6. METHODOLOGY
The Advisory Board appointed a Technical Working
Group (TWG), with scientists from the Czech Republic,
Poland, Finland and Germany to develop a methodology
and prepare guidelines for the harmonisation of the
various classification ~~ approaches for regional
applications. Their meetings took place in Harz
(Germany), in Brandys (Czech Republic) and in Warsaw
(Poland). The elaboration of the standardised
methodology was based on the exchange of experiences
obtained from national projects within the TWG meetings.
For the operational part of the project only the damage
on spruce were considered since this tree species is the
most-investigated so far.
The classification of forest damage on beech (Fagus
sylvatica) and Scots pine (Pinus sylvestris) is also of
great interest, because damages on these tree species
were observed increasingly. Since the investigations are
still in a research phase, the mapping of these tree
species were excluded from the experiment until the
study results within the national projects will lead to
operational damage classification procedures.
The recommendations of the TWG were structured
according the following issues:
e Choice of the appropriate remote sensing data
e problems of damage classification and role of
ancillary data
e harmonisation of damage class definitions according
to
* proposals of damage class definition to be applied in
the test sites
e classification of the satellite data
e verification of the results
Following these guidelines and recommendations, the
standardised methods had to be tested in smaller, yet
representative areas within the national projects. One
important factor that had to be considered in this
procedure was the feasibility to transfer the classification
methods to larger forest areas. The results had to be
presented in standardised maps with a harmonised
legend.
7. HARMONISED APPROACH FOR THE EXPERIMENT
The following section describes the results of the
discussions of the Technical Working Group (TWG).
Based on the findings of other research studies and the
experience from their own national projects, the
members of the TWG developed guidelines for the
harmonisation of the various damage classification
approaches for the regional application. In the following,
recommendations for a standardised methodology for
applying satellite data for forest damage assessment are
presented.
7. 1 Remote Sensing Data
Due to the large area to be covered and the spectral
complexity of forest damage classification, Landsat TM
data were recommended by the TWG in this particular
experiment. The spectral resolution of these data has
been proven to be useful for forest damage classification
in various studies and the main damage areas can be
covered at reasonable costs. Since the sensor shows a
spatial resolution of 30 * 30 metres, a mapping scale of
1:50,000 can be provided accurately. More detailed
inventories on local level by means of infrared aerial
photographs or feasibility studies with other satellite
sensor data can be performed within the national projects
and are not an element of the LAOE.
Infrared aerial photographs are explicitly recommended
for supporting the determination of suitable training areas
for the digital classification and for verification. The
application of the AFL interpretation key ensures the
standardised evaluation of damage levels by visual
photo-interpretation. For financial reasons, infrared aerial
photographs should only be taken for relatively small
forest areas, giving a representative survey of different
degrees of damages, types of forest management and
topographic conditions.
7.2 Problems of Classification and Ancillary
Information
®
Forest damage is only one of a number of factors
influencing the spectral response of stands. If the
separation of damage classes is not possible by relying
on spectral signatures only, ancillary information, if
available, is needed for obtaining a more precise
classification result. Particularly, in classifying different
forest-damage levels whose spectral characteristics differ
only marginally, the overlapping of object-typical (damage
typical) spectral information by other parameters might
be so severe that a satisfactory classification into many
different damage classes without ancillary data is very
difficult. Because of the diverging data in the different
countries, it can not be assumed that the availability and
quality of ancillary information is homogeneous among
the participating countries. This fact was not seen as a
major problem by the TWG as long as different kinds of
ancillary data do not inevitably lead to results which are
no longer comparable.
The TWG came to the conclusion that the incorporation
of ancillary information is quite helpful in improving the
classification accuracy but only those ancillary data
should be used, that are already available in digital
format. Digitising large amounts of data, just for the
purpose of optimising satellite classification, is too time-
consuming and thus not feasible for small-scale
classifications.
7.2.1 Seperation of Forest and Non-Forest
Basically, a separation of closed forest and non-forest
using TM data is possible. However, using the signatures
736
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996
CNY NY If IM "rN $e 6 ow
NY YY 7^ t = 5 X —
ml
