IAPRS & SIS, Vol.34, Part 7, "Resource and Environmental Monitoring", Hyderabad, India,2002
8. CLASSIFY IMAGES USING NDVI PROFILES AND
KNOWN CROP CALENDARS
The approach used in $4 to stratify and differentiate relevant
vegetation cover profiles using 4-years decadal SPOT-4
Vegetation 1-km NDVI images, also offers options to identify
spatially areas having different crop calendars. Use of a (mosaic
of) single high-resolution image lacks the required temporal
information.
The relation and interpretation quality between country level
and local-level classified 1km NDVI time series is explored to
ascertain the link of both with crop calendar information.
For India, the available time series of images were subjected to
an unsupervised classification routine to define 30 classes
(Figures 9 and 11); the choice of class numbers relates with the
number of mixed categories by pixel that will be differentiated.
Similarly, also the time-series sub-set of W-Nizamabad was
subjected to the routine to generate 30 area-specific classes.
Profiles of the latter classes were visually compared and
generalized into 14 classes / NDVI profiles (Fig. 10 and 12).
May.Jun.Jul.'01 Aug.Sep.Oct.
s
Nov.Dec.Jan. Feb.Mar.Apr.'02
Figure 9. Location of Study Area in India on a 1km Resolution
Spot Vegetation NDVI Image (RGB - Feb, Mar, Apr’02).
| NDVI data from decadal Spot-Vegetation Images; 1 km pixels Original classes
—15
20
di! Dryland |___ 25
*
pF 1i. il : —27
\ ‘j AW
1 w^] \l—29
11745 ; ^: 3l —1,2,23
VEO TA QNM ufus i) 18,1924
KM . i
| | (| 2530
|
ng
100$ 2" VA
wi 3 / A| 212226
| p y | 14,16,17
| X i 34
50 r^ 8,10,12
i 6,7,9
Clougis
—5,11,13
Apr'98 Apr'99 Apr'00 Apr'01 Apr
Figure 10. NDVI Curves of 14 Cover Types in W-Nizamabad
based on 4 Years Decadal 1km Spot Vegetation Data.
4 Classes 10 km
Figure 11. Unsupervised Classified Image (30 classes) of Fig.9.
Enlargement of Nizamabad, and It if Classified Separately.
200 A fo)
NS
M ^. |; i
EE
' ops
b v
VN hdavy
Cloud
100 1
Z7
= Ÿ Clouds
/
X
VV Vc.
50 «——————- M LAM lul,
Figure 12. NDVI Curves of 14 Cover Types in W-Nizamabad.
The enlargements shown in Figure 11 show that the spatial
differentiation in classes is highly similar (colors are user
defined). The area specific classification shows however (after
merging classes from 30 to 14) a more likely, less-scattered,
spatial stratification with more mixed categories. Figure 13
shows that the latter interpretation is acceptable; note
specifically the location of paddies (bluish on the image).
Figure 12 simplifies Figure 10 and lends itself better for
comparisons with crop calendar information. Note that several
curves represent class ‘mixtures’ as they occur in a 1km grid.
1" 1 je pem M m o m
—
dad