Seasonal Variations in the Synoptic Climatology of Air Pollution in Birmingham, United Kingdom

A synoptic typing approach was undertaken to examine the seasonal relationship (winter versus summer) between air mass types and pollutant concentrations of O 3 , PM10, NO x , NO 2 and CO in Birmingham, United Kingdom from 2000 to 2015. Daily means of seven surface meteorological variables were entered into a P-mode principal component analysis. Three principal components explained 72.2% (72.9%) of the variance in winter (summer). Cluster analysis was used to group together days with similar PC scores and thus homogeneous meteorological conditions. Six clusters provided the best air mass classication in both seasons. High pollutant concentrations were associated with anticyclonic types. In particular, tropical (polar) continental air mass type was most likely to produce extremely high concentrations in summer (winter). In winter, a sequence of Polar Continental (cool and humid) and Binary Mid-latitude Anticyclonic Maritime – Sub-Polar Cyclonic Maritime (cold and dry) induced severe pollution episodes in all pollutants. Whilst the mean duration of severe pollution episodes varied little between winter and summer (O 3 was an exception, with severe episodes lasting 20% longer in summer), high pollutant extremes were more common in winter. This was due to more favourable meteorological conditions (e.g., temperature inversions) and increased anthropogenic emissions during the cold season. with across all ve


Introduction
The Industrial Revolution brought considerable wealth and development to the coal-powered industrial towns of Britain, including the Black Country to the west of Birmingham and wider West Midlands area. Economic development came at a cost, with nitrogen oxides (NO x ), carbon monoxide (CO) and ozone (O 3 ) emissions in the Black Country being 50 times higher than present day concentrations (Hatton, 2017). The considerable epidemiological impacts led to the Clean Air Acts of 1956, 1968 and 1993 being passed to regulate domestic, commercial and industrial emissions (Abbott et al., 2012). Birmingham is Britain's second largest city, with a population of 1.22 million. Multiple industrial sources, a major airport and congested roads result in high pollutant emissions. In particular, combustion from diesel engines accounts for the majority of the city's NO x emissions. The Environment Act 1995 produced a national air quality strategy, which gave jurisdiction to local authorities such as Birmingham City Council to review and tackle their air pollution problem (DEFRA, 2018). In 2003, Birmingham was designated as an Air Quality Management Area (AQMA), in which pollutant concentrations must be regularly monitored and policies enacted to reduce levels of harmful pollutants such as NO 2 and PM10. Despite this, the concentration of NO 2 in Birmingham is up to 50% higher than recommended levels (DEFRA and DfT, 2017). Bloss (2018) reported that poor air quality causes 900 premature deaths in Birmingham every year. In order to satisfy annual and winter pollutant targets, Birmingham City Council are introducing a £8 charge for older private cars to enter the city centre from June 2021.
Since the 1980s a growing body of research referred to as the 'Delaware Approach' (Kalkstein and Corrigan, 1986) has established that the physical and thermodynamic properties of the urban boundary layer play a primary role in in uencing the dispersion of pollutants (McGregor and Bamzelis, 1995; Shahgedanova et al., 1998). This synoptic typing Delaware Approach adopts an air mass approach to understanding the relationship between weather and pollution. The advantage of such an approach is that it considers the synergistic behaviour of a multitude of atmospheric variables rather than analysing variables such as temperature and humidity in isolation (Tselepidaki et al., 1995;McGregor, 1999). Crucially, the grouping of multiple surface weather variables in the air mass approach allows one to infer whether conditions are likely to favour pollutant accumulation or suppression. Principal component analysis (PCA) and cluster analysis (CA) are often used to derive the air mass types. PCA identi es groups of highly correlated variables over time. The resulting principal components are linearly independent and explain the maximum amount of variance possible (Yarnal, 1993). Many researchers then cluster the PC scores to identify the air mass types. Studies that have used the PCA and CA approach to synoptic typing include McGregor and Bamzelis in Birmingham, UK (1995) and Shahgedanova et al. (1998) in Moscow. Cheng and Lam (2000) found that air mass types with low wind speeds, low temperatures and anticyclonic conditions were conducive to extremely high CO concentrations in Hong Kong. Kalkstein and Corrigan (1986) showed that maritime air masses -characterised by low pressure and high winds -resulted in low SO 2 concentrations in Delaware, US due to effective mixing and ventilation. Similarly, CO concentrations halved with maritime polar air masses in Moscow (Shahgedanova et al., 1998).
Many studies have acknowledged the tendency of urban air pollutants to persist (Chen et al., 2016). This persistence is a complex phenomenon governed by many inter-related factors. Chelani (2013) showed that CO persistence was dependent on preceding concentrations. Ozone persistence is not dependent on antecedent levels; but is primarily dependent on NO x and CO emissions as precursor sources, and strong sunlight (Varostos et al., 2003). Lonati et al. (2011) found that stable anticyclonic air mass types led to persistently high PM10 concentrations in Milan, Italy. The implications of severe PM10 episodes for daily hospital respiratory admissions in Birmingham were explored by McGregor et al. (1999). Xie et al. (2015) found that a 10 µgm -3 increase in PM2.5 was associated with a 0.27% increase in ischaemic heart disease mortality.
Despite the well-established nature of the weather-pollution relationship, seasonal differences in this association have received comparatively little attention. For instance, McGregor et al.'s (1999) application of the air mass type to Birmingham considered the winter only. A winter-only study might not be generalisable to other seasons, given the large seasonal variations that exist in weather conditions and pollutant emissions. Domestic energy use increases signi cantly in colder winter months, with people more likely to use their cars for shorter journeys. This paper uses a synoptic typing approach to derive air mass types for both winter and summer in Birmingham, so enabling a comparative analysis of the meteorological controls on pollutants between the two seasons to be conducted. The objectives of this paper are as follows: (i) to derive winter and summer air mass types for Birmingham; (ii) to assess whether a relationship exists between the air mass types and mean and extreme pollutant concentrations; and (iii) to examine the seasonal differences in the above relationships This paper is structured as follows. Section two describes the collection and analyses of data. Results are described in section three and then discussed in section four. Conclusions are drawn in section ve.

Data And Methodology
Seven meteorological variables were obtained from the British Atmospheric Data Centre (BADC) for Coleshill (10km east of Birmingham City Centre): mean sea level pressure (MSLP), air temperature (TEMP), wind speed, wind direction, relative humidity (RH), cloud cover (CLOUD) and rainfall (RAIN). Hourly values of these variables were abstracted for winters (DJF) and summers (JJA) from 2000 to 2015, and used to calculate daily means. The sample sizes were 1,354 and 1,380 days for winter and summer respectively. Wind direction cannot be entered into the PCA because it is not on a linear scale. To solve this problem, wind speed and direction were used to calculate the zonal (U) and meridional (V) wind components, which were then entered into the PCA.
Daily mean concentrations of O 3 , PM10, NO x , NO 2 and CO were abstracted from DEFRA's air quality archive. Unfortunately, data were not available for one site for the complete 2000-2015 period, so two sites had to be used: Birmingham Centre (2000-4) and Tyburn . It was decided that Birmingham Tyburn (4.5 km north east of the City Centre) would make up the majority of the dataset given its closer proximity to the meteorological station at Coleshill. The distance between Coleshill and Tyburn (8 km) is acceptable here because air masses typically cover areas greater than hundreds of square kilometres (Kalkstein and Corrigan, 1986). The Tyburn readings are taken next to a busy arterial road and so are signi cantly higher than readings at Birmingham Centre.
In order to produce a homogeneous time series, regression models were used to calibrate the 2000-4 values at Birmingham Centre to those at Tyburn (table 1). The regression equations were also used to in ll any missing values at Tyburn from the other site. Calibration of CO was not performed because it was only measured at Birmingham Centre until 2008.
Daily means of MSLP, TEMP, U, V, RH, CLOUD and log 10 RAIN (dry days were replaced with 0.01mm) were entered into separate P-mode PCAs for winter and summer. The correlation matrix was used because it standardises the variance of each variable to unity, thus eliminating the problem of different measurement scales. It is desirable for each variable to load strongly on to one principal component (PC). This is called simple structure. In the winter PCA, MSLP, TEMP and CLOUD were complex variables; and loaded strongly on to multiple PCs. An orthogonal Varimax rotation was thus applied to the winter PC axes in order to achieve better simple structure. No complex variables were identi ed for the summer PCA, which meant that rotation was unnecessary. PCs with eigenvalues greater than one were retained because these explain more variance than one original variable (Kaiser, 1960).
Cluster analysis was used to group together days with similar PC scores and thus homogeneous meteorological conditions. Again, separate cluster analyses were run for winter and summer using the group average method. Whilst Ward's method is most popular, it priorities equal cluster sizes at the expense of physical air mass interpretation by grouping extreme days with non-extreme days (Kalkstein et al., 1987). By contrast, the group average method focuses on 100% likeness of adjoining cluster cases by prioritising minimisation of within group variance and maximisation of between group variance (Shahgedanova et al., 1998). The squared Euclidean distance was used as the similarity measure. To determine the number of clusters in both seasons, a dendrogram and screeplotted agglomeration schedule were used, the latter of which plots the co-e cient of fusion against the number of clusters (Gower and Barn eld, 1975).
These methods allow the researcher to deduce signi cant between-cluster heterogeneity and thus unique air mass types.
An independent samples one-way analysis of variance (ANOVA) was conducted for winter and summer to investigate whether pollutant concentrations vary between the air mass types. Logged values of NO x , PM10 and CO were utilised to satisfy the normality assumption of ANOVA. Whilst ANOVA reveals whether the air masses are different as a whole, it does not show whether one air mass shows a statistically signi cant difference from the other air masses. As such, a Tukey test was performed, in which signi cant differences in pollutant concentrations between pairs of air masses were identi ed.
For each pollutant and for both summer and winter datasets separately, a Chi-Square test was used to determine whether the most polluted days (standardised scores > 2) are disproportionately associated with certain air mass types. The null hypothesis (H o ) of the test assumes that the number of most polluted days is proportional to the frequency of occurrence of that air mass type. For example, if an air mass occurred on 20% of days throughout the entire data set, then H o assumes that this air mass accounts for 20% of the most polluted days. The Chi-Square test was also used to ascertain whether three thresholds (high, very high and exceptionally high; see table 2) for poor air quality were more likely to be exceeded in summer or winter.
A severe pollution episode was de ned as a sequence of at least four days with concentrations on all days above the 90th percentile. The dominant air mass type for each episode was noted.

Principal component analysis
For the winter, three PCs explain 72.2% of the variance. PC 1 (explained variance = 26.0%) has a strong positive association with log 10 RAIN and V; and a strong negative relationship with MSLP (table 3). PC 1 captures the dominant mode of variability in winter weather conditions, which tend either to be mild and wet with low pressure (high positive PC 1 scores) or cold and dry with high pressure (high negative PC 1 scores). RH and CLOUD have high positive loadings on PC 2 (explained variance = 24.5%); U is the only variable with a strong negative PC 2 loading, which suggests that increased RH and CLOUD are associated with easterly ows. This is logical because easterly winds are cold in winter, which results in a higher RH and often overcast conditions. By contrast, westerly winds are often accompanied with clearer skies. PC 2 can thus be interpreted as the state of the sky. PC 3 (explained variance = 21.8%) is a thermal component. It has a strong positive relationship with TEMP and CLOUD; and a weaker positive relationship with V, U and log 10 RAIN. On days with high positive PC 3 scores, temperatures are higher than average, with moderate south-westerlies, extensive cloud cover and drizzle. Days with high negative PC 3 scores are much cooler and clearer.
Three PCs explain 72.9% of the variance during the summer. PC 1 (explained variance = 39.2%) has a strong positive relationship with log 10 RAIN, RH and CLOUD; and a strong negative association with MSLP. On days with high positive PC1 scores, low pressure areas will be present over Britain, bringing cloud and signi cant rainfall. Days with high negative PC 1 scores are characterised by anticyclonic conditions, with warmer temperatures and blue skies. PC 1 is thus a hygrometric and pressure component. PC 2 (explained variance = 18%) is capturing the strong co-variability between south-easterly air ows from Continental Europe and warmer temperatures. PC 3 (explained variance = 15.7%) has a strong positive association with U, and a moderate inverse association with RH and MSLP. Days with high positive PC 3 scores have strong westerly winds and moderately low pressure, in which depressions driven by strong Atlantic westerlies travel eastwards over Britain. On days with high negative PC 3 scores, the wind tends to blow from the east and MSLP is higher. PC 3 is thus a zonal wind component.

Winter air mass types
For the winter, the dendrogram and agglomeration schedule indicate that the retention of six clusters is most appropriate. The properties of each air mass type are now described ( gure 1 and table 4).

Type 1: Anticyclonic Tropical Maritime
With south-westerly winds, conditions are relatively mild with high cloud cover.
Above average pressure suppresses the amount of rainfall. Typically associated with the passage of warm fronts, although cold and occluded fronts can on occasions bring heavier rainfall to the English Midlands. High pressure is positioned over France and Iberia, with depressions moving eastwards between Scotland and Iceland.

Type 2: Unstable Tropical Maritime
The wettest cluster with the lowest pressure. Birmingham is affected by deep Atlantic lows, cold fronts and strong south-westerly winds.

Type 3: Polar Continental
Much cooler than the rst two air mass types, and so has a higher relative humidity. The wind direction is variable, but always from a cold, continental source. Pressure is usually high; rainfall is moderate, with occasional heavy downpours. Nocturnal radiative cooling brings morning fog, which dissipates by early afternoon.
Type 4: Binary mid-latitude anticyclonic maritime -sub-polar cyclonic maritime (AM-SCM) (McGregor, 1999) Northerly winds drive very cold conditions, with low relative humidity and cloud cover. Typically features a well-developed mid-Atlantic anticyclone, with a depression of moderate intensity lying between Iceland and Scandinavia. Rainfall decreases as the mid-Atlantic high moves towards Britain.

Type 5: Extreme Arctic Maritime
Occurs on only three days. Characterised by north-westerly winds, which bring bitterly cold temperatures.

Type 6: Returning Polar Maritime
This air mass is characterised by fairly mild temperatures, pressure close to the long-term average, strong westerly winds, and moderate cloud cover and rainfall. Depressions track eastwards between Iceland and Norway. When the isobars are traced back into the western Atlantic, they originate from the Greenland and Canada region.

Summer air mass types
Six clusters were again considered to provide the best air mass classi cation. The properties of each air mass type are now described ( gure 2 and table 5).

Type 1: Tropical Maritime
Predominantly wet and cloudy, with moderate temperatures and strong south-westerlies. This air mass typically originates from the Azores and contains active frontal systems. Rainfall was as high as 33.6mm on one day. Type 2: Transitional Polar Arctic Low pressure centred over southern England or northern France leads to cold north-easterly winds transporting Scandinavian or Arctic air to Birmingham. This is by far the wettest air mass type. Anticyclones in the mid-Atlantic and Eastern Europe are common in this situation.

Type 3: Stable Anticyclonic Tropical
Characterised by moderate westerlies, temperatures and cloud cover. Whilst the air has a long sea track, high pressure over southern Britain is the dominant feature and results in mainly dry conditions.

Type 4: Mild Stable Polar Continental
High pressure and north-easterly winds usually bring low rainfall. Signi cant low-level cloud cover brings morning fog, with potential for heavy rain when there are disturbances embedded in the ow and a longer North Sea fetch.

Type 5: Tropical Continental
The driest and warmest air mass type, with low cloud cover and a gentle south-easterly breeze. High pressure centred in the southern North Sea leads to settled conditions.

Type 6: Polar Maritime
This cluster contains only 16 days. Temperatures are on the cool side, with light to moderate rainfall. A low is typically situated over north-west Scotland, which brings strong westerlies to Britain.

Pollutant variability with air mass type
All ve pollutants show statistically signi cant differences between the six air mass types at the 99.9% con dence level in both winter and summer (table 6). As Snedecor's variance ratio F is considerably greater than one, there is far more variance in pollutant concentrations between the air mass types than within types. For all ve pollutants, the F ratio is larger in the winter. This means that concentrations vary more with air mass type in winter than in summer.
Difference matrices are used to represent the results of the post-hoc Tukey tests. In these matrices, matrix cell entry denotes a statistically signi cant difference in the concentration of the given pollutant at the 99% con dence level.
In the winter ( gure 3), NO 2 and O 3 have the most signi cant pairings with 12 and 10 respectively whilst PM10 has the fewest at six. Returning Polar Maritime (air mass type six) is responsible for the largest number of signi cant pairings (20), with Extreme Arctic Maritime (type ve) having the least pairings (eight).
During the summer ( gure 4), PM10 and CO have the largest and smallest number of statistically signi cant pairings at 11 and six respectively. The Transitional Polar Arctic (type two), Stable Dry Anticyclonic Tropical (type three) and Tropical Continental (type ve) all have the largest number of signi cant pairings (16). This means that these three air masses are the most distinctive in terms of concentrations. In particular, air mass type three shows statistically signi cant differences at the 99% con dence level with types two and four across all ve pollutants.

Distribution of air mass types on the most polluted days
Winter air mass type ve was excluded from the Chi-Square test because it occurred on only three days and so violated the 20% rule for expected counts. The test results show that the most polluted days in winter (Z score > 2) show a signi cant association with the air mass type at the 99.9% con dence level (table 7 and gure 5). All pollutants show a statistically signi cant relationship with summer air mass type at the 0.01 level or better (table 7). In particular, high NO x , PM10 and CO concentrations occur disproportionately under summer air mass type three, accounting for 49%, 51% and 65% of high concentration days respectively ( gure 6). Air mass type six is not present for any pollutant except O 3 .

Severe pollution episodes
A severe pollution episode was de ned as a sequence of at least four days with concentrations above the 90th percentile.  concentrations for all pollutants are higher in the winter. The exception is O 3 , which has greater concentrations and longer episode duration in summer. This is to be expected given that increased summer sunshine hours are conducive to increased photolysis. NO x has much higher concentrations in winter than summer, whilst PM10, CO and NO 2 are only slightly higher.

Cross-seasonal comparison of pollution extreme values
The thresholds for high, very high and exceptionally high concentrations are de ned in table 2. The Chi-Square test is applied to determine whether these thresholds are more likely to be exceeded in winter or summer. Apart from O 3 (p = 0.024), the pollutants all show a statistically signi cant difference in their frequency of occurrence between winter and summer at the 99.9% con dence level ( gure 10). These high pollutant thresholds are far more likely to be exceeded in winter than in summer, with the exception of O 3 . There were 317 days in winter when NO 2 exceeded 50 µgm -3 but only 18 such days in summer.

Discussion
Principal component analysis The ANOVA (table 6) and post-hoc Tukey test ( gure 4) results showed statistically signi cant pollutant concentration differences between summer air mass types. These differences disproportionately occur when Tropical Continental (AT 5), which has the highest concentration for three of the ve pollutants, is compared to the other ve air mass types. Tropical continental air masses are typically characterised by anticyclonic conditions that are conducive to low cloud cover, low wind speeds and atmospheric stability (Comrie, 1992). This limits the ventilation and dispersion of pollutants because advective mixing is suppressed and a strong atmospheric subsidence exists, resulting in anomalously high pollutant concentrations. Cheng and Lam (2000) found that SO 2 concentrations in Hong Kong increased six-fold from cyclonic to anticyclonic air mass types; in this study, mean NO x in Birmingham under the Tropical Continental air mass is almost double that of the Polar Maritime air mass type. The passage of the Tropical Continental air mass over Europe's industrial and urban centres is also conducive to high pollutant concentrations.

This study indicated a signi cant difference in O 3 concentrations between air mass types in both winter and summer. Winter Returning Polar Maritime (AT 6)
and summer Tropical Continental (AT 5) possessed the highest mean concentrations of 55.0 µgm -3 and 60.7 µgm -3 respectively. These air mass types are characterised by low cloud cover. The resulting increase in radiation causes NO 2 to dissociate NO 2 into NO + O, which are the precursors of O 3 formation via photolysis (McGregor and Bamzelis, 1995). This link can be corroborated by the low NO 2 concentrations that occur in air mass types associated with less cloud cover; e.g., winter AT 6 mean = 24.4 µgm -3 .

Distribution of air mass types on the most polluted days
In winter, Polar Continental (AT 3) and AM-SCM (AT 4) occurred more frequently than expected on the most polluted days. These air masses are characterised by cold, anticyclonic conditions, with air originating from the north and the east. These conditions are conducive to temperature inversions and therefore pollutant stagnation. This supports the ndings of McGregor (1999), which found that the sub-polar continental anticyclonic air mass type (similar to this study's Polar Continental) was disproportionately represented on the most polluted days. Conversely, Unstable Tropical Maritime (AT 2) is under-represented on days with high concentrations.
During the summer, Stable Anticyclonic Tropical (AT 3) and Tropical Continental (AT 5) disproportionately occur on the most polluted days for all pollutants.
These air mass types are characterised by warm to hot temperatures and light wind conditions. A neutral to stable atmosphere suppresses vertical motion by impeding buoyancy and turbulent eddy motions. This leads to poor ventilation and reduced dispersion of pollutants (Cheng and Lam, 2000).

Severe pollution episodes
In winter, anticyclonic air masses Polar Continental (AT 3) and AM-SCM (AT 4) were found to be most favourable to severe winter pollution episodes. Crucially, McGregor et al. (1999) found that these two air mass types tend to occur together in a sequence during the winters of 1988-94 in Birmingham. Elevated respiratory response to pollutants can occur due to the changing nature of persistent winter anticyclonic weather from ne, cold and dry (AM-SCM) to overcast, cool and humid conditions (Polar Continental) advected from the north-east. Episodes of the AM-SCM air mass type creates a 'sensitising mechanism' by which cold, stable conditions occur in conjunction with elevated pollutant concentrations to prime the population's respiratory response, which is activated by the persistence of cool, moist ows of the Polar Continental air mass. While surface temperature inversions are a primary contributor to severe pollution days, inversions of nocturnal origin typically last only six to 12 hours; and are thus rarely a primary causal mechanism of winter severe pollution episodes (Kassomenos et al., 2007). High concentrations are unlikely to persist under the Tropical Maritime (AT 2) and Returning Polar Maritime (AT 6) air mass types, which proves that the advection of strong Atlantic westerlies disperses pollutants. summer. This corroborates the ndings of Widawski (2015), who showed that a marked temperature inversion in Warsaw (Poland) coincided with the highest levels of NO x . Increased levels of NO 2 and CO in winter months can also be attributed to their signi cantly reduced in uence on photochemical reactions occurring in photolysis for O 3 formation (Cichowicz et al., 2017). The meteorological contribution to signi cantly higher NO x concentrations in winter may be of secondary importance. A three-fold increase in winter cloud cover would be expected to decrease photolysis such that the reduced reaction of NO 2 and sunlight should directly decrease NO x production. Thus, the possible anthropogenic contributions to seasonal differences will now be explored. to winter. During the winter, the use of electricity, central heating and gasoline (mainly vehicle engines) increases greatly because of the colder conditions. Gavin (2014) reported that electricity demand increased by 23% during the winter in the West Midlands. Moreover, fewer people walk and cycle in the winter and instead take to pollutant-producing vehicle transport (Yang et al., 2011). It follows that increased winter anthropogenic heat emissions in cities result in CO, PM10 and especially NO x being released at a greater rate.
Pollutant concentrations found in this paper often exceed EC (2019) and WHO (2005) guidelines, especially in winter. This has signi cant implications for human health, given that those with respiratory diseases are at considerable risk during elevated concentrations of pollutants such as NO 2 and PM10. McGregor et al. (1999) demonstrated the relationship between respiratory-induced hospital admissions in Birmingham and air mass types. Their study revealed that PM10 concentrations exceeded the 50 µgm -3 threshold 22% of the time during anticyclonic polar continental air mass types, which correlated signi cantly with higher respiratory admissions. Pollutant concentrations also play a role in determining heart disease mortality rates in Birmingham (McGregor, 1999).

Conclusion
This study's primary aim was to examine whether a relationship exists between pollution and weather types, and whether this association varies seasonally.
Principal component and cluster analyses were used to derive the air mass types, which were then related to pollutant concentrations. High concentrations are associated with anticyclonic types. In summer, such air mass types are characterised by low wind speeds, which leads to decreased advective mixing. During the winter, similar meteorological conditions are often combined with nocturnal radiative cooling. This creates a temperature inversion near the ground, which also suppresses the vertical dispersion of pollutants.
The tropical (polar) continental air mass type was most conducive to exceptionally high and very high concentrations in summer (winter). Turbulence is suppressed and the atmosphere is stable in these air mass types. In winter, a sequence of Polar Continental (cool and humid) and Binary Mid-latitude Anticyclonic Maritime -Sub-polar Cyclonic Maritime (cold and dry) induced severe pollution episodes in all pollutants, which can elevate the respiratory response in the general population by creating a sensitising mechanism.
The mean duration of a severe pollution episode duration varied little between winter and summer due to the strong presence of anticyclonic air mass types in both seasons. Naturally, increased photolysis results in severe O 3 episodes lasting 20% longer in summer than in winter. With the exception of O 3 , high pollutant extremes were more common in winter. This was due to more favourable meteorological conditions and increased anthropogenic emissions during the cold season. Epidemiological evidence indicates that such high concentrations (e.g., one quarter of winter days exceeded the WHO and EC threshold for NO 2 ) can exasperate heart and respiratory diseases in Birmingham.
The main limitation of our paper is that it has analysed weather -pollution relationships at the daily timescale using the mean value of 24 hourly observations. In mobile situations, it is possible for multiple air mass types to occur on the same day. There was also an imperfection in the method used to calculate the wind component vectors U and V, which misrepresented some days' wind ow. For example, on the 23rd January, 2001, the wind direction was averaged as an easterly despite a ve-hour 19mm rainfall event occurring under a strong westerly ow. Thus, the statistical analyses were unable to identify this rainfall event as being a westerly-induced phenomenon. However, a complete reversal in the wind direction (e.g., westerly to easterly; southerly to northerly) rarely occurs at the sub-daily timescale, which means that the above problem does not occur frequently. Some researchers (e.g., Cheng and Lam, 2000) overcame the problem of intra-diurnal variability by using six hourly averages. Whilst our paper uses 15 years of data, it is acknowledged that using linear regression to correct the values at Birmingham Centre to those at Tyburn between 2000-4 might have stripped out potentially extreme values at the latter site over this time period. This is because the regression line's position is constrained by having to pass through the bi-variate mean centre.
Future research should replicate the principal component and cluster analyses presented in this paper at the sub-daily timescale (e.g., six or 12-hourly means).
Diurnal variations in pollutant concentrations would rstly have to be removed (e.g., calculate hourly standardised scores) before these sub-daily air mass types could be related to air quality data. Another avenue of further research could compare pollutant-air mass type relationships to health indicators, such as respiratory admissions or ischaemic heart disease data. Multi-variate analysis of variance could be used to test statistically the tripartite relationship between air mass types, pollution and concurrent health impacts rather than simply inferring it. Such research could be an invaluable device for authorities to forecast scenarios of concern to public health. This could be communicated to the public by means of a synoptic index. This paper has only used surface meteorological variables. Variables that capture vertical atmospheric stability (e.g., vertical temperature gradient) could be added to the analysis.
This study has clearly shown that an objective, synoptic airmass approach can be used to distinguish comprehensively the pollutant controls of different weather types. Given that pollutant guidelines were often breached during the study period, a concerted effort in the future to understand the complex relationship between meteorological conditions and anthropogenic-pollution relationship may have crucial epidemiological and societal implications.

Ethics approval
The submitted work is original and has not been published elsewhere in any form or language. Consent to participate Air mass type pollution difference matrix for the summer. Pollutants that appear in the matrix show a statistically signi cant difference at the 99% con dence level between the pair of air mass types shown.

Figure 5
Comparison between the percentage occurrence of the air mass types across the whole winter and the percentage occurrence of days with high pollutant concentrations Figure 6 Comparison between the percentage occurrence of the air mass types across the whole summer and the percentage occurrence of days with high pollutant concentrations Figure 7 Number of severe winter pollution episodes per air mass and the relative dominance of that air mass in the episodes. The green line represents the percentage duration of pollution episodes in which the dominant air mass type (AT) is present.  The arithmetic mean length in days of a severe pollution episode in summer and winter (bars), overlaid by their average concentrations (lines).