reas and ifs
oeen done in
fine the ric
thern part of
terested area,
the auxiliary
'S selection.
k space and
imation. It is
ring and the
y 30 minutes
oncerning 10
| activities.
s prepared
d bv geo-
2) covering
ftware for
)n respec
for ground
3. METHODOLOGY
Concerning to the scope of project (crop hect-
garage estimation / cultivation mapping), the
procedure could be explained in three stages as
below:
- Satellite data processing / classification.
- Rice cultivation mapping.
- Ground samples checking using GPS.
3.1 Satellite data processing
31.1 Since the TM images used in the project
werc raw data, using 1:50,000 scale topomaps
and collection of sufficient GCP(s), the TM
images were geometricsally corrected and the 8
quarter scenes were mosaicked.
3.1.2 Using 1:50,000 scale administrative maps
the boundary of Gilan province was transfered to
the previously corrected images and the surr-
ounding part was ommited.
3.1.3 Concerning to the noticeable forest area in
the mentioned province and no rice cultivation
activities in this part, the forest area was ommited
from the data file to decrease the disk space and
computing time.
3.1.4 Concerning to the reed-beds and scattered
small group of trees within the agricultural areas,
a primary unsupervised classification was
performed on the TM data to define the whole
agriculture area that mainly comprise the rice
cultivation.
3.1.5 The obtained results of former stage were
ground checked and the proper parts were
selected to be used as training sites for further
classifications.
3.1.6 Using the data obtained from unsupervised
classification and the areas defined as training
Sites, a supervised classification was performed
on the raw TM bands and spectral separability of
ice against the other crops/vegetation was
evaluated.
31.7 To raise the spectral separability of rice
against the other vegetation, different data mani-
Pulation inclusive of Spectral Ratio, Principal
461
Component Analvsis, filtering and the like were
applied and the best data combination (two PCA
and three TM bands) with the maximum spectral
separability was selected.
3.1.8 Using the results of the former stage, the
final supervised classification was done and the
three different kinds of rice eventually were
united as total rice class in the Gilan province. At
the end of this stage the total area of 193557
hectars was defined as rice hecterage.
It 1s mentioned that, regarding to the relative con-
fusion matrix, an accuracy of 94% was resulted
for the above rice hecterage.
3.2 Rice cultivation mapping
As explained before, Gilan province is divided
into 11 townships and the total rice production
Obtain from these townships. Therefore, after
completion of the mentioned project, using GIS
facilities, 1:100,000 scale rice cultivation maps
relative to the different townships with 30 by 30
minutes of geographic coordinates were
produced.
Since the rice distribution was concerned as main
class to be depicted, on the produced maps, the
other vegetation is shown as "others". The
mentioned maps not only are being used for
agricultural planning / crop monitoring but would
be concerned as an important part of information
needed for agricultural database of Gilan.
3.3 Ground samples checking using GPS
The TM digital data applied in the project were
not the real time data. Therefore, to assess the
accuracy of obtained results the following
manner was applied:
3.3.1 Randomly selection of 5x5 pixel windows
(Ground samples) on the classified images and
extraction of centre coordinates of the windows.
3.3.2 In this stage, the constant geodethic
points(Bench mark) installed by Ntional Carto-
graphic Centre of Iran and by differential
positioning method, the ground samples were
correctly located and percentage of various types
within the samples were assessed. At the end of
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996