inect the deepest
jugh information
| contours. Their
odels (DEMs) or
program called
orithm of Profile
h PPA from two
branch reduction
ire automatically
with the original
per for automatic
yme cartographic
tracting drainage
determine water
Seemuller, 1988;
sity of drainage
sibility of terrain
etric structure of
ck type of a land
res are essential
tion studies and
oe lines) can be
al terrain model
us elevations by
y important for
ng the ridge side
ie high avalanche
on, some of the
ler for different
es usually accept
two-dimensional
. They generally
itify the features.
rocesses have not
wing still is done
e main reason for
lata trends within
most automatic
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
[n this study, some of the terrain forms depending on elevation
data for deriving topographical map information have been
investigated. Then, an automatic line extraction algorithm
coded firstly in FORTRAN by Chang et al. (1988) with the
name PPA program have used for extracting the terrain skeleton
lines by expending the effect of the branch reduction step. The
algorithm is prepared to demonstrate the feasibility of digital
elevation models and computer programming techniques in
automation for extracting ridge and valley line systems of
topographical relief. Finally, some of automatic valley line
extraction results have compared with a topographical map
which is drawn by a human operator from photogrammetric
stereomodel.
2. FORMS OF DTM AND DEM DATA SETS
The terrain surface model is most commonly described either as
a DEM, or as a digital terrain model (DTM) in literature. The
form of DEM is defined as a regular two dimensional array of
heights sampled above some datum that describes a surface.
The other description of DEM is regular gridded matrix
representation of the continuous variation of relief over space.
On the other hand, the form of DTM contains elevation
information with the addition of some explicit coding of the
surface characteristics such as breaks in slope, drainage divides
etc. Examples of DTMs include the triangulated irregular
network (TIN), digital contours with form lines, and the richline
model of Douglas that uses ridge, valley and form lines to
define an elevation model (Wood, 1996).
The points of DTM or DEM data set can systematically be
collected using four different methods according to geometric
disposition of the heights points on terrain surface (Figure 1).
Type I Type lI
ee E t TT
Type III Type IV
Figure 1. Height points data sets (Yoeli, 1984)
Type I: Regular distances height points are ordered with
squared grid as a DEM,
Type II: Irregular distances height points are distributed along
the equidistant parallel profiles,
Type III: Horizontal arrays of equal height points,
Type IV: Randomly distributed height points.
The height points in Type I can be collected manually or
automatically from photogrammetric stereo models. The height
points in Type II can be collected manually where the slope is
changed along the equidistant parallel profiles from
photogrammetric stereo models. The height points in Type III
can be collected manually along the equal horizontal heights
from photogrammetric stereo models or along the contours by
digitizing from hardcopy topographical maps. The height
points in Type IV can be collected where the slope is changed
from photogrammetric stereo models or land surveying.
Adequate logics for the analytical search of skeleton lines of the
relief can, in principle, be formulated for all four types of
D'TMs and there is no need to transform the Types II, III, IV, if
initially so given, into a DTM of Type I by interpolating a
secondary DTM superimposed on the original points (Yoeli,
1984). For example, Aumann et al. (1991) and Tang (1992) are
derived skeleton lines from digitized contours to generate high
quality DTMs. Their data set is like Type III which is acquired
from the contour maps. On the other hand, the simple matrix
form of elevation values as Type I is the most efficient data to
be able to processing with programming language. Due to its
easy integration within a geographic information system (GIS)
environment, the use of gridded matrix representation of the
continuous variation of relief, which means DEM according to
Wood (1996), has become widespread (Figure 1, Type 1). For
the availability of gridded matrix representation, most scientists
have used DEM form in order to extract the terrain skeleton
lines in their studies (Yoeli, 1984; Wood, 1996; Chang, et al.,
1998).
3. AUTOMATIC EXTRACTION OF TERRAIN
SKELETON LINES
FROM DEMs USING ‘RIDGEVALLEYAXISPICKER’
PROGRAM
The basis of the PPA algorithm is written by Chang in Visual
Basic in three steps; target recognition, polygon breaking and
branch reduction. After the three steps, the program produce
smoothed or non-smoothed skeleton line segment coordinates in
a text file. Figure 2 shows the steps of the program.
The main difference of the *RidgeValleyAxisPicker' program
from the original PPA program has been appeared on profile
recognition in target recognition step. There is no profile in
‘RidgeValleyAxisPicker’ program and all height points are
target. Another difference is in branch reduction step. Because
half of the profile length is unknown, there is no rule how many
segments have been deleted for any branches in the step.
In the study, the effect of branch reduction step of the
*RidgeValleyAxisPicker' program is modified by deleting
branches iteratively to obtain more suitable skeleton lines to the
land form.
3.1 Target Recognition
Target recognition procedure is carried out in 'ConnectAll and
‘SortSegment’ subroutines in the ‘RidgeValleyAxisPicker’
program. Both of the subroutines start automatically. Firstly,
‘ConnectAll’ ties all of the DEM points which have elevation
values, that means points are upper than sea level, with line
segments. Secondly, ‘SortSegment’ sorts all of the segments
according to their weights which are calculated by total
elevation value of two points of a segment. Briefly, all of the
DEM points, upper than sea level, are selected as target and tied
with line segments, before the segments are sorted in target
recognition step (Figure 3).
