Research Methodology - Historical Land Use/Land Cover Classification Using Remote Sensing - page 80

LANDSAT-ETM+/SLC-Off_Corrected/ data
(the bands: 1,5, and 7_30m)

TERRA-ASTER data (the bands: 1,2, and
3_15m)

Gram Schmidt
Spectral
Sharpening-
Algorithm

Fusion (sharpening) using the panchromatic
band (8)_15m

Geometric correction/GCPs

Subset all images (to remove the distortion
on the margins)

Radiometric adjustment (atmospheric
correction/ATCOR-2 + radiometric
normalization/iMAD

Radiometric adjustment (atmospheric
correction/ATCOR-2 + radiometric
normalization/iMAD

Geometric
regestration
(image to image)

Mosaic image (3 bands with 15m)

Mosaic image (3 bands with 15m)

Radiometric
normalization
(iMAD)

Layer stacking

ETM+_p173r035-scene_(1,2,3,4,5, and 7
bands)_15m (cover the west area, that
ASTER-data not covers it)

Fused data set (mosaic)

Subset (ERB-
borders)

The final fused data set (cover the all study
area)

Mosaic

Fig. 5.11 Fusion- and mosaic-concept for LANDSAT-ETM+ and TERRA-ASTER data set,
acquired in May and August 2005

The merged data were fit for further digital classifications, since the spectral
separability for LULC- and crops- classes/six spectral bands of the merged data
was better than the spectral separability of the original data. This was because
there were only three spectral bands. Therefore, merged images were used for
visual interpretation and for features extraction (classification).
Figure 5.11 explains the concepts followed to generate the final fused and
mosaiced data set of the ERB-borders based on TERRA-ASTER & LANDSAT-
ETM+ images.
The ASTER data did not cover the entire study area, only the first three bands
with a resolution of 15 m. This data was before tested on the separability among
the extracted classes of interest from the study area and compared to the same
ASTER data after they were merged with the three spectral bands (1, 5 and 7) of
LANDSAT-ETM+. However, these three bands were useless in the classification
process due to their low separability when compared with the results of spectral
separability that resulted after fusing with the other three bands of ETM+ (see
Tables 5.4 , 5.5 ). Therefore, ETM+-scenes which were corrected SLC-off data
were used. This data had similar temporal coverage to the ASTER data. The idea
was to increase the spectral resolution which in turn, increased the spectral sep-
arability between classes. These offered classification results with higher accuracy
rather than using only the three spectral bands of ASTER data.
Because the ASTER data had a spatial resolution of 15 m, and in order to
benefit from this to compare results with the results of MSS-60 m, and TM-30 m,