The trees of the same size, just in a different relation to each
other, can be delineated in very different way. As it can be
noticed, depending on delineation philosophy, crown size can
significantly change. Trees in stands growing usually very close
to each other, they branches penetrate each other and
interindividual competition cause different crow deformation.
So even for single species monoculture stands, accuracy of tree
detection and crown delineation vary regarding to stocking size
and species.
Most commonly Crown Height Model (CHM) is used for crown
delineation. It is always some kind of approximation of real tree
crown surface, interpolated in different way depended of used
algorithm. As we can see relations tree to tree is very different
(Fig. 2) and depends especially on density.
va
: Sigleieos kt 0
; … Group oftrees .” A soins 5
Tree height (cm)
SBEBHSSBERRRLREREEZEER
T
—— |
=
| j
S n n o. S aS NEUSS SUO S SS SS
Distance [m]
Figure 2. Forest stand cross-section in CHM
te Li et de
28288
Ir dl i
|
All methods of crown delineation have some disadvantages
because just single growing trees crowns can be described
accurately. If trees are merged together it is impossible to detect
right shape of crown.
Additionally, paradigm of forest silviculture is changing to
sustainable forest management. Forests are changing in many
areas from pine or spruce monocultures into stands with more
complex species composition, age and height structure. This
will cause additional problems.
1.3 Motivation and aims
Detection and accurate determination of tree segments influence
volume or biomass estimation (Popescu et al., 2003; Kwak et
al, 2010). Figure 1 presents method used in the study to
delineate crown segments. Motivation behind this philosophy
was that height threshold cause smaller segment variations, than
other methods. This follows from the fact that we observe a
higher part of the tree, which often is not distorted by
neighborhood with other trees, while their shape is proportional
to the size of the tree (Fig. 2).
Aim of the study was to find the best selected height threshold
and sampling rule for standing volume calculations in protected
area.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B8, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
2. MATERIALS AND METHODS
2.1 Study area
Area of interest is situated in The Stolowe Mountains National
Park (south-western Poland, near the border with Czech
Republic). Created in 1993, the Park covers an area of
63.39 km?, of which forests accounts for 57.79 km. Forested
regions are covered mainly by Norway spruce (Picea abies Ly.
Spruce stands within the park are in the poor health condition.
Spruce monocultures are very susceptible to the adverse effects
of many abiotic and biotic factors. Deciduous forests originating
from natural regeneration are preserved only in small fragments
in an inaccessible area.
2.2 Field data
The terrestrial measurements were performed during July 2009
by the team of TAXUS SI. Permanent circular sample plots
with a radius of 12.62 m were distributed in a systematic grid
arrangement each 400 meters on the entire area of the Stolowe
Mountains NP. Coordinates of each plot's centre were
measured using DGPS method. All living and dead trees with
dbh28 cm inside the sample plot were measured.
2.3 Airborne Laser Scanner Data
Data acquisition was performed in August 2008 using Altman's
Optech 3100 system installed on the board of Cessna 206 plane.
The main system components are: scanner with LIDAR sensor
head, data storage devices, laptop and remote display. The
position and orientation of the system during the flight was
carried out with a set of Applanix POS-AV 510, which consists
of a system that combines a standard geodetic dual-frequency
GPS receiver and high-end inertial recorder deflection and
rotation angles (Inertial Measurement Unit - IMU).
Sensor 3100
ight hei 700 m
Data i to 5 echoes
| +/- 18°
Scan 100 kHz
Table 1. ALS data characteristics
Digital surface models were generated using TreesVis Software
(FELIS, Germany).
2.4 Volume calculations
The volume of each sampled tree was calculated using formulae
currently applied in the planning of forest management in
Poland (Bruchwald et al., 2000). Stem volume was assumed to
be the volume of ‘merchantable timber’ and the lower dbh
threshold was taken to be 8.0 cm. Plot volume — divided into
the ‘volume of living trees’ and the *volume of dead trees' -
was derived as a sum of the individual trees. Results obtained
from the ground measurements were used as reference data.
2.5 Single tree segmentation
Segmentation was conducted in Halcon MVTec software, in the
following way:
a) 0.5 m resolution Canopy Height Model (CHM) was loaded;
b) Median filter was applied (circle, 1 pixel radius);
c) Primary segmentation was carried out;
d) Se
e) Le
h) W
For fi
minin
calcul
Basec
whick
samp
numb
heigh
of crc
with
as the
with ;
ortho
2.6 |
In th
heigl
wher
In th
carri:
samp
(‘cen
Crow
each
relati
Carri
For ¢
corre
