Previous air pollution research in Christchurch has involved case studies, such as the Christchurch Air Pollution Study in 2000 (Kossmann and Sturman, 2004). To understand the physical nature of meteorological controls of air pollution, all space and time scales need to be considered. As discussed by Lovejoy et al. (2009) and Paradisi et al. (2009), atmospheric processes are closely linked in the spatial and temporal domain. Therefore, local to regional processes operating on daily time scales can be expected to be closely linked to processes at much larger synoptic to hemispheric scales that operate on seasonal to interdecadal time scales. Local meteorological processes that underlie the Christchurch air pollution problem have been previously investigated in some depth (APAC, 1959; Pullen, 1970; Kossmann and Sturman, 2004; McKendry et al., 2004; Corsmeier et al., 2006). However, there is a clear lack of analysis of synoptic climatological influences on air quality in Christchurch. Internationally, many studies have investigated synoptic controls on air quality, for example, Sanchez et al. (1990), Yap and Chung (1977), Heidorn and Yap (1986), McGregor and Bamzelis (1995), O'Hare and Wilby (1995), and more recently Makra et al. (2009), Kuo et al. (2008) and Cheng et al. (2007). For New Zealand, however, investigation of synoptic forcing of air pollution meteorology is sparse. Recently, Khan et al. (2007) studied the relationship between synoptic weather situations and night time ozone (O3) and nitrogen oxides (NOx) concentrations in Auckland. They showed that anticyclonic events influence the variation in concentrations of both NOx and O3. Similarly, Jiang et al. (2005) have related synoptic weather types and winter time NOx concentrations during morning rush hour traffic in Auckland and also found anticyclonic conditions to be most influential. Owens and Tapper (1977) provided a brief statistical investigation of both local and synoptic influences on air quality in Christchurch. However, no other investigations of the influence of meteorology at the synoptic or larger spatial and temporal scales on local air quality have been reported for Christchurch, which provides the opportunity to apply some new approaches to a longstanding environmental problem.
1.1. Air pollution in Christchurch
The city of Christchurch, located on the east coast of New Zealand's South Island, has an air pollution problem that has long been recognized (Gray, 1889; Pullen, 1970; APAC, 1959). According to Statistics New Zealand 2006 census data, approximately 350 000 people reside within the Christchurch urban area (CCC, 2007). The pollutant of most concern, by far, in present-day Christchurch is PM10. PM2.5 also shows enhanced levels during winter, but the scarcity of available monitoring data prohibits climatological analysis at this point in time (Aberkane et al., 2006). Carbon monoxide is still of some concern, and in the past, sulphur dioxide, released mainly through the burning of coal, contributed to the problem. NOx and O3 are of minor importance, mostly due to Christchurch's low population and infrastructural density. Poorly insulated dwellings, together with traditional home heating practices such as coal and wood burning, lead to the release of particulate matter into the urban atmosphere, especially during winter months. Associated health problems for the urban population are common (Wilton, 2001) and hospital admissions are increased during the cold season (McGowan et al., 2002; Hales et al., 2000).
Even though air quality management campaigns started in the 1930s (Wilton and Ayrey, 2002), continuous air quality monitoring in Christchurch has only been implemented since the late 1980s, while quality assured and reliable air quality data are only available from the late 1990s (T. Aberkane pers. comm.). Therefore, air quality data sets in Christchurch have only recently accumulated to the extent that statistically robust time series analysis is possible. New Zealand's National Environmental Standards (NES) require that no more than three exceedances of 50 µg m−3 for PM10 (averaged over the 24-h period between midnight and midnight) be observed in the Christchurch urban airshed per year by 1 September 2016 (Ministry for the Environment, 2011), and no more than one exceedance by 1 September 2020. To put this into perspective, between 1999 and 2006, the NES was exceeded on 25–30 days each year, and these exceedances mainly occurred during the 123-day winter period between May and August. Various measures to reduce anthropogenic emissions have therefore been implemented in recent years and efforts have been undertaken to quantify their impact (Bluett et al., 2007; Scarrott et al., 2009; Appelhans et al., 2010).
1.2. Meteorological effects
A complex pattern of low-level local airflow (Kossmann and Sturman, 2004) associated with highly stable and stagnant near-surface atmospheric conditions allows the emitted particulates to accumulate to high concentrations and leads to reduced air quality in the city. Even though synoptic flow in the region is dominated by winds from westerly directions, surface winds that affect Canterbury during the day are usually dominated by northeasterly directions. The ‘Canterbury North-Easter’ is largely the result of a combination of the sea-breeze circulation and orographic modification of synoptic westerly to northwesterly flow over the Southern Alps that produces a lee side trough east of the mountains (McKendry et al.1986, 1987). During night time, northeasterly winds are less frequent and replaced by weaker low-level winds from the southwest and west. Figure 1 shows the low-level wind field as measured at Christchurch airport (Figure 2) on a seasonal basis, highlighting the dominance of the daytime northeasterly regime, especially during the summer season (DJF). As particulate pollution is largely a winter time problem, the season of interest for this research is JJA. During this season, northeasterly frequency is decreased by almost half (in comparison with DJF) and daytime wind speeds are reduced. This is testament to the fact that reduced thermal forcing, as a result of shorter days and a higher solar zenith angle, decreases the sea-breeze component of the Canterbury North-Easter. This decrease is compensated by an increase in westerly to southwesterly flow, especially during the night. The increased nocturnal westerly flow component stems from enhanced cold air drainage from the Southern Alps and Canterbury Plains over the city. The pattern of diurnal flow reversal in the area has been identified as a major mesoscale driver for air pollution events in Christchurch in previous investigations (Kossmann and Sturman, 2004, Corsmeier et al., 2006). This study extends this research by identifying synoptic conditions that favour the development of this diurnal flow pattern.
In addition to the above-mentioned meteorological effects, the general diurnal pattern of PM10 concentrations shown in Figure 3 also results from variations in emissions. Concentrations were recorded at the main air quality monitoring site in Christchurch, located a few kilometres north of the city centre at Coles Place (Figure 2). Two peaks are apparent in the diurnal cycle, with a minor morning peak at approximately 0900 h and a major peak in the evening at around 2100–2200 h, reflecting enhanced emissions during these times. The morning peak can be mostly attributed to rush hour traffic and industrial emissions. The evening peak is mainly the result of emissions from home heating, although traffic emissions during the evening rush hour also contribute. Afternoons generally show low concentrations due to reduced emission activity and, more importantly, a well-mixed atmospheric boundary layer and stronger wind speeds (Figure 1).
Over the last decade, Environment Canterbury (ECan), the local environmental authority, has carried out several emission inventories of the city and has identified three major sources of PM10 emissions, with varying contributions. Approximately, 80% of all anthropogenic emissions originate from solid fuel burning for residential heating. The remaining 20% is more or less equally divided among emissions from traffic and industry (Scott and Gunatilaka, 2004). However, other natural sources such as wind-blown dust and sea salt regularly contribute to pollution levels, although information on the contribution of natural sources to background levels of particulates is not available.
The main aim of this article is therefore to improve understanding of meteorological influences on particulate pollution in urban environments at a range of temporal and spatial scales, from daily to interannual and local to hemispheric. The focus hereby lies on identifying synoptic weather conditions that control air quality in Christchurch. Investigation of their temporal dynamics is also conducted and links to possible long-term hemispheric forcings are explored.
2. Data and methods
To achieve the aforementioned objectives, local meteorological conditions were classified according to their potential for degrading air quality. This was achieved using classification tree analysis to establish a generalized quantitative relationship between breaches of the NES (exceedances) and corresponding combinations of local meteorological parameters. In contrast to classical correlation and regression-based approaches as used by previous researchers (Owens and Tapper, 1977), a classification-based approach allows not only for identification of relevant relationships between meteorological conditions and PM10 concentrations, but it also provides detailed quantitative information on local air pollution potential in response to varying combinations of meteorological parameters. This is vital, as it enables clear identification of the relationship between particular combinations of local atmospheric conditions and varying degrees of air quality degradation, and these relationships can be expressed as probabilities. These relationships then form the basis for subsequent investigation of synoptic and larger-scale controls.
2.1. Meteorological data
The meteorological data used in this study were recorded at Christchurch airport (Figure 2) and were supplied by the National Institute of Water and Atmospheric (NIWA) Research via their online database ‘CliFlo: NIWA's National Climate Database on the Web’ (CliFlo, 2010).
2.2. Air quality data
Air quality data were supplied by ECan, comprising hourly values of PM10 between 1999 and 2008 recorded at their main (and only continuously recording) air quality monitoring site (Coles Place), which is located just north of the central business district (Figure 2). This site was selected by the regional environmental authorities as it was considered to be where the standard for PM10 is most likely to be breached by the greatest amount in Christchurch, as is required to satisfy the NES. As instrumentation was changed from a Tapered Element Oscillating Microbalance (TEOM) sampler with the inlet temperature set at 40 °C to a Filter Dynamics Measurement System (FDMS) set at 30 °C in 2004, a TEOM–FDMS equivalent dataset was calculated to ensure a continuous record with the highest possible comparability between the two measurement techniques. Although this procedure was undertaken prior to the data being provided to the authors, it is important to be aware that some preprocessing has taken place. For an in-depth description of the calibration technique refer to Scarrott et al. (2009).
In this study, air quality data were aggregated into daily binary values of whether the NES was breached or not, whereas meteorological data were aggregated in several ways (hourly to daily) to capture relevant processes (see Table I for a list of meteorological data used in this research).
Table I. List of meteorological variables used in the classification and regression tree CART analysis
Description (measurement unit)
Minimum temperature of the following day (° C)
Difference between maximum temperature of considered day and minimum temperature of following day (K)
Mean wind speed between 0000 and 0600 h (m/s)
Mean wind speed between 0600 and 1200 h (m/s)
Mean wind speed between 1200 and 1800 h (m/s)
Mean wind speed between 1800 and 2400 h (m/s)
Mean wind speed between 0000 and 1200 h (m/s)
Mean wind speed between 1200 and 2400 h (m/s)
Mean wind speed between 1800 and 2400 h of preceding day (m/s)
Mean wind speed between 1800 h of preceding day and 0600 h of considered day (m/s)
Relative humidity at 0900 h (%)
Mean sea-level pressure at 0900 h (hPa)
Difference in mean sea-level pressure at 0900 h between preceding day and considered day (hPa)
Daily global radiation (MJ)
Accumulated 24 h rain (mm)
2.3. Synoptic classification data
To relate local atmospheric conditions to coarser scale influences, a classification of regularly reoccurring synoptic weather patterns developed by Kidson (2000) was used. On the basis of cluster analysis of 1000 hPa height fields derived from NCEP/NCAR reanalyses (Kalnay et al., 1996), Kidson (2000) identified 12 synoptic classes and related them to New Zealand weather regimes using temperature and precipitation observations (Figure 4). This classification has been used in several studies to assess the influence of synoptic weather patterns in a variety of applications (Beentjes and Renwick, 2001, Jiang et al., 2005). It is available from 1958 to the present day and each day is classified into 1 of the 12 classes at 1200 and 2400 h. Kidson (2000) concluded in his study that, given that the spread of climatic elements within classes is large compared with the differences between classes, this synoptic climatological classification is mainly of qualitative value. However, he also stated that there may be applications in which the classes will be of quantitative use, as, for example, in the frequency analysis of extreme events. As local atmospheric conditions that lead to deterioration in air quality can be considered extreme in the sense that only a narrow range of local meteorological conditions will restrict dispersion sufficiently to permit the build-up of ambient pollutants, a quantitative investigation should therefore be possible.
Mullan (2009) examined expected change in frequency of the synoptic types described by Kidson (2000) by the end of the twenty-first century. He classified daily mean sea level pressure (MSLP) fields from ten Global Climate Models (GCMs) for a 40-year simulation in the twentieth century (commonly known as the ‘20c3m’ run) according to the classification introduced by Kidson (2000). Similarly, Kidson types were diagnosed from daily MSLP fields for a 20-year simulation of the same ten GCMs at the end of the twenty-first century (2080–2100) using the IPCC SRES A1B emission scenario to force the models (the ‘sresa1b’ run). Then, seasonal frequencies of all synoptic types across all ten models for both periods were calculated. Finally, seasonal differences in frequencies were calculated (sresa1b—20c3m) to quantify expected changes (B. Mullan pers. comm.).
2.4. Classification of exceedance probabilities
Classification and regression trees (CART) describe a statistical procedure that was introduced by Breiman et al. (1984). CART has been applied to a wide variety of environmental studies (Hendrikx et al., 2005; Waheed et al., 2006; Zheng et al., 2009). Of most relevance to this study at hand is the article of Slini et al. (2006), which evaluated four different statistical techniques to forecast PM10 concentrations for Thessaloniki, Greece and concluded that CART proved satisfactory in capturing concentration trends. In our case, CART was chosen as the results of this technique are easily interpreted and can be communicated to non-scientific audiences without much explanation, which was important, as the findings of this study are of direct relevance to the local environmental authorities.
A brief summary of the method used in this study is given below, although the reader is referred to Hothorn et al. (2006) for an in-depth description of the recursive partitioning algorithm used here. On the basis of a set of predictor variables, this statistical approach uses recursive partitioning to split the response into a set of classes (nodes) with maximum class purity and arranges the final splits into a decision tree diagram. At each stage of the partitioning, all possible splits are identified using a Monte Carlo approach and a p-value is calculated for each potential split using a suitable statistic (depending on the nature, notably the statistical scale, of the predictor variable) to ensure comparability of the split criteria. Finally, a split is made to produce exactly two nodes using the predictor with the lowest p-value. Each of the resulting nodes then becomes the input for the next level in the decision tree, and the procedure is repeated as outlined above. To avoid overfitting of the classification tree, it is possible to control for minimum node size, that is, the minimum number of observations the resultant nodes must have after each split. Furthermore, the p-value for possible splits can be specified so that splits are only allowed if the split-statistic is significant at a p-value that is lower than the specified level.
In a conservative set up (minimum node size was set to 100 observations and splits were only allowed at p levels < 0.001), almost 1200 daily observations from the years 1999 to 2008 of the binary response as to whether the national guideline of 50 µg m−3 was breached or not (exceedance yes/no) were regressed against a range of predictor variables (Table I). Such a conservative approach was chosen to identify the general conditions that affect the occurrence (or non-occurrence) of an exceedance, rather than the detailed statistical relationships that exist in the investigated data set. Furthermore, this conservative set up addresses the issue of overfitting, even though the utilized algorithm was explicitly developed by Hothorn et al. (2006) to avoid this well-known issue of CART applications. In fact, they concluded that the development of their unbiased recursive partitioning algorithm addresses all previously problematic issues of CART, so that post-classification measures such as pruning and cross-validation become redundant.
Analysis was restricted to winter months (May–August) as this is the period when most of the exceedances occur. As mentioned earlier, air quality observations were collected at the main air quality site in Christchurch at Coles Place, whereas meteorological data were taken from Christchurch airport. Figure 5 shows a comparison of the sites with respect to the key atmospheric variables of air temperature and wind speed, showing a strong relationship (although wind speed is consistently lower at the urban site). To account for the fact that the effect of atmospheric processes are not restricted to any standard time period, lagged information of several meteorological variables was also considered, as is standard procedure in time series analysis. Unfortunately, information on vertical atmospheric structure was not available at satisfactory temporal resolution. However, in an attempt to approximate nocturnal atmospheric stability, information on the cooling rate between the maximum temperature of the considered day and the minimum temperature of the next day was taken into account (Tmax_i_Tmin_i1). Missing data are not allowed in classification tree analysis. Therefore, all days with missing observations of any of the listed variables were excluded from the analysis.
Using a binary categorical response (exceedance/non-exceedance) has a few distinct advantages over using continuous data, such as PM10 concentrations. Firstly, the variance in the predictions is kept to a minimum. Secondly, interpretation of the results is straightforward as the distribution within each final class represents the probability of exceeding the national standard under a given set of meteorological conditions.
3.1. Application of the CART technique
Using the input variables listed in Table I, the algorithm produced a classification tree with five terminal nodes (TNs; see Figure 6—note that TNs were labelled according to their pollution potential from high to low for ease of interpretation). Out of the comprehensive set of potential predictors, three variables were found to most significantly influence splitting of the response into the identified nodes. These are minimum temperature of the following day (split 1—whether a given night is a frost night or not), average wind speed between 1200 and 2400 h (split 2), and average wind speed between 1800 and 2400 h (split 3). As particulate pollution in Christchurch originates mostly from domestic fires, temperature can be considered to be a key driver behind emission release, and wind speed is a major controller of dispersion of pollutants, especially in the absence of information on atmospheric stratification. At a general level, the meteorological conditions for each identified TN can be summarized as follows (note that ‘morning’ refers to the morning of the following day, as represented by Tmin_i1):
TN1: calm conditions during the second part of the day, extremely calm conditions during the evening, frost in the morning.
TN2: similar to TN1, although no major calming effect during the evening.
TN3: calm to very calm conditions during the afternoon and evening, no frost in the morning.
TN4: frost in the morning, windy conditions during the day.
TN5: no frost, windy.
The first two TNs represent conditions that are likely to result in degraded air quality. The meteorological conditions identified by TN5 clearly promote good air quality, and the remaining two classes can be described as intermediate. TN5, the class with lowest pollution potential, accounts for approximately 50% of the original data (622 of 1174 days). The remaining TNs represent between 10 and 15% each of the input data (remember that minimum node size was set to 100 observations). Each of the TNs represents a distinct class of meteorological conditions in the sense that overlaps are not possible.
As mentioned earlier, having a binary response variable enables direct assessment of the pollution potential of each TN. The probability of an exceedance occurring under meteorological conditions as characterized by TN5 is 0.04. It is interesting to note that despite the unfavourable meteorology, exceedances nonetheless occur. Emission contributions can only be speculated, but one possible source in such conditions may be sea spray, given the coastal location of Christchurch. At the other end of the scale, 82% of all days with meteorological conditions as described by TN1 can expect a breach of the national air quality standard for PM10.
3.2. Effects of local meteorology on air quality
Examination of the diurnal variation of PM10 concentrations within each of the TNs reveals that the greatest differences are related to the rate of increase in concentrations in the evening (Figure 7). The difference between TN1 and TN2 is especially pronounced, as a result of the lower wind speeds after 1800 h in TN1 cases. Apart from the difference in magnitude, a notable delay of peak concentrations of about 2 h is apparent in TN1 when compared with TN2. The evening concentrations for TN3 are similar in magnitude to those for TN2, although they tend to peak later. The following analysis focuses on the conditions that lead to the majority of PM10 exceedances in Christchurch, as observed in both TN1 and TN2.
Figure 8 shows average hourly PM10 concentrations for these two TNs with regard to wind direction (left panels) and wind speed (right panels). Note that the y-axis scale is centred around midnight to allow easy analysis of conditions at the time of day that exhibits highest PM10 concentrations. Southeasterly wind conditions yield some of the highest hourly concentrations in both TN1 and TN2. However, it is obvious from the wind data plots that they are associated with a very low frequency of occurrence. They will, therefore, not be analysed here.
It is to be expected that absolute concentrations mirror the exceedance-based classification, so that it is not surprising to see that PM10 concentrations are higher for TN1 than TN2. In addition, the key difference between TN1 and TN2 is the occurrence of lower wind speeds after 1800 h, so that a very prominent increase in PM10 concentrations after 1800 h is evident in TN1. This effect is more or less independent of wind direction, with the exception of southwesterly winds, which exhibit less potential to degrade air quality. Southwesterly flow generally tends to be windier, and thus associated with enhanced turbulent mixing, which in turn increases air pollution dispersion near the surface. This is observed in both classes. Apart from southeasterlies, wind directions that cause highest concentrations are clearly between northerly and westerly. Northeasterlies tend to be associated with lower particulate levels, although this is less so in the case of TN1. In TN2, highest concentrations are associated with flow from northwesterly directions, with a distinct brief maximum at around 2100 h. On the other hand, under northwesterly flow, the peak at TN1 is delayed, and occurs around or slightly after midnight. It is also indicated that higher concentrations can be observed slightly earlier in a short but intense peak (indicated by the dashed blue circle in the top panel of Figure 8) that is associated with winds from the north. The timing (between 2100 and 2400 h) and location (northerly flow direction) of this peak indicates that this maximum is likely to be a result of the transition between daytime northeasterly airflow and night time westerlies. The same signal of a short peak under northerly airflow is apparent in TN2 cases about 3 h later, although it is much reduced in magnitude (see dashed blue circle in the bottom panel of Figure 8). It appears to indicate the same process of local airflow transition associated with elevated levels of particulates. Further analysis of the effect of the transition periods within local wind regimes on PM10 concentrations in Christchurch can be found in Appelhans (2010).
3.3. Influence of synoptic weather patterns on air quality
To investigate meteorological influences on air quality that result from atmospheric processes on coarser spatial scales, TNs from the classification in Figure 6 were analysed with regard to their synoptic forcings. As no information on PM10 concentrations is needed for this, it was decided to extend the period used to investigate synoptic influences over a slightly longer time period to try to capture variability associated with longer-term influences, such as El Niño-Southern Oscillation (ENSO). Thus the analysis presented here is based on meteorological observations between 1995 and 2008.
Figure 9 shows the relative increase or decrease in frequency of the set of 12 synoptic types identified by Kidson (2000) within each of the TNs in comparison with their overall distribution for both 1200 and 2400 h (the times for which synoptic types are classified each day). The further a box deviates from the horizontal line representing the expected frequency distribution, the stronger the relationship between the synoptic type and the exceedance probability class. The width of the boxes is proportional to their overall frequency adjusted to the TN frequency, so that the ratio of widths between all synoptic types within each TN and the frequency of each synoptic type across TNs are kept constant. The colour shading relates to the strength of dependence between the considered categories and is, simply speaking, a visualization of significance levels based on Pearson residuals (Zeileis et al., 2007).
For both daily synoptic classifications (1200 and 2400 h), the pattern is similar, although slight differences are apparent. Given that local wind conditions during the evening are most important for the isolation of TN1 and TN2, greater focus is given to synoptic conditions at midnight. However, the dynamics of progression between different synoptic conditions is also important and will be discussed later. It is obvious that the two high pollution classes are dominated by anticyclonic conditions, with a high pressure system located over the Tasman Sea in close vicinity to New Zealand, denoted by the highly significant increase in the occurrence of synoptic type H, and also HW (although less significant), in TN1. It is suggested that latitudinal variation in the location of this anticyclone is of importance, given the decreased dependence of TN2 on HW. A summary of synoptic influences for each TN is given below (refer to Figure 9(b)).
TN1: local atmospheric conditions are mostly a result of anticyclones located in close vicinity to the west of the country (H and HW) or directly over the (eastern) South Island (HSE).
TN2: similar to TN1, although the high pressure system over the Tasman Sea is likely to be located slightly further north as indicated by increased frequencies of HNW and W at the expense of H and HW. This leads to an increased isobaric gradient over the South Island, which may explain the slightly windier conditions for cases classified in this node when compared with TN1.
TN3: no significant signal is apparent. However, all synoptic types that show anticyclones located to the east of New Zealand show slightly increased frequencies (HE, HSE, NE, and TNW).
TN4: atmospheric conditions in Christchurch as described by TN4 are clearly associated with southwesterly flow, with significantly increased frequencies of HNW and SW, and a marked decrease in types promoting northerly to northwesterly flow (HE, NE, and TNW).
TN5: is dominated by flow from northerly quadrants (HE, NE, and TNW), but also a result of enhanced frequencies of situations that favour zonal flow over the South Island (T and W → westerlies and TSW → easterlies).
In summary, the local meteorological conditions in each TN as identified by classification tree analysis can be clearly associated with distinct synoptic signatures that control large scale airflow, which in response modify local meteorology. However, these synoptic forcings are not stagnant over time and their temporal evolution is associated with smooth transitions from one state to another, which also have an effect on air quality.
3.4. Effects of synoptic transition on air quality
To obtain a clearer understanding of the temporal evolution of weather systems in a given area, it is useful to investigate average persistence of synoptic types and likely transitions between them. Figure 10 shows the relative frequency of persistence and transition for each synoptic type, as well as the mean time of persistence in days. Frequency counts were derived from the time series of synoptic types by comparing each observation i to its immediate successor i + 1. Equality between i and i + 1 was defined as persistence, a change in type between i and i + 1 was counted as transition. The most dominant type, in both absolute frequency and persistence duration, is type H. In addition to H, winter months are dominated by HSE, SW, T and TSW, which is in line with the findings of Kidson (2000). Although HNW occurs reasonably frequently, this synoptic type tends to progress quicker towards other types (i.e. it is less persistent). True transitional types are HW, NE, R and especially W, which on average only persists for approximately 24 h. For his classification, Kidson (2000) analysed the most common transitions between synoptic types and concluded that the strongest transition patterns are in line with expectations from qualitative observations. However, Kidson (2000) took into account all seasons to analyse the common patterns of transition between the types, whereas the analysis presented here is restricted to the months May–August. In addition, he outlined significant seasonal differences in the frequency of the types. Therefore, the main transition patterns were analysed for winter to find common patterns of synoptic evolution with respect to air quality in Christchurch.
Figure 11 shows the nature of the transitions by assigning a probability that denotes the likelihood of one type being followed by another. This reveals that between May and August the most common transition is T → SW → HNW → H → HSE → HE → TNW → T (taking the highest transition probability as a starting point). This is completely in agreement with the most common transition pattern identified by Kidson (2000). Figure 12 shows this transition in more detail, along with pollution potential indicated by association with TNs from the classification tree analysis at each stage, as identified in Figure 9. This transition depicts the eastward progression of a high pressure system embedded in the midlatitude westerlies en route from the Tasman Sea, crossing New Zealand, and finally leaving the region towards the east. As a result of the meandering of the predominant westerly flow in the region, the location of the anticyclone is subject to meridional variations in its location during its eastward propagation. This may result in a variety of weather type sequences depending on whether the anticyclone follows a more northerly or southerly path. In some cases, the pattern of evolution may be completely disrupted, for example, during an anticyclonic blocking episode.
With regard to air quality, the described transition patterns show varying potential for degradation. The most common transition with a clear tendency to create local atmospheric conditions conducive to elevated levels of particulate matter during its intermediate stages is HNW → H → HSE. This pattern is usually followed by northwesterly flow, which, in the case of the Canterbury region, is mostly associated with lee side foehn conditions, and hence a clearing of the atmosphere. In the case of a southward high pressure passage, HW and HSE are the situations with enhanced pollution potential. However, as mentioned earlier (Figure 10), HW generally shows transitional character and therefore does not last long. In addition, the location of anticyclones to the south of the country generally produces easterly flow in the region that is regularly associated with cloudy, windy and possibly rainy conditions. Therefore, this transition can be expected to be more favourable for enhanced air quality. Finally, the transition describing a northern anticyclonic passage can be assumed to yield the least potential for degrading air quality in Christchurch, mostly because the influence of the anticyclone is generally smaller, as it is further away. It is, however, important to recognize that the above analysis provides a general idea of the influence of the evolution of weather patterns on air quality, and that in reality transitions between synoptic weather types can be highly variable.
3.5. Longer-term and larger-scale relationships
It is evident that anticyclonic situations generally favour light winds and clear skies, thus promoting low-level atmospheric stability, and in many cases leading to exceedances of the national standard for PM10 concentrations in Christchurch. In fact, in the period 1999–2008, 43% of all exceedances were associated with these weather types. Therefore, an analysis of the historic variability of their occurrence during winter time should provide a general indication of the long-term variations in the atmospheric conditions that produce poor air quality in Christchurch. As Tmin_i1 was found to be most influential with regard to exceedance potential, a combination of daily minimum temperature values (described by Tmin_i1) and the H and HW synoptic weather types (which, together, account for 66% of all exceedances between 1999 and 2008), should provide an acceptable approximation of historic air pollution potential and its long-term variability. Figure 13 shows deviations from the overall mean frequency of all 123 days between May and August of each year between 1960 and 2008 with Tmin_i1 < − 0.1 °C and synoptic conditions classified as either H or HW (hereafter referred to as HHWT). The smoothed signal was calculated employing a locally weighted polynomial regression model (Loess), which was first introduced by Cleveland (1979). A clear periodicity is apparent in the frequency of the synoptic types in question. Peak occurrences in air pollution potential are seen for the late 1960s, the mid-1980s and the late 1990s. Relatively few occurrences are evident in the mid-1970s and early to mid-1990s. Discrete Fourier transformation (not shown) of the raw counts indicates weak frequency maxima at 4 and 16 years. Autocorrelation in time (also not shown) was also assessed and reveals a strong inverse correlation (r ≈ − 0.5) every 8 years, which is in line with the 16-year signal identified through Fourier transformation. It is not clear what might be controlling this periodicity.
In an attempt to identify the underlying controls of the variations shown in Figure 13, the occurrence of synoptic classes H and HW (HHW) during May to August of each year was compared with mean values of the Southern Annular Mode (SAM) and the Southern Oscillation Index (SOI), averaged accordingly (May–August) (Figure 14). In this case, the period of analysis ended in 2007 due to limited availability of information of the low frequency climatic signals after this date. As the name implies, the SAM is oscillating annually. It describes an oscillation in surface pressure between the mid- and high latitudes of the Southern Hemisphere (Bridgman and Oliver, 2006). Its influence on the weather and climate in New Zealand has been subject to numerous scientific studies (Ummenhofer et al., 2009; Clare et al., 2002), and during austral winter it is usually in its negative phase, which explains the mostly negative values in Figure 14. It becomes apparent, however, that this mode also shows a periodic signal of about 4–6 years on an interannular scale. Furthermore, it is evident that, apart from the period between the mid-1970s and the early 1980s, the HHW signal is mostly in phase with the SAM, especially after 1983. It is not clear what causes the signal to be out of phase during the mid-1970s to early 1980s. No clear relationship is seen for the SOI. However, it is indicated that, in general, positive SOI values seem to favour the occurrence of synoptic types H and HW. This might provide an explanation for the observed phase discrepancy between HHW and SAM signals, at least during the mid-1970s, where a distinct peak in the SOI is apparent. However, an in-depth analysis of longer-term climate forcings on New Zealand weather is beyond the aims of this study.
Finally, having identified a relationship between synoptic weather types and air pollution exceedances in Christchurch, it is possible to interpret the findings of Mullan (2009) from an air quality perspective.
His projections for the winter season are shown in Figure 15.
With regard to the focus of this article, the predicted change that is of most importance is the increase of synoptic type H during winter. This indicates, that under an average climate change scenario it is to be expected that local conditions may become more conducive to elevated levels of particulates in Christchurch, as such conditions are regularly associated with this synoptic type. For the other two important types, HW and HSE, no clear projection of change can be formulated. Types that were identified to be associated with good air quality, such as TNW and TSW, are expected to become less common in the future. To put the expected percentage changes into perspective, for the 123 day period between May and August, a 2% change equals 2.46 days. Overall, the projected changes generally paint a mixed picture with regard to expected changes in air quality, as some types that are expected to enhance air quality in Christchurch are projected to become more common, such as T, W and SW. Therefore, a robust conclusion relating to expected air quality changes towards the end of this century seems unfeasible. Furthermore, changes to the building structure of Christchurch, technological advances and further regulatory measures over coming decades can be expected to show far greater influence on the pollution problem than the subtle changes in weather type frequencies that are presented here.
This study has investigated meteorological influences that control air quality at a range of spatial and temporal scales. It has been possible to identify the local meteorological conditions and larger-scale synoptic weather patterns that are most likely to influence air quality in Christchurch, and then to use this knowledge to investigate controls of variations in air pollution potential over the longer term.
As a first step in the analysis, the CART technique was successfully used to identify meteorological factors that are most closely associated with exceedances of the national standard for PM10. On the basis of observations of temperature and wind speed, CART produced two nodes that were identified as the most significant for the occurrence of exceedances of the PM10 standard. It was shown that 82% of all days with the meteorological characteristics represented by TN1 would expect to experience exceedances. The nature of the relationship between these meteorological factors and air pollution concentrations was then explored, particularly the effect of the evening transition period in the local wind regime, when subtle changes of wind speed and direction occurred and were seen to be related to high levels of air pollution.
Subsequent analysis moved upscale and identified the effect of synoptic weather patterns on occurrence of exceedances of the PM10 standard. Clear relationships were identified between synoptic weather patterns and air quality in Christchurch, based on the Kidson classification of synoptic weather patterns over New Zealand. Occurrence of PM10 exceedances was seen to be associated with dominance of anticyclones (especially synoptic types H and HW) and a lack of troughs (TNW and TSW), while low occurrence was associated with a lack of H and other anticyclonic types, and increased occurrence of TNW and other westerly weather types. It can safely be concluded that synoptic conditions that promote degraded air quality in Christchurch are mainly associated with anticyclones that are located in close vicinity to New Zealand, and it is evident that air quality should be worse when their centres are to the west of the country. Furthermore, passage of an anticyclone over the country enhances locally clear and calm meteorological conditions, promoting elevated levels of particulates.
Further analysis of weather type progression through the New Zealand region showed a clear relationship between typical weather pattern sequences and the rise and fall in occurrence of air pollution exceedances. The latitudinal location of the track of anticyclones across New Zealand was seen to be a major influence on air quality in Christchurch. Generally, the further the centre of the passing anticyclone is located away from the central South Island, the better the air quality is expected to be in Christchurch. On the basis of these results, future research could be aimed at understanding the relationship between synoptic weather patterns and local meteorological conditions, and investigating the mechanisms that cause local variations in such factors as airflow and temperature (with consequent effects on air pollution). This may shed further light on the high variability in PM10 concentrations under seemingly similar synoptic weather conditions, as there is still a large amount of variability in air quality that cannot be explained by the results presented here. Some of this residual variation is associated with socioeconomic factors that influence anthropogenic emissions, but it is expected that a portion of the remaining variability may be explained by further in-depth analysis of mesoscale meteorological variations.
In the final section of the analysis, having identified relationships between synoptic weather systems and air pollution in Christchurch, an attempt was made to use this new knowledge to investigate longer-term variations in air quality. It was found that interdecadal variation in the winter time synoptic conditions likely to affect air quality in Christchurch shows distinct periodic fluctuations that reveal a close phase relationship with variations in the intensity of the SAM. The periodic nature of variations in occurrence of different synoptic weather systems that influence air quality, and their link to variations in the strength of the SAM, highlights that air quality in Christchurch is ultimately influenced by processes that operate on interannual hemispheric scales, and implies that pollution potential can also be expected to vary on a periodic interdecadal basis. This aspect of the results is worthy of further work.
Predicting the future potential of air quality based on an ensemble of climate model runs also proved to be informative, in that effects of changes in the occurrence of individual weather types on air pollution potential up to the end of the present century could be predicted. However, the often counteracting effects of changes in the frequency of individual weather types and the large unknowns associated with such external factors as changes in social and economic conditions, building and heating technology and renewable energy availability means that no definitive prediction of future air quality can be provided at this stage. There is therefore plenty of scope for future investigation in this area.
This research was sponsored through the research programme C01X0405 ‘Protecting New Zealand's Clean Air’, funded by the Foundation for Research Science and Technology. The meteorological data used in this study were supplied by the NIWA Research via their online database ‘CliFlo: NIWA's National Climate Database on the Web’. Air quality data were kindly provided by Environment Canterbury (ECan). The authors would also like to thank Brett Mullan from NIWA for providing useful input for the expected future changes of synoptic conditions under a climate change scenario (Figure 15). The time series of Kidson (2000) weather types was kindly supplied by Jim Renwick at NIWA. Finally, the authors would like to thank the anonymous reviewers for helpful comments and suggestions to improve this article.