Summary, Concluding Remarks and Recommendations - Historical Land Use/Land Cover Classification Using Remote Sensing - page 200

evaluation (see Chap. 5.6 ); unsupervised classification (see Chap. 5.7.1.1 );
supervised classification using the three algorithms of classification (i.e., MLC,
NN and SVM) with the two approaches of classification, i.e., the one stage and the
multi stage classification approaches (see Chap. 5.7.1.2.1 ); post-classification
processing (see Chap. 5.11 ); automated change detection mapping using the pre-
classification approach (see Chap. 5.12.1 ) and the post-classification approach (see
Chap. 5.12.2 ) ; and finally, the accuracy assessment techniques (see Chap. 5.13 ).
The new relative radiometric normalization method that was used in this study,
was, after Canty et al. (2003), favored, where it can be applied automatically. It is
consistent, constant, rapid, parameter free and sensor independent, and is enhanced
by an orthogonal regression.
The aforementioned and used techniques in this study have various alternatives
of sub-technique (e.g., radiometric normalization can be performed using more
than one method, such as 6S, dark object method, histogram matching, etc.) and/or
various parameters (e.g., SVM-algorithm of supervised classification can be used
with various of parameters combinations). Thus, some of these alternatives were
mentioned, discussed and compared to justify the final choice of each alternative
technique and/or parameters that were used in this study.
This study proved that the use of multi-sensor (MSS-1975 and TM-2007) and
multi-scale 60 and 30 m data for change detection mapping is possible. Also, it is
possible to use the multi-sensor ASTER-2005 and LANDSAT-ETM+ data for
LULC-classification.
The available remotely sensed data of ASTER-sensor with low spectral reso-
lution (three bands) and high spatial resolution (15 m) had given worse results
with lower classification accuracy than those obtained after fusing with the data of
LANDSAT-ETM+ to increase the spectral resolution.
New sensors (e.g., ASTER) offer higher accuracy rather than the old sensors
(e.g., LANDSAT-MSS), but also bring new problems, such as the increasing time
of processing because of the higher spatial resolution and the lower local coverage
of each scene, and the increase of geometric errors because of the higher spatial
resolution/pixel size.
Remotely sensed data spatial resolution/scale affects the level of useful infor-
mation that can be extracted from the satellite imagery.
7.3 Recommendations/Outlook
Remote sensing techniques and data (LANDSAT and ASTER) were found to be
very effective in the classification of land use/land cover and in the mapping of
irrigation areas, and to detect and map changes that occurred over a number of years
in the arid and semi-arid area of the Euphrates River Basin in Syria. However, these
approaches were uneven for classification of agricultural crops, where their effec-
tiveness were based on many factors, such as the used remotely sensed data type
and its characteristics (e.g., spatial and spectral resolution, etc.), the agricultural