You are using an outdated browser that does not fully support the intranda viewer.
As a result, some pages may not be displayed correctly.

We recommend you use one of the following browsers:

Full text

Remote sensing for resources development and environmental management
Damen, M. C. J.

Symposium on Remote Sensing for Resources Development and Environmental Management / Enschede / August 1986
Digital classification of forested areas using simulated TM-
and SPOT- and Landsat 5/TM-data
Dept. Luftbildmessung und Fernerkundung, Institute for Forest Economy and Inventory, University of Freiburg, FR Germany
DFVLR (Deutsche Forschungs- und Versuchsanstalt für Luft- und Raumfahrt), FR Germany
ABSTRACT: In the following piece of research both SPOT- and TM-images, as well as true Landsat 5/TM data have
been digitally classified. The results show information about the possibilities of recognizing different types
and age classes of trees, together with a high resolution of the classified forest units. Using both TM- and
SPOT-data it is possible to differentiate between at least three age classes in forest stands. As well as
distinguishing betv^en coniferous and deciduos trees, it is possible to recognize certain tree types in pure
crop stands within these classes, depending on the time of year. The correctly classified forest unit is
influenced by the form and size of the stands, but generally stands larger than an hectare can be recognized.
Ihe classification of the Landsat 5/TM scenes was improved by taking into account topographical information
and by using rnultitemporal data.
The results of the TM-simulation carried out by Kirch-
hof, W. , Mauser, W. and Stibig, H.-J. , were pub
lished in the research report FB-85-49 of the DFVLR
(Deutsche Forschungs- und Versuchsanstalt fiir Luft-
und Raumfahrt).
The digital classification of Landsat/MSS-images has
already been carried out in countries with extensive
forest areas, producing good results. Offering a high
level of efficiency, their use has been shown for ge
neral classifications, for example seperating deci
duos from coniferous forest for the purpose of stra
tification. for detailed inventory methods. However
owing to the limitations of the geometrical resolu
tion, Landsat/MSS data have proved of little use for
forestry classification in Centra], Europe.
The size of the planning units in the intensively
managed forests is usually between 3 and 5 hectares
and the treatment units are often smaller. Whereas an
area of 2,56 hectares is necessary for one pure
Landsat/MSS pixel. New sattelite images from SPOT-
and Landsat 5/TM with a high geometrical resolution
open up possibilities for the use of remote sensing
for forestry purpose in Central Europe.
The strip overflown for the SPOT- and TM-simulation
is situated west of Freiburg. Apart from agricultural
and built up areas, the test area contains typical
mixed deciduos forest of the Rhine valley. The main
tree types are oak, ash and maple. There are also
some single pure crop stands of spruce, douglas firs
and red oak.
A test area to the north-west of Freiburg in the
Kaiserstuhl was chosen for the multitemporal evalua
tion of the Landsat 5/TM data.
This area consisted of two types of forest:
1. "Auwald": deciduos lowland forests near the
Rhine. The main tree types are oak, white beech,
poplar and maple as well as pure crop stands of
douglas firs and pine.
2. Colline to submontane mixed deciduos forest
consisting mainly of beech, white beech and oak,
varying in height from 200 to 550 metres.
The simulation of the SPOT and TM data was carried
out on behalf of the research centre of the European
community (IRC) in ISPRA. The SPOT-simulation took
place on the 26.5.1982 and was performed using a
10 band Daedelus scanner from a height of 7000 metres
by the NGI (National Geographical Institute). The
radiometrical simulation of the SPOT-simulation bands
(S.S.) from the Daedelus bands was carried out by
the CNES (Centre National d'Etude Spatial) (Tab. 1).
The TM-simulation was flown on the 21.7.83 fron a
height of 4000 metres by the DFVLR, using two Bendix-
M2S-Scanners, one of which was modified to work in
the middle infrared wavelengths. For the evaluation
of the data, the scanner bands (T.M.S.) which most
closely approach the TM bands were used (Tab. 1).
The selected Landsat 5/TM-images are the scenes
195/27 taken on the 18.4.84 and the 7.7.84.
In order to include the different tree types and age
classes, the test areas were selected by using ground
truth inventories, the interpretation of infrared
false cölour composites and the .information from fo
rest management plans.
The geometrical rectification of the scenes took
place with pass points on Gauß-Krüger coordinates
using an exponential transformation polynom.
The signature analysis of the test areas was per
formed by comparing histogramms, the reflection
curves of the mean values and the presentation of
the object classes in two dimensional feature space.
The quality of the training areas for the simulation
data used for the maximum likehood method was esti
mated using a confusing-matrix.
The sub-scenes were finally classified according to
the maximum likehood method. For the TM-scene of
April a more simple classification by the defini
tion of thresholds was sufficient for a forestry
stratification, owing to the more distinct reflec
tion differences. For that purpose the maxima and
minima values for all the bands were cited.
Additional information in the form of a digital
terrain model was used for a further stratification
according to height. To test the accuracy of the
classification the results were compared with forestry
planning maps of selected areas.
The image processing systems FIPS (Freiburg's Image
Processing System) of the dept. Luftbildmessung und
Fernerkundung and DIBIAS of the DFZLR were used for
the digital image analysis.