ng
un
ed
se
on
re
lle
th
ly
je
es
Dy
W
nt
1e
pe
ch
d.
ge
pe
to
1e
le
60% is either not possible at all or only with an
unsatisfying degree of precision. One of the reasons for
this is that the potential differences in reflection caused
by needle loss are superimposed by the reflection of the
ground vegetation, which strongly influences the
signature in open stands. However, it can be assumed
that these stands show strong damages and thus belong
predominantly to the category C3 (damaged trees more
than 66%). It can be assumed that C3+ will often be find
in the deforestation class 1 (solution of damage degree
"improvement" after damaged trees are taken out) and
C3- in the deforestation class 2. Thus a subdivision into
different needle-loss categories is not necessary.
Disadvantages of the damage class definition
As mentioned above, it is not possible to isolate the
defoliation symptoms from picture elements integrating
defoliation as well as deforestation features. A precise
separation of defoliation categories is only possible with
the classification within one deforestation stage.
Unfortunately, these both effects may superimpose each
other and thus lead to a less satisfying classification
result of the needle loss.
The problem of recognising damages caused by air
pollution among clearcuts, windthrow areas or other
damages leading to a decreasing crown density cannot
be solved if information on regular cutting operations and
on disaster (storm, snow, ice, insect attacks etc.) is not
available.
A decreasing canopy density leads to an increasing
influence of ground vegetation on the reflection of forest
stands. The typical reflection characteristic of forest
stands can be modified strongly by different types and
phenological stages of ground vegetation. Detailed field
information and knowledge about the reflection
characteristic of the forest floor is necessary to estimate
the influence of ground vegetation on the classification
accuracy.
7.4 Classification
In the final phase of the project the damage-class
definition had to be tested. It must be made sure that
only those methods are applied which are also
operational for small-scale classifications (total test area),
since the classifications in the pilot studies were carried
out for smaller investigation areas only. This is
particularly important for the integration of ancillary
information available for smaller test areas but not
covering the entire region.
It is not necessary to use the same classification
algorithm for all the test areas as long as it is ensured
that the classification of forest damages is carried out
according to the damage classes mentioned above.
739
7.5 Verification
The verification of the classification result must be
directed towards the "Ground Truth" available for the test
area (e.g. aerial photographs or field experiments).
Methods therefore cannot be standardised.
Irrespective of the method applied, it must be ensured
that for all the verification areas in the entire test region
the damage symptoms "needle loss categories in
percentage" are recorded and "crown density" according
to the percentage of canopy density. Only then there can
be a guarantee that the classification of the damage
class definition described above can be verified. One of
the verification results is to be the analysis of the causes
of mis-classifications so that the classification results can
be analysed in detail by the user.
8. RESULTS OF THE STUDIES IN THE TEST SITES
The final image classification and the presentation of its
results was performed in all test sites following the
common methodology approved by the Advisory Board of
the Large Area Experiment for Forest Damage
Monitoring in Europe using Satellite Remote Sensing.
This methodology was based on the suggestions of the
TWG.
The results of the forest damage classification were
printed on coloured maps in scale of 1:100,000 (Polish
Sudety Mountains, Krusne Hory, Harz) and 1:200.000
(Fichtel- and Erzgebirge). The results of the particular
projects are summarized below. Detailed information on
the investigations in the test areas can be derived from
Henzlik, 1995; Zawila-Niedzwiecki, 1994; Reuter &
Akgóz, 1995 and Scharat et al, 1995:
Necessity of ancillary information
The investigations in pilot studies showed that the
integration of a DEM and forest stand parameters
derived from digital forest maps increased the
classification accuracy only slightly. The digitisation of
forest maps is very time consuming. Digitising large
amounts of these data just for the purpose of optimising
satellite classification is not advisable and thus not
feasible for small scale classifications. The thamatic
information of forest maps should only be integrated into
the classification process if it is already available in digital
format.
However, the integration of additional information is
necessary in order to seperate forest from non forest
(open stands «2096 crown closure). This can be derived
either by using the digital forest layer of topographic
maps or by classifying a forest mask through multi-
temporal classification of Landsat-data.
Signature analysis
All pilot studies stated that the influence of stand density
on the spectral signature of Landsat TM data is high. The
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996