446
E. C. White et al.
connected to the rate of incidence of acute respiratory illness [8], and to larger
reductions in crop yields compared to equivalent long-term, lower concentration
ozone exposure [6]. Ozone concentrations are positively correlated with temperature
and stagnancy, and negatively correlated with cloud cover [4], owing to increased
emissions of biogenic ozone precursor compounds at high temperatures and increased
rates of photolysis, among other things. This suggests that ozone pollution episodes
are more likely to happen during heat wave conditions. Previous studies have found
that North American summertime heat waves may be preceded by particular largescale meteorological patterns with subseasonal persistence [7, 9]. This study aims to
determine whether summertime regional ozone pollution episodes in North America
might similarly be preceded by such large-scale meteorological patterns.
71.2 Data
The ozone data used for this study are from the United States Environmental Protection Agency’s Air Quality System (AQS) [11] and from Environment and Climate
Change Canada’s National Air Pollution Surveillance Program (NAPS) [2]. The
ozone concentrations are maximum daily 8-hour average values (MDA-8) in June,
July, and August (JJA) from 1990 to 2016. ERA-Interim reanalysis fields [1] are used
to examine meteorology during ozone episodes.
71.3 Ozone Episode Identification
The AQS and NAPS stations span an area too large for any definition of an ozone
pollution episode to be reasonably applied to all stations in aggregate. To remedy
this, the AQS and NAPS stations are divided into smaller groups using a statistical
clustering algorithm. Specifically, hierarchical agglomerative clustering [12] is used,
with a similarity metric between each possible pairing of stations used as input.
The similarity metric is a Jaccard distance [3] based on daily co-occurrences of
extreme ozone concentrations (i.e. greater than historical 95th percentile at a location)
between two stations. The results of the clustering are shown in Fig. 71.1.
Within each cluster, ozone pollution episodes are defined for all ozone monitoring
stations in aggregate, with the daily 95th percentile ozone measurement, herein called
O95, being the primary quantity of interest. An extreme threshold is defined as
the mean plus standard deviation of O95 in a 5-year rolling window. The rolling
window is applied to remove the influence of decreasing anthropogenic emissions of
ozone precursor compounds over time. An ozone episode occurs when the extreme
threshold is surpassed for 3 or more consecutive days. In the Northeast cluster, this
definition identifies 51 ozone episodes over the 27-year period of interest.
E. C. White et al.
connected to the rate of incidence of acute respiratory illness [8], and to larger
reductions in crop yields compared to equivalent long-term, lower concentration
ozone exposure [6]. Ozone concentrations are positively correlated with temperature
and stagnancy, and negatively correlated with cloud cover [4], owing to increased
emissions of biogenic ozone precursor compounds at high temperatures and increased
rates of photolysis, among other things. This suggests that ozone pollution episodes
are more likely to happen during heat wave conditions. Previous studies have found
that North American summertime heat waves may be preceded by particular largescale meteorological patterns with subseasonal persistence [7, 9]. This study aims to
determine whether summertime regional ozone pollution episodes in North America
might similarly be preceded by such large-scale meteorological patterns.
71.2 Data
The ozone data used for this study are from the United States Environmental Protection Agency’s Air Quality System (AQS) [11] and from Environment and Climate
Change Canada’s National Air Pollution Surveillance Program (NAPS) [2]. The
ozone concentrations are maximum daily 8-hour average values (MDA-8) in June,
July, and August (JJA) from 1990 to 2016. ERA-Interim reanalysis fields [1] are used
to examine meteorology during ozone episodes.
71.3 Ozone Episode Identification
The AQS and NAPS stations span an area too large for any definition of an ozone
pollution episode to be reasonably applied to all stations in aggregate. To remedy
this, the AQS and NAPS stations are divided into smaller groups using a statistical
clustering algorithm. Specifically, hierarchical agglomerative clustering [12] is used,
with a similarity metric between each possible pairing of stations used as input.
The similarity metric is a Jaccard distance [3] based on daily co-occurrences of
extreme ozone concentrations (i.e. greater than historical 95th percentile at a location)
between two stations. The results of the clustering are shown in Fig. 71.1.
Within each cluster, ozone pollution episodes are defined for all ozone monitoring
stations in aggregate, with the daily 95th percentile ozone measurement, herein called
O95, being the primary quantity of interest. An extreme threshold is defined as
the mean plus standard deviation of O95 in a 5-year rolling window. The rolling
window is applied to remove the influence of decreasing anthropogenic emissions of
ozone precursor compounds over time. An ozone episode occurs when the extreme
threshold is surpassed for 3 or more consecutive days. In the Northeast cluster, this
definition identifies 51 ozone episodes over the 27-year period of interest.