A landform classification method with GIS for landscape

Sustainable Development and Planning II, Vol. 2
825
A landform classification method with G.I.S.
for landscape visual analysis purposes
A. Tsouchlaraki
Hellenic Open University, Greece
Abstract
The relief forms are essentially innumerous and their classification has been a
multidisciplinary field of study. This paper develops a landform classification
method, in order to meet visual analysis needs. The main feature of this
classification is that it tries to predict perspective observation conditions and
constitutes a modification of Hammond’s method. Considering that we have a
digital terrain model in the method proposed, the three parameters of
Hammond’s method are adopted (flat slope percentage, hypsometric difference,
flat slope percentage in the upper or lower half of the hypsometric difference
range) while one more parameter is added regarding the observation post. The
four parameters are applied to movable windows of specific dimensions and of
specific movement increment on the digital terrain model. GIS tools are used for
this work.
The study area of this paper is about the Lefka Ori mountain range of the
island of Crete. From the results of the classification 33 relief categories are
derived. In order to examine the degree to which these classification results meet
landscape visual analysis purposes, certain positions were sampled, from which
perspective digital imaging representations of the relief were created, with the
use of a special algorithm. The results show that the proposed classification is
considered successful, since it fulfils to a great extent the description of the
visual impression of an observer for a specified relief form. A further
examination may lead to the determination of all morphological combinations
that can probably appear.
Keywords: landform classification, quantitative parameters selection, landscape
visual analysis, geographical information systems.
WIT Transactions on Ecology and the Environment, Vol 84, © 2005 WIT Press
www.witpress.com, ISSN 1743-3541 (on-line)
826 Sustainable Development and Planning II, Vol. 2
1
Introduction
Landscape visual analysis attempted to reveal the factors that regulate the
complexity and the differentiation of physical space, and to present the
difference between the actual and the visual landscape. It differs from the usual
geographical analysis, which is also an analysis under specialised knowledge, in
that it does not deal with physical environment only as a “reality”, but also as a
space of behaviour and perception. People when moving within a landscape do
not analyse a static map or an aerial photo. Issues of perspective, relative
position, movement, orientation, etc., help at interpreting perception; in this
respect, landscape analysis has been trying to play this role [1].
There are cases when it is desirable to isolate and examine separately the
relief from other elements that compose the visual environment. These cases
include technical works that cause big and permanent changes in the relief, as
well as works whose arrangement depend on the soil’s morphology.
The forms of the relief are practically indefinite in number. Landform
classification is an issue that has preoccupied many scientists, and in particular
surveying engineers [2, 3, 4]. The methods developed differ as to the way they
approach the problem and the aim they serve. In order to meet the needs of visual
analysis, this paper develops a landform classification method, based on existing
experience, but modified to meet the needs of perspective and 3D perception.
The method is applied on Lefka Ori mountain range of the island of Crete,
due to the variety of its relief forms, since it combines an island and an
appropriate inland environment.
2
Developing a landform classification method for visual
analysis purposes
The classification method developed is a modified form of Hammond’s
classification, which was selected as the basis; firstly because it is one of the
most acknowledged relief classification methods [5] and secondly, because it
uses few variables in the quantitative determination of forms, something that is
essential in order to have a limited and logical number of categories. Considering
that a digital terrain model is available, the method proposed adopts the three
quantitative parameters of Hammond’s classification, and one more. These four
parameters are applied to moving windows of specific dimensions. Following
this, a detailed description of the elements of the proposed classification method
is made.
2.1 Review of Hammond’s classification method
According to Hammond’s classification, in order to classify large areas, the
study area is sub-divided into large square parts, 6x6 square miles, namely
10x10 km2 approximately [5]. This size is neither big to include a large variety
of forms, nor small to divide individual slopes. The quantitative parameters used
WIT Transactions on Ecology and the Environment, Vol 84, © 2005 WIT Press
www.witpress.com, ISSN 1743-3541 (on-line)
Sustainable Development and Planning II, Vol. 2
827
are based for their determination on the main stages of geomorphic process.
These parameters are:
1. Flat slopes percentage. Slopes up to 8% are characterised as flat.
2. The maximum hypsometric difference observed within the square part.
3. The percentage of flat slopes observed at the upper or lower half of the
maximum hypsometric difference range.
Each of the variables mentioned above is divided in specific value intervals
(Table 1), while 96 possible combinations may theoretically derive from the
intervals of all variables. However, the application of Hammond’s method to the
Appalachia Mountains showed that only half of the combinations might probably
appear, while most of them at a small percentage [5]. So he suggested a final
classification in five categories: i) purely plane areas; ii) areas that combine
plane and hilly parts (Β3b); iii) plateaux (B3c, B4cd); iv) mountains with small
curvatures on their picks (C3bc, C4acd, C5cd); and v) mountains with great
curvature (D3, D4, D5).
Table 1:
Hammond’s classification.
Parameters
Flat slopes percentage
Maximum hypsometric difference
Flat slope percentage at the upper or lower half
of the hypsometric difference range (this
parameter is not examined when flat slopes
cover less than 20% or when dh <30)
Code
A
B
C
D
1
2
3
4
5
6
a
b
c
d
Value intervals
> 80%
50-80%
20-50%
<20%
0-30m
30-90m
90-150m
150-300m
300-900m
>900m
>75% at the upper half
50-75% at the upper half
50-75% at the lower half
>75% at the lower half
Cole et al examined the minimum possible size of the square part, where
Hammond’s method may be applied, without altering the classification. By this
examination they proved that in parts smaller than 3x3 square miles, namely 5x5
km2 approximately, categories alter in relation to the initial ones and, moreover,
they define this size as the threshold for applying the method [5].
2.2 Quantitative parameters in the classification method proposed
The classification method proposed adopts the three quantitative parameters of
prementioned Hammond’s method, while one more variable is added for the
observation post. The observation post is actually related to the relative elevation
post between the observer and the objects he observes and it constitutes a
significant element in landscape analysis, because it affects the viewshed offered
by the position. Three types of positions are determined [6]: 1) higher, 2) equal,
and 3) lower. At higher position, the observer is higher than the objects he
WIT Transactions on Ecology and the Environment, Vol 84, © 2005 WIT Press
www.witpress.com, ISSN 1743-3541 (on-line)
828 Sustainable Development and Planning II, Vol. 2
observes, while this position offers a larger viewshed and gives him the
opportunity to better observe and perceive shapes and forms. At equal position,
the observer stands at about the same elevation with the observation objects,
while at the lower position he stands at a lower level. As we are interested in the
observation post with regard to the whole view of the relief and not for a certain
sub-area of the relief, we used the criterion of 1/3 of the maximum hypsometric
difference observed in order to determine the position. That is, if ‘h0’ is the
elevation of the observation post, ‘hmin’ the minimum elevation observed in the
window and ‘dh’ the maximum hypsometric difference, then:
h0 - hmin <
1
dh => relatively lower position
3
1
2
dh < h0 - hmin < dh => relatively middle position
3
3
h0 - hmin >
2
dh => relatively higher position
3
This logic is based on three characteristic parts of a hill: a) the foot; b) the
hillside; and c) the peak. Therefore, the position is characterised lower at the first
third of the hypsometric difference, because it is expected to observe mainly the
foot of the elements of the relief and there is in general a limited visibility. At the
second third, the position is characterised as middle, because visibility will be
greater than the one at the lower position, but at the same time narrower than the
one offered by the higher position. Finally, at the last third, where peaks prevail
and visibility has the maximum possible range, the position is characterised as
higher.
2.3 Size and movement increment of the moving window
The previous four quantitative parameters are applied to square parts of 5x5 km2.
This surface area meets two criteria: a) as mentioned above, Cole et al found that
it is the threshold to which Hammond’s method can be applied; and b) the 5 km
distance practically constitutes the middle viewing zone for landscape analysts.
Most landscape analysis systems do not examine distance as a constant
variable, but they make use of a general distinction by dividing it into three main
zones, i.e. close, middle and long distance. Various approaches have been
formulated for determining these zones [6,7,8]. The close viewing zone is
characterised by great details in the texture and the colours of the elements and
covers information from the observation post up to a distance of approximately
1000m (or 800m according to other approaches) [6]. It should be noted at this
point that a very close viewing zone, which is placed by many scientists at the
first 100m and by others at the first 400 m, is not examined in landscape analysis
or evaluation, because it is regarded as a general threshold for the convenient
WIT Transactions on Ecology and the Environment, Vol 84, © 2005 WIT Press
www.witpress.com, ISSN 1743-3541 (on-line)
Sustainable Development and Planning II, Vol. 2
829
stereoscopic observation of the objects [6]. The middle zone is the most
important one in landscape analysis, mainly because it gives a complete
perception of the shapes, the forms and the templates of the elements. This zone
extends from 1 to 4 km, while according to some analysts it reaches 5 km. Due
to the interference of the atmosphere, the colours in this zone are clear, while
their contrast is reduced. Finally, the long zone is dominated by the horizon and
serves as a general background, where only general information about the forms
of the elements is perceived. This zone begins where the middle zone ends and
extends to the point in which the atmospheric conditions offer visibility.
Therefore, by determining square parts of 5x5 km2, the middle viewing zone
is taken into account for the classification of the relief. In contrast to Hammond’s
method, where the square parts determined are static, the method in question
makes use of moving windows. This selection was made because the relative
observation post within each part changes from point to point, while the
researcher is interested in characterising the form of the relief, always in
combination with the observation post. Hence, it should be stressed that
following the application of the method, any landform category that could derive
characterises the specified position under examination and not all other positions
within the square part, as it is the case in Hammond’s method.
The movement increment of the window was set at 2 km. The selection of the
2 km was made based on the close viewing zone, which is set at approximately
1 km radially, around each view point. The movement increment practically
creates a template with elements of 2x2 km2 size. For each element of the
template the observation posts are selected to be in the middle of the sides, while
in each post the orientation towards the centre of the element of the template is
examined (Figure 1).
Figure 1: Dimensions and movement increment of movable window – position
and orientation.
WIT Transactions on Ecology and the Environment, Vol 84, © 2005 WIT Press
www.witpress.com, ISSN 1743-3541 (on-line)
830 Sustainable Development and Planning II, Vol. 2
The observation post is considered to be at the centre of the corresponding
side of the moving window, as it is in Figure 1. This process is repeated for each
element, and therefore the four main orientations are examined, i.e. from north,
south, east and west to the centre of all the elements (see the black arrows in
Figure 1).
2.4 Codification of the combinations
For the final classification of the landforms and in combination with the
observation post, a codification is required in order to have manageable results.
The codification selected in this paper corresponds to a number at each value
interval of each variable (Table 2).
Table 2:
Codification of the classification results.
Parameters
Relative observation post
Window percentage with slopes
smaller than 8%
Maximum hypsometric
difference (dh)
Percentage of flat slopes (<8%)
at the upper or lower half of the
hypsometric difference range
(dh)
Value intervals
Lower (< dh/3)
Middle (dh/3 up to 2dh/3)
Higher ( > 2dh/3)
> 80%
50 - 80%
20 - 50%
< 20%
0 < dh < 30 m
30 < dh < 90 m
90 < dh < 150 m
150 < dh < 300 m
300 < dh < 900 m
dh > 900 m
when flat < 20% ή dh <30
> 75% at the upper half
50-75% at the upper half
50-75% at the lower half
> 75% at the lower half
Code
1000
2000
3000
100
200
300
400
10
20
30
40
50
60
0
1
2
3
4
The final code of each observation post derives from adding up the
corresponding codes that represent the characteristics of a given relief form. For
example, if for a certain position the code is 3454, this means that:
- the observation post is the higher (code 3000);
- less than 20% of the window is covered by flat slopes (code 400);
- the hypsometric difference observed varies in the range of 300 - 900 m
(code 50);
- more than 75% of flat slopes appear at the lower part of the hypsometric
difference (code 4).
By applying this codification technique, where alpharethmetic characters are
avoided, the process of determining the appearance frequency of the various
combinations becomes easier. Actually this codification derives from applying
overlaying tools of a GIS, using as a background the elevation data of a digital
terrain model.
WIT Transactions on Ecology and the Environment, Vol 84, © 2005 WIT Press
www.witpress.com, ISSN 1743-3541 (on-line)
Sustainable Development and Planning II, Vol. 2
3
831
Applying the classification method to the study area
In order to apply the classification, the aforementioned process was planned and
a record was created for the final co-ordinates of the observation posts.
A digital terrain model was used for obtaining elevation data, which was
created by map sheets of the Army Geographical Service, of a 1:50 000 scale,
that is the most common scale in middle scale landscape analysis [9]. The data
resolution was set at 50 m, thus meeting the research requirements. The
application field of the works is depicted in Figure 2.
Figure 2: Research field – Lefka Ori of the island of Crete.
The IDRISI GIS package was used for the processing of the information, for
creating the DTM, calculating the slopes and all other derivative elements. Based
on the DTM created for the study area, a total of 1033 observation posts –
orientations were examined. Certain elements of the template, perimetrically to
the area’s bounds, were not examined, because either further hypsometric
information was needed, or there was a risk to include the sea, an element that
lies beyond the scope of this paper. From the results four maps are created. Each
map corresponds to one of the main orientations, i.e. N, E, S, W, and shows the
serial number of the element of the template and the code of the relief category
that is viewed from each element.
The resulting relief categories are presented in Table 3, with their appearance
frequencies in parenthesis. Practically 33 combinations were produced, out of a
total of 96x3 = 288 theoretically possible combinations. There were no areas in
which more than 80% of their surface was dominated by flat slopes, a fact that
was expected as the region has no large plains.
WIT Transactions on Ecology and the Environment, Vol 84, © 2005 WIT Press
www.witpress.com, ISSN 1743-3541 (on-line)
832 Sustainable Development and Planning II, Vol. 2
Table 3:
Relief classification categories in the study area.
1243
1244
1254
(2)
(3)
(2)
-
1343
(4)
-
1352 (1)
1353 (7)
1354 (39)
1364 (4)
1450 (49)
1460 (142)
2243 (2)
2244 (5)
2254 (2)
2342 (2)
2343 (3)
2344 (2)
2352 (1)
2353 (12)
2354 (36)
2364 (5)
2450 (118)
2460 (231)
3243
3244
(2)
(2)
-
3342
3343
(1)
(3)
3351 (1)
3352 (3)
3353 (8)
3354 (20)
3364 (2)
3450 (96)
3460 (223)
As Hammond also detected, a few combinations may probably appear;
however, in this particular application in Lefka Ori, our purpose is not to draw
conclusions about all possible combinations. Besides, this is already examined
by Ηammond, with the use of a greater research field (section 2.1). The issues of
particular interest in our case were: a) the way this new quantitative parameter
and the use of a moving window affect the results of the classification; and b) the
degree to which the results of the classification fulfill landscape visual analysis
purposes.
More specifically, in order to examine the second issue, certain positions
were sampled for creating perspective digital imaging representations of the
relief, with the use of a special algorithm developed in this regard [10].
According to this algorithm, for all photographic shootings we used the
geometrical elements of a 35 mm camera, with normal lens f=50 mm, strictly
horizontal shootings and a 1.5 m observation height. In order to create the
images, a DTM of the area of the Lefka Ori was used, as mentioned above. The
colour palette consists of tones that darken according to the shading and become
grey according to the distance, in order to simulate the natural observation
conditions. In Figure 3 eight images are indicatively presented out of the total
images created. The printing size of the images is in this case much smaller than
the one of a normal picture, due to limited publication space. Below each image
there are two codes: a) the serial number of the element of the shooting template;
and b) the category of the relief form, that was expected to been seen according
to the classification. The serial number of the element of the template ends to one
of the letters of the Latin alphabet, i.e. n, e, s, w, that represent respectively the
north, east, south or west orientation of the shooting.
4
Output comments
The results differentiate in relation to the ones that could have derived from
Hammond’s classification, exactly as the visual impression offered to the
observer by image 3B in comparison to image 3A differentiates.
WIT Transactions on Ecology and the Environment, Vol 84, © 2005 WIT Press
www.witpress.com, ISSN 1743-3541 (on-line)
Sustainable Development and Planning II, Vol. 2
A
B
C
128n
1243
129n
1343
156n
2244
258w
1450
166n
2450
102s
2243
246e
1460
229w
3460
833
1
2
3
Figure 3:
Relief digital imaging representation in 8 selected positions.
The images 3A and 3B depict mountains with great curvature, whereas in the
first case the position is lower (image 3Α), and in the second one higher (image
3Β); therefore, the viewshed offered by each position differentiates. Similarly,
the same applies to images 2A and 2B, but with a smaller differentiation, where
the first one corresponds to a lower position and the second one to an equal
observation post.
A lower observation post offers also a different visual impression in the case
of mountains with great curvatures (images 2A, 3Α), in relation to plateaux or
mountains having small curvatures on the top (images 1A, 1B). In the case of
images 1A and 1B, the category of the position is not readily perceived. If the
code did not exist, we would have been concerned of whether the position is
lower, equal or higher. This is due to the way we distinguish the observation
posts, namely based on the relative hypsometric difference of each form and not
on certain standardised value intervals, common for all forms. Therefore, it is
expected that in plain areas or in areas dominated by high percentages of flat
slopes, where the hypsometric difference is relatively small, the 1/3 of the
hypsometric difference observed does not produce important visual
differentiations in relation to its 2/3, especially when this hypsometric difference
happens to appear after the middle viewing zone. On the contrary, in
WIT Transactions on Ecology and the Environment, Vol 84, © 2005 WIT Press
www.witpress.com, ISSN 1743-3541 (on-line)
834 Sustainable Development and Planning II, Vol. 2
mountainous areas the observation post differentiates significantly the visual
impression of the result. This problem is however resolved by the second and the
third digit of the codes, where it is obvious whether we are dealing with a
mountain or plain area. Image 3A refers to a hypsometric difference higher than
900m, while images 1A and 1B to a hypsometric difference within the range of
150-300m.
Images 1A and 1B are shootings towards the same direction of two
horizontally (to axis X) continuous square parts (128 and 129). These two
images show that the horizontal movement increment of the moving window is
satisfying, as it is neither too small so as to cover the forms depicted, not too big
to lose the coherence and the sequence of the forms. Images 1B and 1C are
shootings towards the same direction of two vertically (to axis Υ) continuous
square parts. It is obvious that the relief elements that are depicted in the first
case in the middle viewing zone (1B), are represented in the following position
in the long viewing zone. Therefore, the movement increment selected is
considered to sufficiently secure, both horizontally and vertically, the coherence
and the sequence of the relief forms.
When comparing images 1A and 2C, that present the same relief
characteristics, according to the three last digits of the code, and while we would
have expected that image 2C depicts a similar plain landscape to image 1A, this
looks more like a hilly rather than a plain landscape. This fact is due to two
possible reasons: a) the different hypsometric differences of the two forms,
which still belong to the same category (perhaps the two opposite sides of the
range of the specified category); and b) the different horizontal distances of the
minimum and maximum elevation values from the observation post. In the
second case, even if hypsometric difference is the same in both forms, the
perspective height varies and therefore the impression offered.
5
Conclusions
From the abovementioned observations we can conclude that even within the
same classification category, there is still a variety of forms. This is however
inevitable because: a) classification presupposes generalisation; and b) the
perspective observation conditions cannot be accurately measured and predicted
by topographic maps (ground plan), without using a certain representation
perspective. Besides, what matters in this case is to have the classification prior
to the creation of the images, and without calculating the viewshed which is a
time consuming procedure. The classification method in question tries to predict
the perspective observation conditions and not to use them as entry elements. To
summarise, the relief classification proposed is regarded as successful with
regard to the objectives for which it was designed, since it describes sufficiently
to a great extent the visual impression made to the observer by a certain relief
form.
A further research may lead to the classification of the relief throughout the
Greek territory, but also of other areas, and to the determination of all the
morphological combinations that may probably appear. Therefore we could
WIT Transactions on Ecology and the Environment, Vol 84, © 2005 WIT Press
www.witpress.com, ISSN 1743-3541 (on-line)
Sustainable Development and Planning II, Vol. 2
835
compare various areas in order to show the uniqueness of certain combinations
and, based on the results, to proceed to a generalisation of the categories in
greater ones. However, the results of this particular classification are useful in
landscape analysis, even without a further generalisation, taken into account that
the uniqueness of the relief form constitutes a criterion in visual landscape
evaluation [8]. This information together with other data of other variables of the
physical and visual environment, such as vegetation, water resources, etc., may
even lead to the creation of complex landscape maps.
The topic of this paper involves human perception and the physical
environment. The relevance of the visual landscape classification and evaluation
in terms of landscape dynamics, theoretical ecology and eco-informatics is
obvious. It is difficult to model human perception given that there are many gaps
in our understanding of human perception. The use of G.I.S. to model perception
should be encouraged because of its relevance.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
Ananiadou-Tzimopoulou M., Landscape analysis in Designing.
Contribution to Landscape Architecture Research. PhD. Dissertation.
Scientific Journal of the Polytechnic School of the Aristotle University
Thessaloniki (in Greek), 1982.
Bradyn L.K., Landscape classification using GIS and national digital
databases. Landscape Research, Vol. 21, No 3, pp. 277-300, 1996.
Bradyn L.K., Classification of macro landforms using GIS. ITTC journal,
1997-1, pp.26-40.
Dikau R., Brabb E.E., Mark R.M., Landform classification of New
Mexico by computer. U.S. Department of the Interior, U.S. Geological
Survey. Open file report 91-634, 1991.
Cole N, Ferraro M, Mallary R, Palmer J, Zube E., Visual Design
Resources for Surface Mine Reclamation. Institute for Man and
Environment / ARSTECHNICA: Center for Art and Technology
University of Massachusetts. Amherst, 1976.
Felleman P.J., Landscape Visibility. Foundations for Visual Project
Analysis. John Willey & Sons, N.Y., pp. 48-62., 1986
USDA Forest Service, The Visual Management System. Government
Printing Office: Ag. Handbook 462. Washington, 1973.
USDA Forest Service, Landscape Aesthetics. Government Printing
Office: Ag. Handbook 701. Washington, 1995.
Smardon CR, Felleman PJ, Palmer FJ., Decision Making Model for
Visual Resource Management and Project Review. Foundations for Visual
Project Analysis, John Willey & Sons, N.Y., pp. 21-35. N.Y., 1986.
Tsouchlaraki A., Digital Relief Visualisation in Landscape Analysis.
Technika Chronika. Scientific edition of the Technical Chamber of
Greece: I, issue 3, pp. 27-40. Athens (in Greek with English extended
summary), 1996.
WIT Transactions on Ecology and the Environment, Vol 84, © 2005 WIT Press
www.witpress.com, ISSN 1743-3541 (on-line)