International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
J image section of 3
Gifereim daiess
3*3000*620 pixels
the multi dates base of
training 300 600 pixels
the referents map of
Kobonen
Application of HAC
classification of the map
base af training
fassified
Figure 3. Methodology of the application of SOM algorithm
4 RESULTAT
The classification results was validated, on each date, by a
comparison between the calculated spectra of the referents and
the theoretical spectra and with those of the ground for each unit
of landscape. Validation on the ground is carried by the
localization of the class pixels on a georeferenced image
compared to ground data taken on this area.
The other method of validation is made by comparison with the
digital map obtained by photo-interpretation and from the
ground data. This map of validation covers 300*600 pixels. The
difficulty of classification lies especially in the separation of the
various existing sand classes on the ground. It's question to
make separation from the various sand classes of dunes and to
separate also from the sand classes and the coastal dunes.
Algorithm SOM presents a clear improvement compared to the
k-means algorithm, in particular for the classes having high
reflectances (classes of dunes)
Fixed yellow 13.52 78.97 57.43
dunes
costal dunes 17.08 74.26 6235
Deposit of dune 15.32 71.51 58.91
Inter dune 23.68 3.14 63.52
Nouakchotian 6.74 72.43 66.03
Surface
Lagoon of the 13.31 72,34 70.39
Aftout
Wetlands of the 9.68 78.12 71.47
lagoon
zone of water 11.5 82.71 76.11
resurgence
Dense Urban area 14.13 71.83 68.72
Vegetation 7.56 81.01 80.34
Table 1. Results of classification by SOM algorithm
The application of this method, on the totality of the study zones
on the three dates, made it possible to characterize 71% of the
units of landscape of the environment for three dates. Table 2
summarizes the changes of surface states in the three dates, year
89, year 95 and the year 98. We note a strong increase of the
urban areas (180%).
For the Eastern part of the city; even if the large "dunes of
Erguc" did not change appreciably, there is a movement of
sands in the South-western North-eastern direction. It appears
by the growth of deposit layer of dunes and thus by the
reduction of 55% of "inter dunes" class and in 43% of
"nouakchotian area" class.
In the East of the city "the flooded zone" of the lagoon
decreased by the transformation of the pixels into "wetland zone
of lagoon" and into "zone of turbid water 1” (see Figure 5,
wetland zone of lagoon) following the strong hydrous erosion
exerted by the sea in the south of the port.
area %
Class Name year 89 year 95 year 98
zone of breaking of
waves 2,09 1,74 2,17
FOR Turbid water | 8,92 6,55 0,90
_ Turbid water 2 8,04 10,81 6,86
Mo us E US | : _impid water 7,63 7 12 17,27
Figure 4. Classification of the validation zone 06/06/98. Lama wg =
Caption is same then that of Figure 5 caption Wetland 1 1,38 0,55 0,67
Wetland 2 1,99 2,36 3,34
For the class "dispersed urban zone" the percentage of the Wetland 3 2.13 4,47 5.97
classified pixels is lowest whereas the percentage of the false Flooded zone 0.29 0.30 0.24
classes is high. That is explained by the use of the dunes sand in
like materials of building (the sand is extracted from the dunes). Urban zone 4,74 7,50 11,19
See Table 1. Costal dunes 1,20 7,88 6,13
Old dunes 3,48 4,59 4,92
CU Eisen SOM = Ergue dunes 11,85 7,61 10,39
/ eor Kr - S E
Algorithme ‘algorithme means ier dros 4.93 3.30 2.69
Error, class, ‚| classed classed Nouakchotian soil 8,34 4,18 3,62
zone of breaking | 5.8 80.31 74.37 cost arca 0,49 0.78 0,43
of waves |
x > S 2e se 2
Zone humide de la | 9.68 78.12 71.47 Deposit of dunes | 2,32 3S 2:46
lasune Vegetation 0,78 1,14 0,54
Dispersed urban 28.31 69.74 67.58
| area Table 2. Results of classification on three dates of images
506
Inte
Th
res
col
lan
thi.
4.1
Th
ch:
val
to
cli
