469 
Symposium on Remote Sensing for Resources Development and Environmental Management / Enschede / August 1986 
Experiences in application of multispectral scanner-data 
for forest damage inventory 
A.Kadro & S.Kuntz 
Department of Photointerpretation and Remote Sensing, University of Freiburg, FR Germany 
ABSTRACT: For testing the potential use of multispectral scanner data for the inventory of forest damages in 
large areas five test sites in South-west Germany were sensed at three flight altitudes with an 11-channel 
scanner. At the same time, ground truth information in these test sites were obtained and the actual state 
of the forest stands was documented with color infrared (CIR) aerial photographs. The test sites differ in 
morphology, forest types and degree of the actual forest damage.The acquired data were evaluated with a com 
puter aided supervised classification using the maximum-1ikehood method. For verification of the classification 
results for both single trees and stands, the terrestrial ground truth and the CIR-photographs were used. 
This paper presents the classification results and discusses the problems of a computer aided forest damage 
inventory/. 
1 INTRODUCTION 
Since the late 1970's a regional decline affecting 
many tree species has occured in Europe. The urgent 
need for detailed information about the actual si 
tuation of the forests in Germany are of vital in 
terest for both government and forest departments. 
So aerial and ground survey methods have been used 
to get this information. 
But for large areas these methods are time-consuming 
and expensive. So in 1983 a project started at the 
Department of Photointerpretation and Remote Sen 
sing, University of Freiburg, to evaluate multispec 
tral scanner data for forest damage inventory. For 
this purpose data were collected in July 1984 and 
August 1985 on 5 test areas in South-west Germany 
with a Bendix-M2S-Scanner flown by the DFVLR Ober 
pfaffenhofen. This scanner was modified (Table 1 ) 
by the DFVLR to simulate the Thematic Mapper in Land- 
sat 5 (Table 2). The data were collected at altitudes 
of 300 m, 1000 m and 3000 m. A Landsat 5 image from 
nearly the same time, recording the same areas was 
also evaluated. 
Table 1. Spectral channels and wavelengths of the 
modified Bendix -Scanner 
median 
range 
median 
range 
channel 
Ann* 
channel A tun 
3 
515 
50 
9 
720 
40 
4 
560 
40 
10 
1015 
90 
5 
600 
40 
5TM 
1650 
200 
6 
640 
40 
7TM 
2210 
270 
7 
680 
40 
11 
11000 
6000 
8 
720 
40 
Table 
2. Spectral channels and 
wavelengths of Land- 
sat 5 
(TM) 
median 
range 
median 
range 
channel Ann» 
Alum 
channel 
A H»m 
Alnm 
1 
485 
70 
5 
1650 
200 
2 
560 
80 
6 
11450 
2100 
3 
660 
60 
7 
2215 
270 
4 
830 
140 
the different altitudes have the following ground 
resolution (pixel size): 
at 300 m altitude 
at 1000 m " 
at 3000 m " 
at 705 km " 
0,75 X 0,75 m 
2.5 X 2,5 m 
7.5 X 7,5 m 
30 X 30 m 
(aircraft MSS) 
If 
(Landsat 5) 
Fran the 300 m altitude pixels numbering up to 100 
represent one single tree crown. Fran 1000 m and 
3000 m one can evaluate only groups of trees or 
stands and from Landsat images only large stands 
can be evaluated. 
The test site iron which results will be presented 
is mountainous and contains mostly coniferous trees 
(spruce mixed with fir) and some smaller stands of 
deciduous trees (mostly beech). The main interest in 
this project was focused on coniferous trees because 
they are of major interest in german forestry al 
though in a continuing project deciduos species also 
will be investigated. 
The computer aided classification of different damage 
classes is based on the differences in reflection of 
healthy and damaged vegetation in the spectral re 
gion of the visible, near infrared and middle infra 
red part of the electromagnetic spectrum. These spec 
tral differences are assumed for a computer classi 
fication to operate, so the first step was the eva 
luation of the spectral signatures of differently 
damaged tree species. The results of this evaluation 
are presented in a special paper at this symposium 
(Kadro, 1986). For the supervised classification 
and presentation of results special software was de 
veloped at the department. The classification algo 
rithm is a combined box and maximum-1 ikelyhood 
classifier which can also analyse additional infor 
mation given by the user. For example: terrain model, 
masks, modification of the covariance matrix, a pri 
ori probabilities and permanent or temporary condi 
tions for including or excluding pixels during the 
classification process. 
2 RESULTS OF THE COMPUTER AIDED CLASSIFICATION! 
2.1 Frcm 300 m altitude 
The test sites differ in morphology, forest types and 
degree of the actual forest damage to simulate all 
possible inventory problems. The data collected frcm 
For checking the classification from 300 m altitude 
a crown map drawn with a Bausch and Lomb Zocm-Trans- 
ferscope was digitized and used as an overlay (photo 
1).