cale 
rea: 
th a 
ions 
| in- 
d in 
rvey 
ence 
with 
d on 
sual 
o for 
ality 
ition 
reta- 
sing 
hree 
ring 
map 
,odel 
sure 
rials 
third 
liffe- 
rpre- 
r the 
nd 5 
  
  
  
  
  
  
aerial 
ı this 
Since the old map series does not contain clearings, the 
old aerial photos (1981, altitude 3000 m) was essential 
for the up-dating of clearings in this study. 
In the test area, which was chosen for its extended 
forested areas, there was little or no built-up area to 
revise. Built-up areas were therefore not interpreted. 
The evaluation made in this study was based on general 
experience from SPOT images. 
4. RESULTS 
4.1 General 
Air photos have a higher geometric resolution than 
satellite images and are for that reason preferred by the 
interpretors. The NLS does not have so much experien- 
ce of multispectral satellite images and these are in 
some cases considered too blurry at the interpretation 
scale 1:25 000. To fully benefit from these images the 
interpretation methodology needs to be further develo- 
ped. The geometric position accuracy, however, is very 
high in the SPOT images. 
4.2 Adjustments to the interpretations 
Since the new aerial photos in the study and the 
satellite images differed one year, all changes between 
1989 and 1990 were disregarded. 
4.3 Accuracy types and measures 
Three different accuracy types were defined: 
- the completeness, i.e. how many roads, clear- 
tings etc. were detected? 
- the shape consistency of the features. An erro- 
neous shape of a clearing is for instance caused 
by including an area of sapling stands in the 
clearing. 
- the geometric accuracy in the delimitations. 
This depends on the geometric resolution of the 
sensor and preprocessing of the image, and of 
the sharpness or contrast of the particular 
entity. À road is for instance expected to have 
a higher geometric accuracy than a clear-cut, 
depending on narrow zones of deciduous vege- 
tation often delimiting the clearings. 
It is difficult to distinctly separate the accuracy types 
from each other. A clearing only partly interpreted will 
be considered to have an erroneous shape. Alternatively 
it could be regarded as two entities of which one is 
correct and the other wrong (completeness). Also the 
geometric accuracy and the shape could be discussed. 
Including a larger fuzzy area of vegetation surrounding 
a clearing in the interpretation is here considered as 
influencing the shape and not the geometric accuracy. 
The following accuracy measures has been used for the 
above accuracy types. 
Completeness: 
2 x #correctly interpreted clearings 
#clearings in new map + #clearings interpreted 
  
559 
The formula, which was used for clearings, takes into 
consideration both the number of clearings in the true 
map and the total number of clearings in the interpre- 
tations. Thereby, both errors of omission and commis- 
sion are taken into account. 
For roads, a revision almost exclusively means new 
roads, and only errors of omission is likely to occur. The 
number of interpreted new roads compared to number 
of new roads in the true map were therefore used. 
Shape consistency: A subjective visual measure was 
used. For clearings for instance, only those who signifi- 
cantly differ in size compared to the true map was 
regarded, i.e. double or half the size. 
Geometric accuracy: A visual comparison was made to 
the true map. 
4.4 Roads - completeness and shape consistency 
The number of new roads in the test area is only about 
20. No significant difference between the interpretations 
in the different image materials are found. However, 
roads on the actual clearings is found to cause some 
problems in SPOT XS. This is probably not the case 
while working with digital images, since these can be 
contrast stretched interactivally in the different parts 
to enhance different features. The photographic images 
in this study, were stretched in regard to the forested 
areas. 
4.5 Roads - geometric accuracy 
Using the pen chosen by the interpretors for the draf- 
ting (approximate width one pixel), no difference of 
significance to this application is found between the in- 
terpretations. 
4.6 Clearings - completeness and shape consistency 
The results of the interpretations are shown in table 2a 
and 2b. The total number of clearings in the area is 
over 200. 
The results seem to indicate, somewhat surprisingly, 
that the approximate accuracy in revision/interpretation 
of clearings is 65-70%, independent of image material. 
The relative accuracy between interpretation # 1 and 2 
(two interpretors, same area and material) is a little 
higher, around 80%, indicating a high agreement bet- 
ween different interpretors. 
Interpretation # 4 is an outlier, where the interpretor 
obviously has been very careful in the classification. 
Out of 50 interpreted clearings, 49 are correct. 
The interpretations of clearings in the study were 
naturally only made within forested areas of the map. 
In the topographic map, clearings are defined as 
forested areas with tree heights less than 1.5 meters. 
The misinterpretations are therefore mainly areas 
within the zone between clearings and forest, i.e. 
around 1.5 m. It is a known fact that, even in aerial 
photos (9 200 m), it is not possible to reliably interpret 
the 1.5 meter limit.