International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
is focused on dealing with the finite time series, while
conventional theories of time series analysis deal with the
infinite time series.
In the second step, the detected changes in the N"' observation
were discriminated into stationary change and nonstationary
change by comparing them with the change process model
calculated in the first step. Actual observed values of newly
(N^) acquired data were evaluated by comparison with the
predicted value and its deviation (extent of the allowable error).
Recalculation after adding the newly acquired data into the
archive was performed when the change was discriminated as
stationary change. This recalculation improved precision of the
“base fluctuation" model. When the change was discriminated
as nonstationary change, stationarity was recalculated by setting
the newly acquired data as the default value. The method was
able to discriminate the changes into stationary change ((i) in
the Figure 4) and nonstationary change ((ii) and (iii) in the
Figure 4), while conventional methods interpret them as the
same type of change.
D
Stationary * :
Change c Yo
N-1 N
(ii) =
Nonstationary 2
Change = ;
ea Be
N-1 N
Figure 4. Evaluation of stationarity
3. RESULTS
Figure 5 shows the temporal behavior of the backscatter and
stationary change process models in the sample area. Figure 6
shows the area where nonstationary change was detected
through 1992-1998, as extracted by the method developed in
this study. The precision of the detection was evaluated by
comparing the existing spatial data and to the result. In this
study, existing land use data and vegetation data were used as
examples of spatial data which need updating. Figure 7 shows
100m meshed land use data (1991, 1997) and 1:50,000 scale
vegetation data (1994, 1997). These are prepared nation-wide
and are maintained by the national institute. The frequency of
updating them is quite low (i. e. several years). Changes were
observed in these data, and samples of the changes are also
illustrated in Figure 7. Evaluation was especially focused on the
following areas:
540
- Areas where change was detected by both existing data
and the method developed in this study
- Areas where change was detected only by existing data,
- Areas where change was detected only by the method
developed in this study
3.1 Areas where change was detected by both existing data
and the method developed in this study
Area A is an area which changed from bare land to buildings.
Both subtraction of existing data and the developed model were
able to detect the nonstationary change. But it is difficult to tell
*when" this change occurred from the subtraction of existing
data. On the other hand, the result of the developed method
suggests that the change had occurred after 14 May 1997.
Compared to the base fluctuation, deviating elements were
stable until 14 May 1997, and the profile indicates a different
stationary pattern after then. This result suggests that the
nonstationary change occurred between 14/May/1997 and
27/Jun/1997. Several more places which were indicating
obvious nonstationary changes, such as the construction of new
bridge and deforested areas were found in the study site.
3.2 Areas where change was detected only by existing data
Area B is a place which changed from bare land to a university
campus. However, this change was not detected by the method
developed in this study, while subtraction of existing data did.
The reason was that opening of the campus (it was opened in
1990) was not reflected in either the land use data of 1991 or
the vegetation data of 1994. It is clear that “time inconsistency”
(problem (i) in the Introduction), is embedded in these data.
w
= T T "T
EL T = $$.
+ 2, 1
v A
« ,
- . 4
e 2 o Go
sr : © 3
= x 9 1
= :
:
:
:
: 5; 1
SE : o 2 4
a 6 * «|
o^ >
[ Co 99.9 © ane a BM
on ; TO T5 5 a
5 ve Waeso JV 7900 o 3 2 | ud
1 ; | ^ ;
121 1/1 ctu, vtt
1833 1995 1337 Fast
| Area A
|
|
|
FT T T T 3
SI xl
ei: e 9 1
© [ 5
o S.
sk E o 6 o c 1
SL 3 $
SI o 9 9 c o
| o o o o
> o sie a °°.
Sr o2 E
2 o o
to o
e
o o 1
etr 1 geo oz | ; ; Ÿ : : xz]
121 1/1 1/1
1993 1395 1997
Area B
1500€
T
Figure 5. Temporal behaviour of backscatter (point) and
stationary process (line)
Internal
In this
land us
etc.) ca:
(ie. lan
occurre
This pr
The me
some e»
in 3.1 al
