300
J. Soares et al.
ing air quality network or design a new one is the combined use of the model and
clustering analysis [1]. HC is a well-established methodology used to determine the
inherent or natural groupings of objects [2]. The similarity among members is determined by a distance metric, which is used to create a similarity matrix in which data
are cross-compared. This is followed by operations on the similarity matrix which
group data according to their degree of (dis)similarity with respect to that metric.
47.2 Methodology
The methodology presented here is based on the associativity analysis described
in the work of Solazzo and Galmarini [4] and references therein. The HC analysis
uses the hourly or time-filtered time series of observations at different monitoring
stations in Alberta, and analyses this data based on two dissimilarity metrics, 1-R
(R being the Pearson correlation coefficient), and Euclidean distance to understand
the similarity between time series in terms of temporal and magnitude variation,
respectively. Dissimilarity analysis results may be used to rank stations in terms of
potential redundancy, where the lowest levels of dissimilarity may point to stations
being potentially redundant.
The methodology can be applied to hourly resolution or to time-filtered time
series. A Kolmogorov-Zurbenko filter [6] was applied to filter the time series to
remove daily, weekly and monthly or shorter time scales signal from the hourly time
series.
The methodology was first applied to hourly observations, and to hourly model
results extracted at station locations, to assess the model’s ability to recreate the
associations between observed records. We then apply the same methodology to
modelled gridded-data and assess the extent to which model output can be used as a
potential surrogate for observations in clustering analysis.
For this study, observational hourly data was collected for NO 2 , SO 2 , PM 2.5 and
O 3 from the monitoring networks of Alberta for the period between August 2013
and July 2014 was collected. There are nine networks within Alberta (Fig. 47.1b):
Alberta Capital Airshed Alliance (ACAA), Calgary Regional Airshed Zone (CRAZ),
Lakeland Industrial Community Association (LICA), Fort Air Partnership (FAP),
Peace Airshed Zone Association (PAZA), Palliser Airshed Society (PAS), Parkland
Airshed Management Zone (PAMZ), West Central Airshed Society (WCAS), and
Wood Buffalo Environmental Association (WBEA). The data was subjected to additional quality assurance and control procedures to avoid gaps in the time series of
observations: continuous station data should be rejected if their hourly data records
for the analysis period have more than 10% of the total data for the year missing, or
contain data gaps of more than 168 consecutive hours in duration [4].
Model simulations were carried out for the same time period, over a domain
centred over North America with 10 km grid spacing with the Global Environmental Multiscale—Modelling Air-quality and Chemistry modelling system [3]. The
resulting outputs were used as initial and boundary conditions for a nested set of
J. Soares et al.
ing air quality network or design a new one is the combined use of the model and
clustering analysis [1]. HC is a well-established methodology used to determine the
inherent or natural groupings of objects [2]. The similarity among members is determined by a distance metric, which is used to create a similarity matrix in which data
are cross-compared. This is followed by operations on the similarity matrix which
group data according to their degree of (dis)similarity with respect to that metric.
47.2 Methodology
The methodology presented here is based on the associativity analysis described
in the work of Solazzo and Galmarini [4] and references therein. The HC analysis
uses the hourly or time-filtered time series of observations at different monitoring
stations in Alberta, and analyses this data based on two dissimilarity metrics, 1-R
(R being the Pearson correlation coefficient), and Euclidean distance to understand
the similarity between time series in terms of temporal and magnitude variation,
respectively. Dissimilarity analysis results may be used to rank stations in terms of
potential redundancy, where the lowest levels of dissimilarity may point to stations
being potentially redundant.
The methodology can be applied to hourly resolution or to time-filtered time
series. A Kolmogorov-Zurbenko filter [6] was applied to filter the time series to
remove daily, weekly and monthly or shorter time scales signal from the hourly time
series.
The methodology was first applied to hourly observations, and to hourly model
results extracted at station locations, to assess the model’s ability to recreate the
associations between observed records. We then apply the same methodology to
modelled gridded-data and assess the extent to which model output can be used as a
potential surrogate for observations in clustering analysis.
For this study, observational hourly data was collected for NO 2 , SO 2 , PM 2.5 and
O 3 from the monitoring networks of Alberta for the period between August 2013
and July 2014 was collected. There are nine networks within Alberta (Fig. 47.1b):
Alberta Capital Airshed Alliance (ACAA), Calgary Regional Airshed Zone (CRAZ),
Lakeland Industrial Community Association (LICA), Fort Air Partnership (FAP),
Peace Airshed Zone Association (PAZA), Palliser Airshed Society (PAS), Parkland
Airshed Management Zone (PAMZ), West Central Airshed Society (WCAS), and
Wood Buffalo Environmental Association (WBEA). The data was subjected to additional quality assurance and control procedures to avoid gaps in the time series of
observations: continuous station data should be rejected if their hourly data records
for the analysis period have more than 10% of the total data for the year missing, or
contain data gaps of more than 168 consecutive hours in duration [4].
Model simulations were carried out for the same time period, over a domain
centred over North America with 10 km grid spacing with the Global Environmental Multiscale—Modelling Air-quality and Chemistry modelling system [3]. The
resulting outputs were used as initial and boundary conditions for a nested set of