1 
as 
en 
ler 
ge 
as 
us 
NO 
yer 
ck 
ja 
  
For this reason the fragmentation index, originally 
intended as a measure of pattern complexity for 
choroplethic maps (Monmonier, 1974) was implemented 
and applied as a region-based GIS operator (Johnsson, 
1995). The fragementation index is computed as: 
FI = (M-1)/(N-1), 
where M = number of image regions in the categorized 
image, and N = number of pixels in the categorized 
image. The Fl is computed for each map polygon in the 
base map (figure 5). The Fl was implemented in a raster- 
based GIS. 
  
Classified remote sensing data 
) 
  
  
  
NEN VER 
NEN WER 
  
Base map Value map 
Each region uniquely numbered 
N 
VOS 
Output map 
Fragmentation index computed 
for each region in the base map. 
Values stored as region labels. 
(Note: Values above are hypothetical) 
Figure 5. Hegion-based fragmentation index (FI) 
computations in a raster GIS. 
The spatial pattern of landuse categories is being 
recognized as increasingly important within landscape 
ecology and several indices have been developed to 
capture significant spatial patterns (e.g. McGarigal and 
Marks, 1994). These are designed to work on 
categorized image data, and would provide another 
source for qualitative image generalization functions. 
4. POLYGON-BASED CHANGE DETECTION 
- A CASE STUDY 
4.1 Background 
The case study relates to forest management. The study 
was carried out at the Pacific Forestry Centre in British 
Columbia, Canada, within the framework of the SEIDAM 
(System of Experts for Intelligent Data Management) 
project.” The issue was to develop and test a method to 
  
SEIDAM is a project under NASA's Applied Information Systems 
Research Program. The SEIDAM Project is also supported by 
Natural Resources Canada, Industry and Science Canada, the BC 
Ministry of Forests, the BC Ministry of Environment, Lands and 
Parks, and the BC Forest Resources Development Agreement. 
387 
automatically extract clear-cuts from a Landsat TM 
image for revision of forest inventory maps in a vector- 
based GIS. 
Clear-cut detection by multi-spectral image classification 
yielded unsatisfactory result due to the natural 
heterogeneity of the mountaineous test area, and due to 
confusion with other features of similar spectral 
characteristics, such as rock outcrops and roads. 
Instead an approach was adopted that relied on 
comparison of existing forest density data in the GIS 
data base with meausures of forest density computed 
from the image data. 
The study has been described in detail in Johnsson 
(1994). 
4.2 Material and methods 
The method was developed and tested on a digital forest 
inventory map in scale 1:20,000 (BC Ministry of Forests, 
1991), covering an area of approximately 11x14 km 
(figure 6). The digital forest map consists of map 
polygons, which correspond to forest stands (forest 
management units). Each forest stand has a number of 
attributes, stored in a separate attribute database. 
  
Landsat TM 
Forest inventory map 
NS 
Th 
Figure 6. Study area and data 
A subsection of a Landsat TM scene was used as image 
material. The temporal difference between the map and 
the image was approximately a year, during which logging 
was known to have occurred. 
The forest database attribute crown closure provided an 
estimate of expected forest density for each forest 
stand. Crown closure is defined as the percentage of 
ground area covered by the vertically projected crowns 
of all living, commercial tree species in the main canopy, 
rounded off to the nearest 10 % (BCMoF, 1992). The map 
contains forest stands with crown closure ranging from 
0% to 65%. 
A normalized difference image of Landsat TM4 and TM5 
(ND45) was computed according to: 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996