International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
V he arcas of the two catchments from each satellite image have 
heen extracted using overlay masks, The catchments are further 
subdivided into three elevation zones. The. software for the 
above purpose has been developed locally, Figure 7 présents 
the extracted elevation zones of Cordevole river basin from 
standard false color composites acquired on different dates, 
ZONE-1 
£ONE-3 
gE HIT 
20 (WR Remnte Serzıng Dept, Oo 
  
Figure 7 Black and white representation of extracted elevation 
zones drom Standard False Color Composites of Cordevolc 
river basin, 
5.1 Classification of multi-spectral data 
Ihe classification problem has been subject of much rescarch 
during past three decades, Classification. is essentially the 
process of carrying out a mathematical transformation from one 
data space to another. In the remote sensing context this is from 
image digital count. values to. map classes relevant lo the 
application of interest. While certain map classes may bc 
clearly distinguished in the image data, there is often no clear 
relationship between the image digital count values and thc 
map class on a pixel basis, The conventional classification 
process involves the division of the feature space defined by 
the raw or derived data features into disjoint regions cach of 
which corresponds to a separate map class. 
Multi-spectral classification of remote sensing images bas been 
dominated by hard algorithms that produce onc class per pixel. 
An example is the level slice algorithm that uses parallelepiped 
regions in feature (multi-spectral} space. À pixel is labeled with 
the class name corresponding to the parallelepiped that contains 
the pixels feature vectors (Schowengerdt, 1996), Another non- 
parametric method of classification is the Euclidean minimum 
distance classifier. Approaches based on statistical analyses of 
the data and derivation of parameters of distribution 
characterizing each map class in the data set are also available 
and known us parametric methods. A good example of this 
approach is the maximum likelihood classifier. It bases its bard 
decision on a comparison of a posteriori probabilities among 
the candidate classes. All the non-parametric and parametric 
classifiers give different classified information. Hence in the 
present study the classification of the digital satellite data has 
been made using the non-parametric and parametric classifiers 
vis, parallelepiped. minimum distance and maximum 
1224 
likelihood. classifBiers to. categorize snow covered area into 1) 
two classes as snow and snow free arca as well as il) three 
classes as snow. transition snow and aper. The suitability of 
these methods for snow melt runoff forecast has been studied. 
Figure 8 shows examples of the supervised classification of 
elevation zones extracted from various sutellite scenes far Eu 
Vizza catchment using Euclidean minimum distance classifier 
The area covered by each class in each elevation zone has heen 
determined by counting the number of pixels classified under 
cach category and multiplying them with their respective pixel 
dimension. The mixed snow cover representing the transition 
zone was given a weighting of 50% and added to the pure snow 
to obtain total snow-covered area for each elevation zone. 
LAVIZZA CATCHMENT 
Supervised Classification 
Minimum Distance 
LR L-$ 1-4 me : 181 
EN DE d 
e £2 
RTT PET ELIT 18 Jy 1984 
À 
  
Figure 8 Black and white representation 01° multi-spectral 
classification of various elevation zones of La Vizza basim 
using supervised minimum distance classilier 
6. DEPLETION CURVES 
The rate of snowmelt during summer months varies from onc 
catchment to another in different high mountainous regions of 
the world. depending on the prevailing meteorological 
conditions, Consequently the snowmelt depletion curves reflect 
the seasonal decrease of the snow cover as it is influenced by 
the dominant meteorological factors like temperature and 
precipitation. The spatial distribution of the snow cover is 
described by snow cover depletion curves, which summarize 
the percentage areal coverage of the snowpack. 5now cover 
depletion curves have been developed lor. and applied in, 
hydrological models on a watershed or. elevation zone basis. 
Watershed-wide snow depletion curve relationships are used in 
lumped hydrological models such as the National Weather 
Service River Forecast System (Anderson, 1973) to describe 
the snow cover distribution as the snow melts, while elevation