ıhul 2004 
cy of 85- 
ails were 
nge: The 
to GIS 
ses were 
ectories: 
 landuse 
hat there 
s so that 
of 19735. 
e 2 for 
—^G— 
und as 
in 1976, 
vated as 
2000 
  
  
are ground 
fied for 
onment, 
generic 
changes 
ch as G 
dicating 
le point 
ncludes 
puilding 
: human 
ories of 
>G—> 
change 
natural 
ng. For 
er and 
strong 
stare G 
nd G > 
ng the 
le point, 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004 
  
the landuse classes were all confirmed by the five-date ground 
references, the case was regarded as ‘true’ trajectory, otherwise 
it is a ‘false’ case. We have chosen a stratified random sampling 
scheme for selecting sample points of reference data for 
trajectory accuracy assessment. 790 sample points were 
generated using the method as reported by Zhou et al (2004). 
Table 3. Classification of landuse change trajectories. 
being affected by the spatial resolution of data and they 
can be used with care. 
3. If the change of metrics from one time to another is not 
stable, and, though related to the spatial resolution 
obviously, the metrics for the same spatial resolution are 
not comparable, the metrics cannot be used. 
Table 4. Spatial statistics for analyzing spatial patterns of 
landuse changes. 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Level 1 Level 2 Description Trajectory 
classes classes examples 
Unchanged | Grass/wood No change 6.265626 
land 236 
Salty grass No change 5-35-5555» 
S 
Water body No change W->W-—>W->W 
—W 
Bare ground | No change B—>B—B—>B— 
B 
Stable Old Changed to and G—>6>6>C—> 
cultivation remained as C 
cropland since 1994 
New Changed to and S2S56—6G 
cultivation remained as C 
cropland since 2000 
Abandoned Revered from G—>G—>G—C— 
cultivation cropland to other G 
classes in 2000 
Reservoirs/p | Changed to and G—>G>W—W 
onds remained as water —W 
bodies since 1986 
Unstable Grass/woodl | Periodical changes 655566 
and between cover Gand | S 
S 
Flooded Periodical changes G>W-—>G->W 
between cover W 6 
and other types 
Bare ground | Periodical changes G->B—>B->G-> 
between cover Band | B 
other types 
2.3.3 Analysing spatial pattern: The spatial pattern 
of landuse influences the ecological process of movement 
of matter and energy. The spatial pattern of landuse and 
land cover has been actively researched in the field of 
landscape ecology (Miller e/ al 1998, Farina 1998). In this 
study we have selected five variables to analyze landuse 
patch characteristics and landscape patterns. Table 4 
summarizes the computation and interpretation of these 
variables. 
Spatial pattern is different from the area statistics and temporal 
trajectories because the effect of errors cannot be detected from 
ground references and their corresponding statistics directly. 
Generally, landuse pattern parameters should reflect the overall 
trend of landuse change, so that they should not show acute 
fluctuation over long time series. We therefore propose to use 
the time series to assess the effect of multi-resolution imagery. 
|. If the change of metrics for every landuse class from onc 
time to another is stable, the metrics are comparable and 
can reflect the regularity of land use spatial pattern change. 
These metrics can be regarded as a metrics that can be 
used and are not affected by resolution of remote sensing 
data. 
If the change of metrics for every land use class from onc 
time to another is not stable, and obviously related to the 
spatial resolution of the data, these metrics are regarded as 
Na 
  
  
  
  
  
  
  
Abbreviation | Name Equation* Interpretation 
PPU Patch Per Ppu =! Fragmentation of 
(Frohn 1998) | Unit E74 area pattern, with 
higher values 
indicating more 
fragmented areas. 
PAFD Perimeter- PAFD Complexity of 
(Saura and Area p-keg^ area shapes, 
Martinez- Fractal ranging between 
Millán 2001) Dimension 1 and 2 with 
higher values 
indicating more 
complex shapes. 
MSI Mean Ep Irregularity of the 
(Saura and Shape > Ja, shapes, with the 
Martinez- Index MSs XS minimum value for 
Millán 2001) 4m perfect square 
shapes. 
SD Shannon A E Variety and 
(Farina 1998) | Diversity | SP=-2,2 m2 | relative 
i abundance of the 
cover type, with 
the higher values 
indicating more 
diversified 
landuse. 
DI Dominanc DI = Inn — SD | Dominance of one 
(Farina 1998) | e Index landuse class 
over the others, 0 
«D «1. 
  
  
  
  
699 
* where: m = total number of patches of the class of interest; A 
= total area of the study area; p = perimeter of class of interest; 
k = constant; a = area of each class of interest; n = total number 
of classes; and P is the ratio of a class area to the total area, 
which reflects relative importance of landuse types. 
3. RESULTS AND DISCUSSION 
3.1 Area statistics 
From the area statistics (table 5) some major changes of landuse 
can be observed in the past 30 years. For cropland, its area 
increased from 496 on the 1994 TM image to 13% on the 2000 
ETM image. For grass and woodland, its area decreased 5-7% 
from 72-75% on early MSS to 60-66% on later SPOT. TM and 
ETM images. For salty grassland, it decreased from 5-7% on 
the MSS and SPOT to 3% on the TM and ETM images. For 
water body, it accounts for 8-10% on the MSS, 18-23% on the 
SPOT and TM, and 9% on the ETM data. 
In the study area the large-scale reclamation started in 1992 and 
increased very rapidly in the past decade. This is confirmed by 
the increase of cropland shown on the TM and ETM images. In 
early 1980s, a large reservoir formed in sand dune area because