International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004 
This definition allows for the explicit determination of the 
change components, that is, the additions 44, and deletions 
Del 
Ada 7 $2- (64082) 
D = 81-8; x53) 
Therefore, an addition is determined as the difference between 
the new dataset and the common elements between the two 
temporal datasets, while a deletion is determined as the 
difference between the old dataset and the common elements 
between the two temporal datasets. The common elements of 
the two datasets are determined as the spatial intersection set of 
the two temporal datasets. 
A detected spatial change could be caused by differences in 
positional accuracies between the two datasets. The significance 
of change can be expressed based on accuracy tolerances and 
minimum sizes. To account for positional inaccuracies, 
appropriate spatial buffers are generated around the two 
temporal features during the change detection operation, while 
the minimum sizes satisfying the specifications are handled 
using appropriate spatial filters. The buffering and filtering 
operations are used to keep only the actual changes. Whatever 
vector segments are outside the buffer zones are considered as 
changes. If the new features from the S; data are outside the 
buffer applied to S, features, changes are considered as the 
actual additions. If the old features from the S, data are outside 
the buffer of the S; features, changes are considered as the 
actual deletions. 
3. CASE STUDY I: WATER BODY AREA 
EXTRACTION FROM LANDSAT 7 IN NORTHERN 
CANADA 
The Landsat 7 ETM+ ortho-images of 15m and 30m spatial 
resolutions constitute the image layer of the Canadian national 
data framework. Under the Geomatics for Northern 
Development Program of the Earth Sciences Sector (ESS) of 
NRCan, particular attention has been given to the mapping of 
Northern Canada both for the completion of the 1:50 000 scale 
coverage and for the updating of existing decades-old data. 
Water body areas, such as lakes and rivers are the most 
' predominant features in this northern region of the country. The 
plethora of water bodies covering the northern areas 
necessitates the need for rapid approaches for their recognition 
and extraction. A common approach to automatically extract 
water bodies is by using land cover supervised or unsupervised 
classification. However, the existence of other features with 
similar reflectance, such as glaciers, ice caps, wetlands and 
shadowing from the mountainous terrain leads to low 
separability of the thematic classes and increases the confusion 
level of the results. This results in low accuracy classification 
and low reliability requiring afterwards significant amounts of 
interactive editing. 
A semi-automated approach for the extraction of water features 
from Landsat 7 ETM+ imagery has been applied based on 
image processing and GIS tools combined with a spatial 
constraint. The approach is based on the principle of 
determining and establishing conditions that uniquely 
characterized the water bodies in order to increase the success 
of recognizing and extracting these particular features from the 
Landsat 7 ETM+ imagery. The conditions established are based 
614 
on: a) the spectral properties (digital numbers) of the water 
bodies as they recorded in the various bands, b) the notion that 
water bodies are located in areas with zero or minimum terrain 
slope, and c) the intersection of conditions-derived spatial 
layers using the AND Boolean operator. 
      
> Mr uem + 
fe tt née se 
v A 4 e 
Ca. PAR. MR a ru. date 
Figure 2: ISODATA 
clustering using Landsat 7 
bands 4,5,7. 
Figure 1: Thresholding on 
Landsat 7 band 5. 
  
Figure 4: Terrain constraint 
Figure 3: Spatial intersection 4: 
(slope between 0-4 degrees). 
of thresholding output and 
water class from ISODATA. 
  
  
  
Figure 6: Commission and 
omission error areas (shown 
in orange). 
Figure 5: Extracted water 
bodies. 
The water bodies recognition and extraction approach is based 
on the application of threshold, spectral, spatial and Boolean 
operators. First the image was edge sharpened. A thresholding 
operation was applied on the band 5 based on initial reading of 
the digital numbers of sample of water bodies and the band 
histogram followed by a median filter for noise reduction (Fig 
Inter. 
  
1). A 
band 
(Fig. 
outpt 
areas 
An a 
small 
were 
degre 
consi 
of sl 
water 
deter 
comn 
not) : 
for th 
Moni 
of ve 
vulne 
chang 
Progr 
invest 
regior 
other 
assess 
optior 
and ir 
coasts 
neces: 
and b 
chang 
south- 
Harbo 
aerial 
The C 
semi-e 
systen 
coastli 
  
Figure