pollution (Tai Po, Sha Tin, Islands) corre-
spond well with densely vegetated, sparsely
populated, and elevated regions. Addition-
ally, the regions (Yuen Long and North)
close to mainland China show high
PM2.5
concentrations which might be affected by
the particle pollution in Shenzhen.
To further compare the performance of
the four models (
OLS
,
GWR
,
GTWR
,
IGTWR
) for
annual
PM2.5
predictions, the mean absolute
error (
MAE
) and standard deviation (
STD
) of
errors between the
PM2.5
estimations and
ground measurements of three years at 14
monitoring stations are calculated. The
MAE
in Figure 5a shows the average abso-
lute difference between the model derived
PM2.5
and ground observations at different
stations, ranging from 5.3 to 12.4. The
STD
of errors shown in Figure 5b ranges from 6.7
to 15.3, indicating the errors dispersion of
different monitoring stations. From the com-
parison, we can see that the
IGTWR
annual es-
timations of
PM2.5
agree best with the ground
measurements among the four models.
Discussion
In this study, the
PM2.5
estimations from
OLS
,
GWR
,
GTWR
, and
IGTWR
are validated
against ground observations. The annual
PM2.5
levels for different models are also
compared. The
GWR
-based models are
Figure 4.
IGTWR
derived mean
PM2.5
distributions averaged over
AOD
available days and ground
PM2.5
observations in Hong
Kong for (a) 2012, (b) 2013, (c) 2014, and (d) three years.
Figure 5. The mean absolute error (
MAE
): (a) and standard deviation (
STD
) of
error,and (b) between the
PM2.5
estimations and ground measurements at 14
stations from 2012 to 2014 (The abbreviations of the station names are used
in the figure. They are
CW
(Central/Western),
E
(Eastern),
KT
(Kwun Tong),
SSP
(Sham Shui Po),
KC
(Kwai Chung),
TW
(Tsuen Wan),
YL
(Yuen Lung),
TC
(Tung Chung),
TP
(Tai Po),
ST
(Sha Tin),
TM
(Tap Mun),
CB
(Causeway Bay),
C
(Central), and
MK
(Mong Kok).).
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
December 2018
767
