calculating spatio-temporal distance (He and Huang, 2018),
this method calculates the cosine amplitude with taking the
reference date fixed at the beginning of the cycle, and the
scale factors among spatial, temporal and seasonal distances
are not well considered. Thus, the seasonal complex nonsta-
tionarity has not been fully exploited. In this paper, we pro-
pose an improved
GTWR
(
IGTWR
) model that incorporates sea-
sonal variation by introducing a new calculation of seasonal
distance for spatiotemporal weight determination. The
IGTWR
model is established for
PM2.5
predictions in Hong Kong based
on
AOD
, meteorological and land use data. In order to improve
the accuracy and spatial resolution of
PM2.5
predictions, the
simplified aerosol retrieval algorithm (
SARA
) (Bilal
et al.,
2013; Bilal
et al.,
2014; Bilal and Nichol, 2015; Bilal
et al.,
2017; Bilal
et al.,
2017) is employed to produce
AOD
data with
500 m spatial resolution. The
IGTWR
model is also compared
with the ordinary least squares (
OLS
),
GWR
, and
GTWR
models
for performance assessment. In the end, the spatial patterns
of annual
PM2.5
levels in Hong Kong from 2012 to 2014 are
shown based on the proposed
IGTWR
model.
This paper is structured as followings. The next section
introduces the data collection from ground stations and
satellites, followed by a description of the principle of
GTWR
model and our improvements. Next, the comparison between
IGTWR
and other methods based on
PM2.5
predictions and
observations on sites is presented. The annual predicted
PM2.5
distributions from 2012 to 2014 are also presented at the end
of this section. The conclusions are given in the last section.
Study Area and Data Collection
Study Area
Hong Kong is located in the south of China, from 22°08
′
to
22°35
′
Latitude and from 113°49
′
to 114°31
′
Longitude, and
is surrounded by the sea on three sides. Hong Kong is a city
with high population density, which has over 7 million
residents within only 1,100 km
2
. In recent years, the closer
economic cooperation with Pearl River Delta (
PRD
) region,
coupled with the increase in population and traffic, have
placed increasing pressure on the atmospheric environment
of Hong Kong (Bilal
et al.
2013). The number of hazy days
in winter from 2000 onwards is 5 times higher compared to
that of the 1980s (Roza, 2010). The high air pollution in 2012
caused 3,096 premature deaths, 151,300 hospitalizations, 7.16
million doctor visits, and 39,499 million dollars costs (Clean
Air Network (
CAN
) 2013). Therefore, more accurate prediction
of
PM2.5
spatio-temporal variations is necessary for the moni-
toring, management, and protection of atmospheric environ-
ment in Hong Kong.
Ground PM2.5 Data
The hourly
PM2.5
concentrations data from 01 January 2012 to
31 December 2014 were downloaded from the website of the
Environment Protection Department (
EPD
)
.
gov.hk/
). The
EPD
started to build up air quality monitoring
stations from 1995, with the aim of monitoring the content of
air pollutants (
NO
2
,
SO
2
,
NO
,
PM10
,
PM2.5
, etc.). These stations
are evenly distributed in Hong Kong and their locations are
shown by points in Figure 1.
MODIS AOD Data
The MODerate resolution Imaging Spectroradiometer (
MO-
DIS
) was aboard satellite Terra launched in 1999 and Aqua
launched in 2002. Its scan width is 2,330 km, and it can
obtain global observation data at least once per day. The
overpass time of Terra and Aqua in Hong Kong is between
10:00 and 11:00 local solar time (
LST
) and between 14:00 and
16:00
LST
, respectively.
MODIS
has 36 spectral bands with
wavelengths of 0.4 - 14 μm at three spatial resolutions of 250
m, 500 m, and 1000 m. These bands can be used to retrieve
data of aerosol, water vapor, land surface temperature, and
ocean color (Yap and Hashim, 2012).
AOD
is usually retrieved
Figure 1. The location of
PM2.5
monitoring stations,
AERONET
stations, and meteorological stations.
762
December 2018
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
