e medium: borders clearly visible but not obvious; only small
but still distinct differences to the neighbour stand; reduced
visibility in the stereo model e.g. edge of the photo.
e Difficult: no clear border lines; no obvious difference to the
neighbour stand; bad visibility in the stereo model e.g.
hidden position due to exposition.
3.3 Selection of the test objects
For the testsite in Germany 5 stereo models, which is 25% of
all existing stereo models have been selected randomly. Within
the 5 stereo models all border lines of the different forest units
have been visually classified into one of the three different
complexity classes.
For the Italian testsite a stratified sampling was necessary, due
to the fact that the area is very inhomogeneous. Therefore three
geomorphologic units were first delineated and then in each
subunit the samples randomly selected. One subunit was
represented by 2 stereo models, while each of the other two
subunits were represented by 3 stereo models. The total of 8
stereo models was again 25% of all existing stereo models. The
border lines of all forest units in the selected stereo models
were again visually classified into one of the three complexity
classes.
3.4 Delineation procedure
In order to value the delineation accuracy 10 testpersons were
selected which provided the same education background and
had all basic photogrammetric knowledge. Five persons even
had an extensive experience in aerial photointerpretation and
photogrammetric measurement. To achieve standardisation for
the delineation an interpretation key was developed to provide
information on
the subject of delineation
the photomaterial
the test area
the stands which had to be delineated
the rules of delineation and
the successive steps for interpretation
Important for the comparative valuation of the delineation
results were the rules for the delineation. The rules comprised
for both testsites the following points:
® the stands had to be delineated according to their
homogeneity in tree species and age,
® the delineation line between unequal high stands or
between forest and non forest had to be located in the
centre of tree crowns of the higher stands,
e for equal high stands the delineation line had to be located
in the centre of the deepest visible point between these
stands,
e discontinuities had to be taken into consideration for areas
with more than 30 m width (approximately two tree
crowns,
e in case only tree species are changing areas with more than
60 m width or at least one hectare size had to be taken into
consideration.
For the interpreters it was not necessary to identify the type of
tree species but they only had to register the change in tree
species. Aside to the stereo model each interpreter had a paper
copy of the infrared airphoto, where all stands which had to be
delineated were marked by numbers. Each interpreter had to
delineate the same stand twice. During the process of
delineating the interpreter was able to see his delineation in the
stereo model through the VIDEOMAP device.
3.5 Calculation of the delineation accuracy
Due to the assumption that all delineated lines represent
equally the truth, there is no ‘true line’ to which the test
delineations can be referred. On the other hand this means that
all lines represent a certain inaccuracy. In order to quantify this
inaccuracy all delineation lines were first transferred into
ARC/INFO GIS and reduced to x and y co-ordinates. Due to the
deviation between the delineation of different interpreter the
stand border was then represented by a net of two-dimensional
lines. After the lines were framed by a minimum and a
maximum polygon a measure of the variation width was
possible. For the calculation of the delineation variation subsets
out of the polygons in form of polygon pieces were selected
representing equally each complexity class. In order to quantify
the variation of all delineations each selected polygon piece
was buffered starting one time from the maximum polygon
piece and second time from the minimum polygon piece. The
buffers were step by step enlarged till all existing delineation
lines in that polygon piece were included. For each buffer step
around the maximum polygon respectively around the
minimum polygon the length of delineation lines included was
calculated. This calculation provided a distribution of line
length starting from the minimum respectively maximum
polygon piece. It was then assumed that the buffer section with
the highest density of lines, respectively the section with the in
total longest line distance, is the buffer section with the highest
probability to represent the true stand border line. After the
distribution was then transformed in that way that the buffer
section with the maximum length of lines was the reference the
demonstration and calculation of
line length over distribution width (buffer section) became
possible.
4. RESULTS
All together 36763 m delineated stand border was evaluated.
63.7% of the border lines were classified ‘easy’, 20.3% were
classified ‘medium’ and 16% were classified ‘difficult’.
4.1 Evaluation of the 1:7000 aerial photographs
All together 7169 m of stand border was delineated in the large
scale photographs. Table 1 gives the distribution over the
different complexity classes.
Tab.1: Evaluated border length in each complexity class (scale
1:7000)
Complex |testarea | |testarea2 |testarea 3 | total
class
easy 1168 m 976 m 2144m
medium 693 m 1170 m 596 m 2432 m
difficult 968 m 626 m 999 m 2593 m |
The investigation shows for the complexity class ‘easy’ 80% of
the delineation lines within 2 m distance and 100% within 51
(Fig. 1). For the complexity class ‘medium’ 100% of the
delineation lines are included within 20 m, but already 59% of
the lines are included within 5 m and 9196 within 11 m. (Fig.
1). For the complexity class ‘difficult’ all delineations are
within 57 m (100%), but already 99% are within 41 m, 90%
are within 27 m and 60% of all lines are within 12 m (Fig.l)
The wide distribution of lines for the 10096 in the complexity
356
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996
1 inae im narrant
