FOREST PARAMETER DERIVIATION FROM DTM/DSM
GENERATED FROM LIDAR AND DIGITAL MODULAR CAMERA (DMC)
M. Heurich®* S. Schadeck®, H. Weinacker® ,P. Krzystek®
“Bavarian Forest National Park, Department of Research - marco.heurich@fonpv-bay.bayern.de
"Munich University of Applied Sciences, Department of Geoinformatics - krzystek@geo.fhm.edu
‘Freiburg University, Department of Remote Sensing and Landscape Information Systems —
holger.weinacker@felis.uni-freiburg.de
KEY WORDS: Forestry, DEM/DTM, Laser Scanner, Digital Imagery, Individual Tree Detection
ABSTRACT:
Ecological analysis and forest planning need sophisticated information about the structures of forests. In recent years new sensor
systems like laserscanner and digital airborne cameras emerged on the marked which fulfil the high requirements for forest
applications. Also, new methods for automated delineation and feature extraction of individual trees have been developed. In this
study images of the brand-new Digital Modular Camera were used to automatically generate a DSM by image matching techniques.
The resulting density of matched 3D points was 18 pts/m?. After the robust filtering the cleaned point cloud showed a mean density
of 11 pts/m°. The photogrammetric DSM was subsequently compared to a manual stereoscopic measurements and ground surveys.
Tree height measurements from stereoscopic measurements showed a better accuracy than heights derived from laserscanning DSM
and the photogrammetric DSM. The applied method of DSM generation using a feature-based matching approach could successfully
reconstruct deciduous canopy surfaces with almost the same accuracy as the laser scanner did. However. the method failed especially
when single coniferous trees were present in the plot by cutting the tree tops and underestimating the lower areas between the trees.
Several ways to improve the matching strategy of the present algorithm for canopy reconstruction are discussed. The potential of the
photogrammetric and laserscanner DSM's for automated tree detection could be clearly shown by applying a single tree delineation
algorithm which is based on a watershed approach.
1. INTRODUCTION
For decades humans have been manually interpreting acrial
photographs of forests: counting stems, classifying species and
stands. Also tree heights were estimated using stereoscopic
acrial photographs. Because of its importance to forest
inventory early studies examined the crown surface with
photogrammetric methods (Hildebrand et al. 1974).
Since the conventional methods are pretty time intensive only
stand delineation's are performed with the help of aerial
photographs in central Europe. Forest assessment on a single
tree level is done by field based sampling of trees. For a wider
use of remote sensing in forest inventory new cost effective
methods have to be developed. Automated individual tree
crown mapping (delineation and feature extraction) is a main
focus of forest inventory research. Accurately mapped and
classified crowns are useful to attain several management goals
(e.g. volume calculations or habitat modelling). The following
two preconditions appear mandatory.
Firstly, sensors with high spatial and spectral resolution are
needed which render possible the characterisation of forest
structures like canopy surfaces or the detection of single trees.
Some of the new sensors which emerged in the last years fulfil
these preconditions to a large extent. Laserscanning is a
technology with great potential for applications in forestry.
Since laser beams partly penetrate the forest surface Digital
Surface Models (DSM) and Digital Terrain Models (DTM) can
be generated by suitable filter techniques. Depending on the
measuring density of the laserscanner it is possible to achieve a
spatial resolution of less than half a meter. Also, the new digital
airborne cameras like the DMC (Z/I-Imaging), ADC (Leica-
Geosystems) or the HRSC-A (DLR) provide high resolution
84
panchromatic and multispectral imagery which can be used to
automatically extract 3D objects and their characteristics.
GSD’s of 10 cm or less are possible mainly depending on the
image scale. For instance, Hese at al. (2000) report about the
generation of DSM in forest areas from HRSC-A imagery with
a post spacing of Im at a GSD of 12cm. Although spaceborne
optical sensors are reaching meanwhile a half meter resolution
and airborne radar sensors (InSAR) are getting under the 1m
level, these sensors cannot contribute so far to a detailed
analysis of forest structures on a tree level due to their limited
DSM resolution they can provide.
Secondly, the increase in resolution implies a paradigm change
in techniques of digital image analysis, since algorithms
developed for coarse resolutions are not directly applicable to
high resolution imagery. Therefore, new methodologies for
automatic detection of trees have been developed in the recent
years. There exists several approaches like segmentation
(Gougeon 1999), finding local maxima (Dralle 1997), edge
detection techniques (Brandberg and Walter 1999) and
watershed approaches (Person et al. 2003).
In this study images of the brand-new Digital Modular-Camera
(DMC) were used to automatically generate a DSM by image
matching techniques. The photogrammetric DSM was
subsequently compared to a laserscanning DSM, manual
stereoscopic measurements and ground surveys. The potential
of the photogrammetric DSM's for automated tree detection
could be clearly shown by delineation of single trees using a
method developed by the University of Freiburg. This research
is embedded in the research project “Evaluation of remote
sensing based methods for the identification of forest
structures”, which has been established to investigate different
airborne sensors like laserscanner, InSAR and digital cameras in
Internatic
combinatioi
automatic d
2.1 Stud:
The researc
Park which
border to t
Within the
alpine spru
partly Mou
mountain. f
European b
pseudoplata
1100 m Fin
ash and bir
valley bottc
ponds.
2.2 Field
Three samp
100m were
mixed fores
September
breast heigl
determined
measuremei
following t
stem positi
DGPS. The
and was «
evaluated ai
SILVA 2?
(Pretzsch et
determined
(50) is a ty
environmen
Feistenháng
European b:
890m above
slope is 15.
typical for t
plot Spe
cies
22 Piece
Fagu
Acer/Tilia
30 Pice
60 Fagu
Ace