2004 International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
———
a inherent spectral characteristic compared to July image. This 5
Were introduced a reduction in the number of subclasses for each 3k
class and therefore, an improvement was observed in the sí
— accuracies of several classes and in the overall accuracy. The T
+
overall accuracies of the classified original bands and the
optimum four bands were quite close. The accuracy of the
S classified PCA bands was found to be 69.294. In August, the
overall accuracy of the parcel-based classified original bands
was 70.8%. After applying the regional masking in the
cem classification, the overall accuracy was computed as 73.894.
Applying the multi-temporal masking technique together with
the regional masking to original bands (1 to 5 and 7) of the
+ August image provided an overall accuracy of 81% (Table 2)
The improvement in the classification accuracy was gained
A 2 =
due
ato
over
urc
M Uncultivated/Bare soil
=
Bl Pepper
B Water melon
2] Cauliflower
Bb Ton:
through the masking of the classes which were accurately
puc classified on previous dates (May and July 2000). The resulting
classification accuracy of the multi-temporal masking technique
was 10% more accurate than the parcel-based classification of
qe the original bands of the August image only. The
multi: misclassifications might have been caused largely by spectral
confusion which was not overcome through parcel-based Ë
analyses, multi-temporal analyses, available ancillary data and g
using remotely sensed data in hand. 5 :
tep of #2
ivated August Parcel-based Multitemporal a £j
à with Producer's User's Producer's User's 3
output Class Acc. % Acc. % Acc. % Acc. %
gu : 2
cation Corn 76.0 87.4 77.4 86.2 E <Z
ımmer Residue 702. 0:989 933: 4000» f|
Tomato 85.4 72.3 86.6 773 RS
Sugar beet 60.4 95.3 80.0 100.0 5
Nr
Clover 50.0 100.0 100.0 100.0 Figure 4. The multi-temporal classification output of the
Pasture 100.0 40.0 100.0 66.7 August image.
tant =
Bol m Pepper 50.9 32.9 52,8 38.4
is the CONCLUSION
S Watermelon 57.1 80.0 6413. 750
»pared, Unc/Bso 22.4 42.9 57.1 23.5 Nowadays, crop estimation and monitoring become one of the
ses for Rise 90.0 75.0 80.0 100.0 major applications in remote sensing. In this study, mapping of
Cauliflower 571 27.0 531 81.0 the summer (August) crops was performed using a multi-
Overall 70.8 813 temporal masking technique together with the parcel-based
à 'erall : ; anii sit = erede cs :
of the T ; ; e : i oi ; analyses of the classified images. The discrimination of the
: l'able 2. Producer's and User's accuracies for each class, for : >
at was : gL agricultural land cover classes in the study area, Karacabey,
: parcel-based and multitemporal analysis of August : : P ; *
ccurate was carried out using variations in spectral responses
E srt
originali bands. :
classes = s throughout the growing season. The proposed methodology
y were ; ; nn includes a synergistic combination of maximum likelihood
dv As a summary, the accuracies of the parcel-based classified u^ ;
c study : : : : classifier, parcel-based analyses, and the multi-temporal
^q August images were found to be higher than the July images. It : ;
of the oa e masking technique.
reels IS not logical to make a comparison with the May image since
i oh the May image contains significantly less number of classes
The limitation of the spatial resolution of Landsat 7 ETM+ was
than the August image. The classified original all bands à sci ; ;
to be © dis s = : overcome through fusing the multispectral bands with the
; provided higher overall accuracies than the other classified ; Hp : GNE
overall | = ar : j : panchromatic band. The spectral variations introduced within
em fh band sets. However, several classes that provided low accuracy - E . :
rom the nil s 2 ; : agricultural parcels were resolved through incorporating the
ins {X m the classified original all bands yielded higher accuracy in ; : ; : :
les 10 th a A im : e spatial context kept in a GIS vector polygon layer. The
at and € other band sets. For example, the classified PCA channels s a >. i
5% an f the . : classification accuracy attained from the parcel-based
B Of the May image yielded an accuracy of 100 percent for iati ed ane . D(^
ng, n Clover. While the f Boel à t hieher] classification was improved after introducing the PC channels
fi ‘> White the accuracy of sugar beet was 2 percent higher in : ;
assified the classified PCA icr i e: | fi d pen | il b and the best four bands as well as masking the selected crops
; assifie AA bands than the classified original all bands : ; ; ; ;
bserved for Jui f e. classitied origina for which the phenological evolution were taken into
ywer by fus. consideration. The class labels were assigned to each
agricultural parcel in a field specific manner in the integration
of remote sensing and GIS. The accuracy of the final classified
y image data was increased from 70.8% (for parcel-based classification
* eleven only) to 81.3% for the 2000 summer crops (August)
( of the classification.
erefore,
of their
1089
