Due to the high spatial variability inside the cork oak forest of Ben Slimane, and because the aerial photos dated back to 1985, it was not possible to identify all the forest land units in relation to the land cover on a working scale, through photo-interpretation. For this reason the adopted methodology was focused on the multitemporal analysis through the enhancement of the 32-bits NDVI values in April and August 2000. The multitemporal analysis permits detection of variations occurring throughout the year and the adopted technique was based on the temporal image differencing (Lillesand and Kiefer, 2000).The saturation produced by herbaceous layers in natural land units induced high DN values in Band 4, present all over the image so that January NDVI wasn’t taken into consideration for the analysis of the woody cover.
After the field work it was possible to recognise three main different kinds of forest formation. The starting point was to discard the cork oak cover as a discriminating parameter. The range of this cover varies from 5 to 60%, with an average of 30.
The variation of the second layer of shrubs could be detected in the satellite image as influencing the NDVI values in the two months. The presence of a dry and close second layer of Cistus monspeliensis (with a cover greater than 50%), together with a variable cover of the other shrubs, induces in the August 432 false colour composite Landsat image, a variation to the brown colour of the forest red colour because of the lower reflection in Band 4. The subtraction of August NDVI DN from April NDVI DN has a high value that allows identification of these land units.
When in the second layer evergreen shrubs like Rhus pentaphyllum, Pistacia lentiscus, Olea oleaster and Myrtus communis predominate, the colour variation is lower.
The last identified land cover is the forest formation characterised by a cork oak cover greater than 40% and without a second layer of shrubs. This type of land cover has a very low NDVI difference due to the approximately constant vegetation reflectance in Band 4.
In all the three forest types the third layer influences the DN fluctuation, especially Asphodelus microcarpus that is very vigorous in February and March and dries out in August.
In order to obtain the final land units of the Ben Slimane forest, the methodology described below was applied.
The subtractive image between April and August NDVI showed higher DN values where an higher reflectance variation occurred, such as in the close Cistus monspeliensis cover.
Working with a worksheet (Table 9) it was possible to establish a relationship between the field observations on cork oak, Cistus spp., shrubs and NDVI variation; the scatter plot between vegetation cover and NDVI differencing values is shown in Figure 9. As formerly stated, the fluctuation is mainly related to the predominance of Cistus spp. that, which dries out during summertime and so does not photosynthesise therefore reducing the reflectance in Band 4. The NDVI value was calculated through an average (3*3 pixels) around the relevé location.
|
Quercus cover (%) |
Cistus cover (%) |
Shrubs cover (%) |
NDVI differencing |
|
45 |
3 |
2 |
23.50 |
|
30 |
25 |
25 |
31.00 |
|
30 |
60 |
30 |
33.75 |
|
20 |
40 |
50 |
38.50 |
|
30 |
15 |
5 |
40.50 |
|
15 |
35 |
40 |
43.50 |
|
25 |
85 |
20 |
60.00 |
|
30 |
70 |
10 |
64.00 |
|
40 |
80 |
50 |
65.25 |
|
30 |
85 |
15 |
66.50 |
|
40 |
60 |
20 |
66.75 |
|
35 |
50 |
45 |
74.75 |
|
5 |
85 |
30 |
75.25 |
Table 9 - NDVI values
Figure 9 - NDVI differencing scatter plot
The three classes identified (Figure 10) have the following values ranges:
Through the Pseudo Colour Transformation (PCT) the subtractive NDVI image was classified in relation to the three classes that were found. A better and clearer image was obtained after a 5*5 median filter was applied to reduce the noise. The final thematic map was then checked and digitised.
- bright green: Cork oak forest without undergrowth
- dark green: Cork oak forest with Mediterranean shrubs
- brown: Cork oak forest with Cistus monspeliensis
Figure 10 - Final result of the NDVI multitemporal analysis in the forest of Ben Slimane