the same period ( Table 2) (Table 2). 15 In areas with intensive pig and poultry farming, there is a significant downward trend in NH3 concentrations from the analysis of annually averaged data (22 % (MK), p = 0.02; 21 % (LR), p = 0.06) that is consistent with, but not as large as the decrease in estimated NH3 emissions from this sector over the same period (39 %) ( Table 5). By contrast, in cattle-dominated areas, there is evidence of a small increasing, but non-significant trend in NH3 concentrations (+12 % (MK); +3.6 % (LR): annually 20 averaged data), despite the decline in NH3 emissions from this sector since 1998 (11%) ( Table 5). At background and sheep dominated sites, NH3 concentrations increased (non-significant) over the monitoring period (Table 5). These increases in NH3 concentrations at background (+17 % (MK); +13 % (LR): annually averaged data) and sheep dominated sites (+15 % (MK); +19 % (LR): annually averaged data) are consistent with decreasing SO2 emissions (and to a lesser extent NOx emissions) associated with a change in the PM from (NH4)2SO4 to NH4NO3, the latter being volatile and releasing NH3 in warm weather. 25 Particulate NH4 + represents a secondary pollutant formed from NH3 and oxidation products of acidic gases such as SO2 and NOx. As the emissions of these acidic gases have reduced over the past years, the ratio between NH3 and NH4 + has increased from 1.8 to 2.8 between 1999 and 2014. These changes are consistent with observed decreases in particulate SO4 2and NO3concentrations that are associated with decline in SO2 and NOx emissions over the same period. This effect appears to be of 30 sufficient magnitude to explain the lack of overall decrease in NH3 concentrations, where the decrease in NH4 + is larger than for NH3 at corresponding sites. Overall, UK annual particulate NH4 + concentrations decreased by 47 % (MK) and 49 % (LR) for period 1999 -2014, associated with a slower formation of particulate NH4 + in the atmosphere from gas-phase NH3.
The findings are consistent with a parallel change in partitioning from particulate NH4 + to gaseous NH3 as also detected in Hungary, the Netherlands and Denmark. 35 Until now, only a modest commitment has been agreed to reduce European NH3 emissions. By contrast, SO2 and NOx emissions have decreased over Europe over the past decades, and are projected to decrease further under the revised Gothenburg Protocol and revised NECD. As a result, the importance of NH3 relative to oxidised N and SO2 emissions is expected to continue to increase over the next decades, playing a significant role in the formation of fine PM and contributing 40 to ecosystem effects through N deposition. With longer atmospheric lifetimes of gaseous NH3 and little commitment to reduce emissions, combined with climate warming effects tending to increase NH3 emissions, there is a substantial risk that exceedance of the NH3 critical levels may increase in the future, exacerbating the threat to the most sensitive semi-natural habitats. The growing relative importance of reduced nitrogen to total acidic and total nitrogen deposition indicates that future strategies to tackle acidification and eutrophication will need to include measures to abate emissions of NH3. Flechard, C. R., Massad,R. S.,Loubet,B.,Personne,E.,Simpson,D.,Bash,J. O.,Cooter,E. J.,Nemitz,E.,and Sutton,M. 25 A.: Advances in understanding, models and parameterizations of biosphere-atmosphere ammonia exchange, Biogeosciences, 10, 5183-5225, 10.5194/bg-10-5183-2013, 2013. Fournier, N., Pais, V. A., Sutton, M. A., Weston, K. J., Dragosits, U., Tang, S. Y., andAherne, J.: Parallelisation and application of a multi-layer atmospheric transport model to quantify dispersion and deposition of ammonia over the British Isles, Environmental Pollution, 116, 95-107, 10.1016/s0269-7491(01) 00146-4, 2002. 30 Fournier, N., Tang, Y. S., Dragosits, U., De Kluizenaar, Y., and Sutton, M. A.: Regional atmospheric budgets of reduced nitrogen over the British Isles assessed using a multi-layer atmospheric transport model, Water Air andSoil Pollution, 162, 331-351, 10.1007/s11270-005-7249-0, 2005. Fowler, D., Sutton, M. A., Smith, R. I., Pitcairn, C. E. R., Coyle, M., Campbell, G., and Stedman, J.: Regional mass budgets of oxidized and reduced nitrogen and their relative contribution to the nitrogen inputs of sensitive ecosystems, Nitrogen, the 35 Confer-N-S, edited by: VanderHoek, K. W., Erisman, J. W., Smeulders, S., Wisniewski, J. R., and Wisniewski, J., 337-342 pp., 1998. Tang, Y. S., Dragosits, U., Theobald, M. R., Fowler, D., and Sutton, M. A.: Sub-grid variability in ammonia concentrations and dry deposition in an upland landscape. In: Air surface exchange of gases and particles: Poster proceedings. (Eds. D. Fowler, C.E.R. Pitcairn, L. Douglas and J.W. Erisman) 48-57, 2001b. Tang, Y. S., and Sutton, M. A.: Quality management in the UK national ammonia monitoring network. In: Proceedings of the International Conference: QA/QC in the field of emission and air quality measurements: harmonization, standardization and 5 accreditation, held in Prague, 21-23 May 2003 (eds. Borowiak A., Hafkenscheid T., Saunders A. and Woods P.). European Commission, Ispra, Italy, 297-307, 2003. Tang, Y. S., Cape, J. N., Braban, C. F., Twigg, M. M., Poskitt, J., Jones, M. R., Rowland, P., Bentley, P., Hockenhull, K., Woods, C., Leaver, D., Simmons, I., van Dijk, N., Nemitz, E., and Sutton, M. A.: Development of a new model DELTA sampler and assessment of potential sampling artefacts in the UKEAP AGANet DELTA system: summary and technical report.  Atmospheric Environment, 38, 4045-4055, 10.1016/j.atmosenv.2004.03.051, 30 2004 van Zanten, M. C., Wichink Kruit, R. J., Hoogerbrugge, R., Van der Swaluw, E., and van Pul, W. A. J.: Trends in ammonia measurements in the Netherlands over the period , Atmospheric Environment, 148, 352-360, 10.1016/j.atmosenv.2016.11.007, 2017 1a, 2a, 3a and monthly mean datasets 1b, 2b, 3b) from the UK National Ammonia Monitoring Network (NAMN). The following are shown: the p-value, median annual trend (Sen's slope, in µg NH3 y -1 ) and the relative median change over the selected time period (in %). For the MK tests, the 95% confidence interval (CI) for the trend and relative change are also estimated.    c Relative change calculated based on the estimated annual NH3 concentration at the start (y0) and at the end (yi) of time series (=100*[(yi-y0) /y0]) computed from the slope and intercept (=100*[(yi-y0) /y0]). Table 3: Summary of Mann-Kendall (MK) and Seasonal Mann-Kendall (SMK) time series trend analysis on grouped NH3 concentration data (annually averaged and monthly mean data) from the UK National Ammonia Monitoring Network (NAMN) for four different emission source sectors. The following are shown: the p-value, median annual trend (Sen's slope, in µg NH3 y -1 ) and the relative median change over the selected time period (in %). For the MK tests, the 95% confidence interval (CI) for the trend and relative change are also estimated.