Geography Reference
In-Depth Information
2005
). Scan statistic is also generalized to space-time to handle ST event report
datasets (Kulldorff et al.
1998
; Kulldorff
2001
).
The computational efficiency was further optimized by Neill and Moore (
2004
)
by employing a top-down pruning of the search space. A Bayesian version of scan
statistics is also proposed where an inference is used instead of the frequentist
hypothesis testing (Neill et al.
2006
). A further step was made to model the
emergence of an outbreak inside a spatiotemporal subspace (Neill et al.
2005
). The
null and alternative hypotheses were modified as follows: H
0
: the probability that
one instance is at risk in the cluster for each day (
t
min
,t
min
C
1
,
:::
,W
) is the expected
value p, and H
1
: The above probability in days
t
min
,t
min
C
1
,
:::
,
Tare
q
p
where
1
q
1
q
2
:::
q
T
. Computational methods were also provided for the efficient
computation of this pattern.
17.5.6
Remote Sensing Image Change Detection
Techniques of change detection in image processing for remote sensing have been
extensively discussed in the literature. Typical methods on bi-temporal change
detection include simple differencing (Rosin
1998
; Rosin and Ioannidis
2003
),
where a value difference between pixels at the same location from two images
are compared with a threshold for change determination, hypothesis testing, where
the null hypothesis of no change is tested against the alternative hypothesis of
change using significance test or likelihood ratio test (Barkat
2005
; Aach and Kaup
1995
; Thoma and Bierling
1989
; Dempster et al.
1977
; Fukunaga and Hayes
1989
).
Another approach for image change detection is predictive modeling (Hsu et al.
1984
;Yakimovsky
1976
; Elfishawy et al.
1991
; Jain and Chau
1995
).
17.5.7
Opportunities for Cross-Fertilization
ST cluster detection techniques are mostly applied on disease mapping and crime
analyses. We argue that other application domains can also benefit from these
techniques as they may also face the problem of cluster/hotspot detection. For
example, a pattern of extreme rainfall change at regional scales due to global
climate change is of interest to climate science research. To do this, the underlying
statistical models may need to switch from Binomial or Poisson distributions to
those modeling tail behaviors of continuous variables, such as the Generalized
Extreme Value (GEV) distributions (Coles
2001
), and the spatial dependencies
among locations may need to be addressed.
The concept of return period (
Wikipedia 2012-R
) is used to characterize
the expected years before seeing the next climate events (rainfall, temperature)
of the same intensity. A return period of a rainfall event is 100 years means that the
probability that the intensity of this rainfall is exceeded in any year is 1/100
D
0.01.
Search WWH ::
Custom Search