XXIX-B3, 2012 
Te practiced per year, 
-, SUgarcane, cassava, 
[ profile by the end of 
planted once a year in 
vegetation activity in 
tivating season due to 
ncluding high albedo 
1g lots, and roads) had 
lation throughout the 
le 
  
+ 
61 
f the 2010 classified 
accuracy was 80.3% 
total of 2,000 pixels 
lass, the two classes 
were single-cropped 
rigated rice (91.9%). 
| for field crops class 
jlassified into other 
as more difficult to 
of upland crop fields 
red. Moreover, the 
he discrimination of 
   
  
1 
i 
  
0 100 200 400 Kilometers QE 
MM À———À ; 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B3, 2012 
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia 
  
Double-cropped rice 
Triple-cropped rice 
Field crops 
    
"1 Forests/orchards/perennial trees 
Built-up areas 
Water bodies 
Figure 3. Spatial distribution of cropping patterns. 
  
  
  
  
  
Classification results 
Ground reference (pixels) Single- Double- Triple- Fielderops Forests/ Built-up Total 
cropped rice cropped rice cropped rice orchards areas 
Single-cropped rice 1,774 6 11 195 34 183 2,203 
Double-cropped rice 0 1,838 372 76 126 294 2,706 
Triple-cropped rice 0 109 1,547 16 19 8 1,699 
Field crops 3 14 19 1,350 80 13 1,479 
Forests/orchards 211 26 37 363 1,735 100 2,472 
Built-up areas 12 7 14 0 6 1,402 1,441 
Total 2,000 2,000 2,000 2,000 2,000 2,000 12,000 
Producer accuracy (%) 88.7 91.9 73.38 67.5 86.75 70.1 
User accuracy (96) 80.5 67.9 91.1 913 70.2 97.3 
Overall accuracy (%) 80.3 
Kappa coefficient 0.76 
  
  
  
Table 1. Results of the classification accuracy assessment. 
6. CONCLUSIONS 
The objective of this study was to develop a classification 
approach for mapping major cropping patterns in Southeast Asia 
using time-series MODIS data. The data were processed using 
wavelet transform and ANNs. The results indicated that filtered 
NDVI patterns reflected temporal characteristics of different 
cropping patterns. This information was important for selecting 
training patterns used in ANNs classification. The ANNs applied 
to the filtered time-series MODIS NDVI data confirmed its 
validity for mapping cropping patterns in the study area. The 
results archived by comparisons between the classified map and 
the ground reference map indicated the overall accuracy of 
80.3% and Kappa coefficient of 0.76. The lowest producer 
accuracy was observed for the field crops class (67.5%) due to 
temporal confusion in discriminating between this class and 
other classes. The methods using wavelet transform and ANNs 
for mapping major cropping patterns in Southeast Asia from 
filtered time-series MODIS NDVI data could be transferred to 
other regions in the world to replace costly field investigations. 
REFERENCES 
Atkinson, P.M., Curran, P.J., 1995. Defining an optimal size of 
support for remote sensing investigations. /JEEE Transactions on 
Geoscience and Remote Sensing, 33, pp. 768-776.