ul 2004
able 6).
acy (all
ectories
ation of
location,
ontrary,
Tequent
indaries
S
y cases
iy also
landuse
| Combi-
nations
2a was
ted for
'ea has
: stable
which
hanged
stable
| while
t since
ge and
anduse
ould be
stories,
higher
ange.
indices.
varies
r time
s. after
of the
greater
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
landuse diversity, resulted in increasing SD and decreasing DI
from 1973 to 1986. From 1994 to 2000, the landuse diversity
increased with the large scale cultivation, resulted in higher SD
and lower DI from 1994 to 2000. Note that the SD and DI
showed reverse trend during the period of 1986 to 1994, while
decreasing SD and increasing DI were observed. This could be
largely affected by the increasing number of landuse types
(from four to five due to the addition of cropland). It is,
therefore, reasonable to conclude that the temporal change of
SD and DI indicates the trend of landuse diversity and they are
not sensitive to the spatial resolution of remotely sensed images,
but could be affected by the changing number of landuse
classes.
Table 7. Metrics of all landuse classes for five-stage data
Metrics Classes 19073: 1976 — 1986 .— 1994 - 2090
Cropland - - - 1.515 — 1.799
Grass/ woodland | 0.189 0.584 2.525 0884 1278
PPU [Salty grass 0331 1042 22073 0.378 0.884
Water body 1136 2987 7.623 1.484 0.694
Bare ground 0.521 1200 1673 0816 0915
Cropland - - - 1.230 1.255
Grass/ woodland | 1.307 1.345 1.304 1328 1.358
RS PAFD |Salty grass 1.216 1224 1301 1237 1.241
Water body 1.202 4309 1239 1.308, 1.3967
Bare ground 1.120: 1.249 1255 1.300 1284
Cropland - - - 1874 2.614
Grass/ woodland | 6.325 7.741 13.690 13.828 11.096
MS! |Salty grass 1551. 1.867 2729 2304 4.812
Water body 21713) 2998 4451 3007 3204
Bare ground 2088 2786 3706 3018 3274
Pattern {SD | 0.823 0.889 0.996 0.882 1.084
mecs ly ^T 0563 0498 0390 0728 0526
In this study, the PAFD values did not show significant
variation between landuse classes. In general, the cropland
showed generally less complex shapes than natural land cover
types, particularly grass and woodland and water body, but
difference in PAFD values are quite small (about 0.10 — 0.15).
This result also shows that the PAFD is not sensitive to the
spatial resolution of the images.
By definition the MSI is related to spatial resolution. The higher
is the spatial resolution, the more details on the object shape are
revealed. and thus the higher MSI is observed. This study
confirmed the assumption as the MSI showed generally higher
values on SPOT, and decreasing values on ETM, TM and MSS,
closely related to their spatial resolutions. Comparing landuse
types. croplands obviously showed the least irregularity
suggesting the fundamental difference between human-induced
landuse and natural land cover patterns. An exception, however,
was observed on salty grass that showed less MSI than the
cropland in 2000. This could be due to that only a small area of
salty grass had left after the reclamation in late 1990s.
In this study the result of PPU showed some effects of spatial
resolution indicated by general higher value on the SPOT image.
However, this pattern was not well supported by the other
evidences. The PPU values did not show a recognisable pattern
701
in relation to landuse types, nor on spatial resolution. As a ratio
of patch numbers and area, PPU can largely be affected by a
number of factors including spatial resolution of the image,
classification accuracy and post-classification sorting methods,
thus the real spatial pattern that may be revealed by PPU could
be well masked.
According to the above analysis, it is suggested that SD, DI and
PAFD are not sensitive to the spatial resolutions of multi-sensor
images, while MSI is closely related to the spatial resolution.
All these four indices have demonstrated good usability as
indicators of spatial pattern of landuse/cover types in this study.
PPU, on the other hand, did not present itself as a reliable and
meaningful indicator for the spatial pattern analysis in this
study.
Benson and MacKenzie (1995) and Frohn et al (1998) stated
that spatial resolution had important effect on most landscape
metrics. However, Wickham and Riitters (1995) stated that
landscape metrics should not be dramatically affected by the
change in pixel size up to 80m. These results appear to be
inconsistent. However, taking into account the difference of the
metrics discussed in their works, the results of this study
confirm the previous work to some extent. For example, the
metrics discussed by Benson and MacKenzie (1995) were
percent water, number of lakes, average lake area and perimeter,
the fractal dimension, and texture, of which some are similar to
MSI in principle, thus it is understandable that the influence of
spatial resolution was emphasised. On the other hand, Wickham
and Riitters (1995) used DI so that it is expected that his finding
is confirmed by this study.
4. CONCLUSION
The ability to discriminate the landuse/cover types varies
significantly for multitemporal images because of various
spatial and spectral resolutions of images acquired by different
sensors. The area statistics are capable of showing the general
landuse change trends, but the uncertainty caused by area
fluctuation due to classification errors may play a significant
role to create misleading results. In this study, the poorer
classification results were found in association with lower
spatial resolution, demonstrated by the higher fluctuation of
area statistics results. Concentration on the detected change area,
e.g. the temporal trajectory analysis, therefore, seems to be a
better and more promising approach.
In the past 30 years, less than 40% of the study area was
unchanged, while the stable and unstable change accounted for
more than 6096. Unchanged trajectories show the original
condition of land cover: stable change trajectories show most
human-induced changes; while unstable change trajectories are
relatively less significant since they tend to show natural (i.e.
reversible) land cover change and also contain most of
classification errors. Therefore, for landuse change detection
study, we recommend that the focus should be on the analysis
of unchanged and stable change trajectories, especially the
stable change trajectories.
For spatial pattern analysis, this study suggests that SD, DI and
PAFD are not sensitive to the spatial resolutions of multi-sensor
images, while MSI is closely related to the spatial resolution.
Al these four indices have demonstrated good usability as
indicators of spatial pattern of landuse/cover types in this study.
PPU. on the other hand, did not present itself as a reliable and