l.-F. Khreim and B. Lacaze: Land Cover Change Detection in the Mediterranean Area
203
This radiometric correction was however not easily feasible with the data produced
by ESA, so that we decided to apply only a scene-to-scene radiometric normalization using pseudoinvariant features. This technique is based on the assumption
of constant reflectance selected in-scene elements. Differences in the grey-level
distributions of these invariant objects are assumed to be a linear function and are
corrected statistically to perform the normalization.
The 1986 image was selected as reference to normalize the other two images. An
image by image geometric correction was then applied to make them
superimposable. The 1977 and 1993 images were corrected relative to the image
of 1986. Three false color composite images of the study area were realized
(bands MSS 7,5,4 in RGB) (Fig. 1,2,3). Visual analysis of these images allows
one to recognize the main changes that occurred during the 1977-1993 period,
such as the urban growth of Lattaquie city, the variations in hydrographic outlines,
especially for the AI Kebir stream after a dam building, the decrease in land
vegetation cover in the North-Western part of the study area as a consequence of
forest fires.
The normalized difference vegetation index was calculated for each date. In order
to avoid the negative values of the NDVI, we chose to use the following formula:
255*NIRlNIR+R <=> 255*MSS7/MSS7+MSS5.
Therefore, the values vary between 0 and 255 instead of -1 and + 1.
The vegetation index images were visualized by monochromatic display, where
NDVI values are in the range (0-255), dark grey indicates low NDVI values
whereas light grey indicates high values (Fig. 4). The visual comparison of these
three NDVI images can give indications about the biophysical changes for the
1977-1993 period. Subtracting a pixel's NDVI value on one date from the
corresponding pixel NDVI on another date constitutes another way by which to
quantify and locate NDVI variations (Fig. 5).
But the interpretation of such documents is often difficult since grey levels are
hardly perceptible by human vision.
4
THE RGB-NDVI COMPOSITE IMAGE TECHNIQUE
In order to facilitate the interpretation of the changing NDVI values for each date
and to study changes that occurred in the 1977-1993 period, a simple and logical
technique was applied to visualize changes by combining the NDVI calculation
and RGB image display functions. In this study, NDVI of 1977, 1986, 1993 were
assigned to red, blue and green respectively (Sader and Winne, 1992).
In the resulting RGB-NDVI color composite image (Fig. 6), white, black and grey
indicate no major change in NDVI values between the three dates while colors
illustrate change in vegetation cover. For example, magenta is created by a high
NDVI value 1977 and 1986 and a low value in 1993 (green biomass reduction);
green is created by a low NDVI intensity in 1977 and 1986 and a high value in
1993; yellow represents high NDVI value in 1977 and 1993 and low in 1986. For
more details, the reader is referred to Table 1.
203
This radiometric correction was however not easily feasible with the data produced
by ESA, so that we decided to apply only a scene-to-scene radiometric normalization using pseudoinvariant features. This technique is based on the assumption
of constant reflectance selected in-scene elements. Differences in the grey-level
distributions of these invariant objects are assumed to be a linear function and are
corrected statistically to perform the normalization.
The 1986 image was selected as reference to normalize the other two images. An
image by image geometric correction was then applied to make them
superimposable. The 1977 and 1993 images were corrected relative to the image
of 1986. Three false color composite images of the study area were realized
(bands MSS 7,5,4 in RGB) (Fig. 1,2,3). Visual analysis of these images allows
one to recognize the main changes that occurred during the 1977-1993 period,
such as the urban growth of Lattaquie city, the variations in hydrographic outlines,
especially for the AI Kebir stream after a dam building, the decrease in land
vegetation cover in the North-Western part of the study area as a consequence of
forest fires.
The normalized difference vegetation index was calculated for each date. In order
to avoid the negative values of the NDVI, we chose to use the following formula:
255*NIRlNIR+R <=> 255*MSS7/MSS7+MSS5.
Therefore, the values vary between 0 and 255 instead of -1 and + 1.
The vegetation index images were visualized by monochromatic display, where
NDVI values are in the range (0-255), dark grey indicates low NDVI values
whereas light grey indicates high values (Fig. 4). The visual comparison of these
three NDVI images can give indications about the biophysical changes for the
1977-1993 period. Subtracting a pixel's NDVI value on one date from the
corresponding pixel NDVI on another date constitutes another way by which to
quantify and locate NDVI variations (Fig. 5).
But the interpretation of such documents is often difficult since grey levels are
hardly perceptible by human vision.
4
THE RGB-NDVI COMPOSITE IMAGE TECHNIQUE
In order to facilitate the interpretation of the changing NDVI values for each date
and to study changes that occurred in the 1977-1993 period, a simple and logical
technique was applied to visualize changes by combining the NDVI calculation
and RGB image display functions. In this study, NDVI of 1977, 1986, 1993 were
assigned to red, blue and green respectively (Sader and Winne, 1992).
In the resulting RGB-NDVI color composite image (Fig. 6), white, black and grey
indicate no major change in NDVI values between the three dates while colors
illustrate change in vegetation cover. For example, magenta is created by a high
NDVI value 1977 and 1986 and a low value in 1993 (green biomass reduction);
green is created by a low NDVI intensity in 1977 and 1986 and a high value in
1993; yellow represents high NDVI value in 1977 and 1993 and low in 1986. For
more details, the reader is referred to Table 1.