Geography Reference
In-Depth Information
N
N
w
z
(
t
)
z
(
t
)
ij
i
j
i
1
j
1
G
st
N
N
z
(
t
)
z
(
t
)
i
j
i
1
j
1
(5)
where
s
I
,
s
C
, and
s
G
can be taken as different formats of space-time cross-
correlation (or cross-product) models (Getis, 1991);
Z
is the target variable of
interest;
i
and
j
are indices of total
N
spatial units;
w
ij
is an element of the k-
order-neighbor spatial weighted matrix (1
st
, 2
nd
,
……
, k
th
);
_
t
_
and
are the
z
z
t
t
are the variances.
The local measures of spatio -temporal autocorrelation can be derived by
decomposing a global measure into particular spatial neighboring units.
In the experiment section, we will evaluate how different spatial and
temporal neighbors (lags) affect the results of three spatio-temporal autocorre-
lation measures for the mobile phone call activities in a city.
and
means of variable
Z
within a time lag, while
3.
D
ATA
P
ROCESSING
In this research, the dataset contains a week of about 74, 000, 000 anony-
mized mobile phone call detail records (CDR) in a large city from a Chinese
telecommunication operating company. The CDR data lists the information of
caller, receiver, mobile base stations, date, time, duration et al. (Table 1).
As shown in Figure 4, every time when a user (caller/receiver) made a
call, he/she was geo-referenced to a corresponding mobile base station that has
a unique longitude/latitude position. The coverage area of each mobile base
station can be expressed as a Voronoi polygon for call activity analysis and
termed as a ―cell‖.
In this Voronoi partition, all phone calls within a given polygon are closer
to the corresponding mobile base station than any other station.
Generally, urban central regions have a higher density of mobile cells (the
coverage area of each cell is smaller) than the outer suburb regions.
Different data subsets have been extracted and processed for various
research purposes such as human mobility modelling (Kang et al., 2012b),
