1203
INTERGRATION OF HIGH RESOLUTION AERIAL IMAGES AND AIRBORNE LIDAR
DATA FOR FOREST DELINEATION
Zuyuan Wang, Ruedi Boesch, Christian Ginzler
Dept, of Land Resource Assessment, Swiss Federal Research Institute WSL
zuyuan.wang, ruedi.boesch, Christian - ginzler@wsl.ch
WG VII/6
KEY WORDS: Aerial images, Airborne LIDAR, Image Segmentation, Texture, Forest Delineation, Remote Sensing
ABSTRACT:
Forest boundary is one of the important parameters in Swiss National Forest Inventory (NFI). It is mainly delineated by aerial photo
interpretation according to the defined “Forest/Non-forest decision” rules at each sample plot. However, it is not suitable for local
assessment of forest stands with such kind of coarse grid sampling design. This paper presents an approach intending to delineate
forest boundary automatically by integrating use of aerial images and airborne LIDAR. Empirical tests show that the proposed
method offers an automatic process of forest boundary detection for various aerial images in a promising way. However, it’s a
challenge task to describe NFI forest /non-forest definition with automatic computer-based method and current study provides an
encouraging basic for further development and testing.
1. INTRODUCTION
Forest is an indispensable foundation of human life. It fulfils
multiple functions such as providing wood and foodstuff,
protecting soil from erosion and stabilising the climate on a
regional and global level etc. A third of Switzerland is covered
in forest. While the plateau has only relatively little forest, by
contrast the south side of the Alps is particularly rich in forest.
In order to record in detail the current state and change within
the Swiss forest in a representative and reproducible manner,
the Swiss NFI has been carried out for three times by the
Federal Council: between 1983 and 1985 (NFI 1), 1993 and 1995
(NFI2) and from 2004 to 2007 (NFI3) (Brassel,2001, Brassel
and Brandli,1999) . A forest inventory is the procedure used to
obtain information on the quantity and quality of the forest
resource and many of the characteristics of the land on which
the trees are growing(Husch, et al.,1982) . The Swiss NFI
covers all essential data on land use, giving the surface area, the
growth and the condition of the forest. Apart from the above
important forest management indicators, the NFI is designed to
be a multi-purpose inventory which entails a high demand on
the methods implementation, flexibility in respect to the
contents, inventory perimeters as well as data analysis.
The employment of aerial photograph is involved in Swiss NFI
so that the cost of the ground survey could be reduced. An
important application of aerial photograph lies on the
classification of plot samples in forest and non-forest areas.
Each aerial photo sample plot was classified according to the
so-called “Forest/Non forest decision”. The Forest Boundary
Line (FBL) determines the border line between normal forest or
shrub forest and non forest. It allows the evaluation of the forest
width and forest inter-space. These values are needed, in order
to reach a forest/non-forest decision. The interpreted boundaries
at each sample plot belong to a regular 500m grid with 25 raster
pointes. However, local assessment of the forest area and forest
parameters is not suitable due to such a coarse grid design.
Furthermore, subjective interpretation results and production-
oriented uncertainty can’t be avoided despite the careful
training.
During the last decade, NFI research has focused on utilizing
remote sensing data in forest inventory. Remote sensing data
can make significant contribution to regional and global forest
cover assessment. It is cost-efficient information for forest
inventory and monitoring purposes(Pekkarinen,2002a). Satellite
images have been widely applied in different forest inventory
and monitoring tasks, e.g. estimation of forest characteristics
(Lu, et al.,2004, McRoberts, et al.,2007, Tuominen and
Haakana,2005), change detection (Desclee, et al.,2006, Yen, et
al.,2005), etc. Haara et al.,(2002) propose a tree species
classification method using semi-automatic delineation of trees
on CIR (color infrared) aerial images. LIDAR (Light Detection
and Ranging data) is an optical remote sensing technology that
measures properties of scattered light to find range and/or other
information of a distant target. This technology has applications
in archaeology, geography, geology, geomorphology,
seismology, remote sensing and atmospheric physics. A number
of studies reveal the successfully use of LIDAR-based
techniques to estimate tree and stand attributes such as height,
crown diameter, basal area and stem volume(Donoghue, et
al.,2007, Morsdorf,2004, Naesset,1997).
There are about 7000 aerial images as well as LIDAR data
covering the whole Switzerland in NFI3 which can be used for
extracting forest features. These data allows the implementation
of modem image processing methods and NFI may be partially
automatic analyzed. In addition, the advantage of achieving
reproducible and consistent result will be offered by using the
available remote sensing data. Improvements are expected in
areas where it is possible to replace or supplement decisions
based on expert opinions, subjective interpretations, or
estimation with measurable quantities.
This paper suggests an integration algorithm of high resolution
aerial images and LIDAR data for the improved automatic
forest boundary delineation. Image segmentation method is
applied for obtaining homogeneous patterns while LIDAR data