Page 174 - IJES Special Issues for AIEC2016
P. 174
74 © Shareef, Husain, and Alharbi 2016 | Optimal Air Quality Monitoring Network
The predictive equations were developed by re- 2. Methodology
gressing the passive monitor measurements at
the 22 monitored schools on land-use variables 2.1 Interpolation Methods
derived from GIS. A universal Kriging interpola-
tion method was found to perform better than Interpolation predicts values for cells in a raster
other methods when a comparison was made be- using a limited number of sample data points,
tween various interpolation methods by devel- which helps in predicting unknown values for
oping EU-wide high resolution air pollution any geographic station. Five interpolation meth-
maps (Beelan et al., 2009). The best method to ods were selected to estimate the concentrations
model concentrations was selected on the basis of air pollution at the unknown stations. The se-
of predefined performance measures, correla- lected methods were (1) Inverse Distance
tion coefficient and RMSE. A generalized mono- Weighted (IDW), (2) Spline (SPL), (3) Ordinary
tonic regression based on B-splines with an ap- Kriging (OK), (4) Universal Kriging (UK) and (5)
plication to air pollution data was proposed by Natural Neighbor (NN). These methods have
Leitenstorfer and Tutz (2007) and the spline been widely used in estimating air pollution con-
method was used to estimate the varying health centrations.
risks from air pollution across Scotland by Dun-
can (2011). The IDW uses a method of interpolation that es-
timates cell values by averaging the values of
The methods summarized above are very useful, sample data points in the neighborhood of each
well established and have been implemented processing cell (ESRI Redland). The closer a
widely. However, it appears a simple GIS based point is to the center of the cell being estimated,
methodology would further reduce the complex- the more influence or weight it has in the averag-
ities of AQMN design. The basic advantage of us- ing process. This has been used in several in-
ing GIS is that it organizes geographic data in stances to interpolate air pollutant concentra-
such a way that the decision making process be- tions (Wong et al., 2004; Hoek et al., 2002; Jerrett
comes easy. In addition to this, it provides sev- et al., 2013).
eral advanced functionalities to manage statisti-
cal and spatial data, interpolate the data to create The SPL uses an interpolation method that esti-
smooth surface, extract data from the interpo- mates values using a mathematical function that
lated surface and create algorithms to automate minimizes overall surface curvature, resulting in
the process. Furthermore, it creates results that a smooth surface that passes exactly through the
can be visualized in interactive maps, which will input points. This method was found superior in
further simplify the decision making process. varying health risk from air pollution (Duncan,
Taking the cue on these advantages, this paper 2012) and was also recommended by
proposes a simple and innovative process to op- Leitenstorfer and Tutz (2006).
timize AQMN by using GIS, interpolation meth-
ods and historical data. The existing stations are Kriging is an advanced geo-statistical procedure
systematically eliminated by creating several in- that generates an estimated surface from a scat-
terpolated maps and comparing it with the ob- tered set of points with z-values. While Kriging is
served values. The number of stations that can a weighted combination of monitor values, this
be eliminated is governed by the pre-defined method also uses spatial auto correlation among
performance measures criteria. Recently an in- data to determine the weights. Generally Kriging
creasing trend of air pollution has been observed has two different forms—ordinary and universal
in Riyadh city of Saudi Arabia and there is an em- Kriging (Burrough, 1998). In the ordinary Krig-
phasis on frequent air pollution measurements ing the mean value is assumed constant and
(Alharbi et al., 2015), hence Riyadh is used as a determined during interpolation, and universal
case study to test the proposed methodology. Kriging assumes that data follows a known
trend. Ordinary and universal Kriging have pre-
viously been used with success to model ozone
(Liu and Rossini, 1996) and particles (Cressie,
2006) at the local scale and to model broad scale
variations in background air pollution (Lefohn et
Science Target Inc. www.sciencetarget.com
