04 
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXX V, Part B4. Istanbul 2004 
  
airphotos, in order to detect mixed pixels or spectral 
signature confusions. Finally, a confusion matrix was 
computed. The correspondence between interpreted 
orthophotos and maximum likelihood classification has 
been performed for each sampling point. The 
percentage of coincidence between 'ground data' and 
classified IKONOS image can be measured either as the 
number of coinciding points derived from the sum of 
the confusion matrix principal diagonal versus total 
number of points (1286) or with Kappa 
coefficient. Taking into account the results obtained 
through photo-interpretation results, the legend is 
improved, new training zones are defined (with their 
area in proportion with the occurrence of the class), 
then a new classification is performed (Fig. 4). 
2.5 Discussion of Results 
The 1286 plots were interpreted on the 
orthophotos, using the 12 items of the classified 
IKONOS imagery. There was a problem of 
interpreting *Broad-leaved forest under smoke? 
(class 4) versus *Broad-leaved forest'(class 3) and 
‘Phrygana under smoke’(class 6) versus Phrygana 
(class 5) since the air-photographs were acquired 
before the forest fire while the IKONOS imagery 
was acquired after the fire. We decided not to use 
the classes 4 and 6 (classes 4 & 3 were merged 
under the label 3, and classes 6 & 5 were merged 
and labelled 5).Another classical problem was related 
to the refinery area, which is a restricted zone. The 
Hellenic Military Geographical Service erased this area 
from airphotos. Pixels belonging to each class on 
IKONOS image were compared with the air photos in 
order to detect mixed pixels or spectral signature 
confusions. The number of well classified points versus 
the total number of points is 752 out of 1286, giving an 
agreement of 59%. Kappa coefficient is K = 0.471. 
The training areas have been revised, taking into 
account the general remarks on confusions between 
some classes (see previous paragraph) as well as 
orthophotos interpretation on the sample points. The 
revised legend is the following : 1.Pine Forest (Pinus 
halepensis), 2.Dense Forest (Q.coccifera/Phyllirea 
media), 3.Broadleaved Forest/Bush (Arbutus Erica), 
4.Broadleaved Forest/Bush  (Arbutus/Erica) under 
smoke,  S.Phrygana,  6.Phrygana under smoke, 
7.Firescars, 8.Crops,orchards, 9.Bare soils (croplands, 
tracks, rocks), 10.Highway, plant, buildings, 11. Water, 
I2.Shadow The result were shown on the classified 
image.It has to be noticed that classes 10 and 11 
(* Highway‘ and “ Plant, buildings *) were finally 
merged; class 6 (* Phrygana under smoke) is often 
classified as * Phrygana * (class 5), as the light smoke 
plume does not strongly influence the Phrygana spectral 
response ; class 12 (* Water *) is unclassified because a 
mask had been previously applied on sea; some 
confusions remain with shadows (class 13). Finally, the 
number of well classified pixels versus total number of 
pixels is 263 982 out of 329 900, giving an accuracy of 
80.02% ; overall Kappa coefficient is K=0.6609. 
Considering that classes “ Water “ and ‘ Unclassified ‘ 
may be merged due to the mask, the agreement 
becomes 88.12%. 
     
  
633 
3. CONCLUSION 
The groundtruthing reliability depends on the correct 
location of the sampling points. Provide orthophotos on 
a convenient scale are available, time and cost can be 
saved by using these data in place of field work. In the 
given example, the classification accuracy was 
noticiably improved (59% to 88%) by this method. The 
resulting classified image of this 5km x 5km area gives 
a realistic view of the vegetation conditions in the 
Greek mediterranean coast. In the surroundings of an 
important port served by a highway all along the coast, 
there are many settlements with accompanying 
croplands. The pine forest on the lower slopes is widely 
moth-eaten by new settlements. Behind, the bush with 
Arbutus and Ericaceae members has been degraded and 
replaced by a low discontinuous phrygana. The impact 
of fire is conspicuous on this scene. In the upper zone 
only, on the Yerania slopes, the bush is still present, as 
well as the Quercus coccifera forest, above 200m 
elevation. 
4. REFERENCES 
BARBAROUSSI, V. 1988.  Spatiocartes et 
cartographie de la végétation méditerranéenne : 
application en Grèce (Corinthia). Thèse Université. 
Toulouse 3 (France), 93 p. 
BARBERO, M. et QUEZEL, P. 1976, Les 
groupements forestiers de Gréce centro-méridionale. 
Ecologia Mediterranea, n?2, 89p. 
CODJIA, C.J, et DOMINGO, E., 1998. Méthode 
pratique d'amélioration des classifications dirigées dans 
l'étude de l'habitat de la ville de Cotonou (Bénin) La 
réalité de terrain en télédétection : pratiques et 
méthodes, 13-17 Ocober 1997 (Montréal : AUPELF- 
UREF), pp.21-27. 
DEBAZAC, E.F., et MAVROMMATIS, G., 1971. Les 
grandes divisions écologiques de la végétation 
forestiére en Gréce continentale. Bulletin de la Société 
Botanique de France, 118, 429-452. 
FONTES, jJ, ESTEVE, P, et GASTELLU- 
ETCHEGORRY, J.Ph., 1998. Validation de résultats de 
traitement d'images spatiales appliqués à l'étude de la 
dégradation des écosystémes ligneux secs du Burkina 
Faso. La réalité de terrain en télédétection : pratiques et 
méthodes, 13-17 October 1997 (Montréal : AUPELF- 
UREF), pp.135-143. 
FRANKLIN, S.E., WULDER, M.A., and GERYLO, 
G.R., 2001. Texture analysis of IKONOS panchromatic 
data for Douglas-fir forest age class separability in 
British Columbia. International Journal of Remote 
Sensing, 22, n?13, 2627-2632. 
HOTYAT, M., 1998. De la donnée satellitaire à 
l'analyse stationnelle : recherche méthodologique pour 
une approche de la forét francaise. La réalité de terrain 
en télédétection : pratiques et méthodes, 13-17 October 
1997 (Montréal : AUPELF-UREF), pp. 55-61. 
IMBERNON, J. 2003. Image IKONOS et étude 
environnementale. Camps de réfugiés du HCR en