INTRODUCTION
Geomorphometry, or simply morphometry, provides a quantitative description
of the shapes of landforms and is derived using a combination of mathematics,
engineering and more recently, computer science. According to Blaszczynski
(1997), landforms are defined as specific geomorphic features on the earth’s
surface, ranging from large-scale features such as plains and mountain
ranges to minor features such as individual hills and valleys. A topographic
position such as a hilltop, flat plain, valley, etc., is intuitively important
for physical and biological processes acting on the landscape. Natural
habitats of plants, erosion potential and solar radiation are directly
related to landform patterns and the relative position with a landform
(Blaszczynski, 1997).
In the past, geomorphometric properties have been measured by calculating
the geometry of the landscape manually (Horton, 1945; Miller, 1953; Coates,
1958). Unfortunately, measurement of such properties from topographic
maps can be time consuming. In the 1960s and early 1970s, the generally
availability of computers made possible more complex, statistically-based
methods to identify landscape features (Chorley, 1972; Evans, 1972). More
recently, advances in computer technology, increased processing power,
new spatial analytical methods and the increasing availability of digital
elevation data have re-oriented geomorphometry (Pike, 1999) and promoted
the development of computer algorithms for calculating and discriminating
geomorphometric properties of the Earth’s surface.
Digital Elevation Models (DEMs) or Digital Surface Models (DSMs) are
digital representations of topography or terrain and have been in use
since the early 1970s. DEMs and derived data sets (slope, aspect, surface
area, hydrographical pattern and shaded relief) have been exploited by
investigators for geomorphologic and geomorphometrical studies (Franklin,
1987; Skidmore, 1989; Ventura and Irvin, 2000). Nowadays, terrain analysis
is one of the more interesting and engaging types of geographic analysis
in order to describe topographic position (Speight, 1990). Terrain data,
when used in conjunction with multispectral imagery, also provides rapid
and useful information on landscape geology, lithology, soils, land use
and landcover, lineaments etc.
The drainage basin, or watershed, is the fundamental unit in geomorphology.
This study focuses on the upper part of the Kozdere basin, which includes
Yazoren polje, due to its large variety of topographic features (Hosgoren,
1981). The main objectives of this research are to semi-automatically
identify and classify the landforms within the watershed by applying advanced
spatial statistics and image processing algorithms to DEMs. Morphometric
techniques described in this paper are used to measure and mathematically
model the Earth’s surface and provide objective and quantitative descriptions
of landforms (Bates and Jackson, 1987; Denizman, 2003).
MATERIALS AND METHODS
The study area is located in Marmara Region in Turkey between latitudes
39°17’N and 39°22’N and longitudes 27°44’E and 27°53’E
(Fig. 1). It includes the settlements of Yazoren-Urbut
(Fig. 1, 2). The total basin area is
~44.2 km2 and the average elevation is ~380 m above mean sea
level.
This area experiences a strongly seasonal Mediterranean climate. Summers
are quite hot and dry while winters are warm and wet. According to the
Balikesir meteorological data, the average temperature is nearly 14.5°C
and the total amount of rainfall in a year is 545 mm (Tagil, 2004). The
geomorphology of the area reflects Alpine and post-Alpine tectonic activity.
The Yazoren-Urbut graben basin separates the existing mountains.
Field surveys revealed that land cover is extremely heterogeneous due
to the large variety of adjacent topographic features. For instance, cultivated
areas alternate with non-cultivated and natural lands are mixed with crop
cultivations (Fig. 2).
Caltepe formations and Yuntdag volcanics are widespread in the region,
while Akcakoyun formations are seen on only a small part of the study
area. These three formations occupy most of the mountainous portion of
the region. The Caltepe Formation, formed during the Upper Triassic period,
is the oldest autochthonous unit in this area and consists of limestone
originating in subduction-accretion units of the Palaeo-Tethys Ocean (Tekeli,
1981; Ercan et al., 1990; Fig. 3). The Caltepe
Formation is overlaid by sandstone, mudstone and limestone from the Early
Cretaceous Akcakoyun formation. The Akcakoyun Formation is unconformably
overlaid by the Yuntdag volcanics from the Late Miocene to Pliocene. The
Yuntdag volcanics cover a large areal extent and are composed of agglomerate
and tuff. Quaternary alluvium is the most recent formation and is restricted
to the valley bottom west of Yazoren.
DATA PROCESSING TECHNIQUES
Input data: The data used in this study consists of: (a) topographic
maps from the Turkish Military Geographic Service on a scale of 1:25,000 (Contour
interval 10 m); (b) a Landsat Enhanced Thematic Mapper Plus (ETM+) satellite
image taken on July 2, 2000 (spatial resolution: 28.5 m); (c) a geologic map
from the General Directorate of Rural Services on a scale of 1:25,000; (d) a
Digital Elevation Model (DEM); (e) GPS (Global Positioning System) data collected
in the field by using a handheld GPS receiver.
Nearest neighbor resampling was used to georectify the LANDSAT image
to the Universal Transverse Mercator map projection (UTM Zone 35, WGS84)
using 50 Ground Control Points (GCPs) with an RMS < 1 pixel.
|
| Fig. 1: |
Topographic overview of the study area and its location |
|
| Fig. 2: |
Topographic overview of the study area and its location |
|
| Fig. 3: |
Topographic overview of the study area and its location |
The GIS software programs ArcView 3.2 and ArcGIS 9.2 were used for the
morphometric analyses. Data analysis was divided into two main sections:
1) digital terrain analysis, using general morphometric techniques to
analyze the entire surface and 2) digital landform analysis, involving
specific morphometric methods to examine particular landforms in the study
area.
DEM: DEM was generated by first digitizing 10 m interval contours
from 1:25,000 Turkish Military Geographic Service topographic maps. Then,
a 20 m DEM was interpolated from contour lines using the Topo to Raster
tool in ArcGIS 9.2, which converts vector topographic lines to a raster
DEM surface using the ANUDEM algorithm.
Watershed delineation and study area: The drainage network, watershed
and sub-watersheds used in this paper were delineated automatically from
the DEM using the Arc Hydro tools of ArcInfo. Flow direction, flow accumulation,
stream definition, stream segmentation and watershed delineation were
determined using standardized raster methods described by Djokic et
al. (1997) and ESRI (1997). The raster analysis area in this study
includes the study area watershed plus a 500 m buffer. This buffer is
included to account for a common problem in raster analysis, in that many
analytical functions require information from a neighborhood around each
grid cell. When analysis is limited to the exact extent of the study area,
then cells at the edge of the study area may not have complete neighborhoods
available. The 500 m buffer used in this study guarantees that all grid
cells will have a full neighborhood available when calculating neighborhood
statistics.
Landcover classification: Landcover was classified using the Guided
Clustering method, which is a hybrid supervised-unsupervised classification
approach outlined in Messina et al. (2000). A 3×3 median filter
was applied to remove speckled pixels, random pixels in the middle of
a main class and to refine the classified maps. Attributes were assigned
to classes using GPS data, Normalized Difference Vegetation Index (NDVI),
field knowledge, air photos and different band combinations. Three different
landcover classes were extracted: forest, shrub and brush rangeland and
bare exposed soil and rock (Fig. 4). The bare exposed
rock and soil class also includes agricultural areas and built-up areas.
The overall accuracy of the landcover classification was 80%.
Morphometric analysis (Topographic attributes): The topographic
attributes of slope, aspect, curvature (planform), topographic wetness
and stream power were computed from the DEM (Fig. 5).
Slope and aspect maps show the magnitude and direction of the vector tangent
to the topographic surface pointing downhill at a point.
Planform curvature, calculated using standard ESRI functions based on
the algorithms of Zevenbergen and Thorne (1987), provides a measure of
how water converges or diverges as it flows through the landscape.
|
| Fig. 4: |
Landcover classes classified from a Landsat Enhanced
Thematic Mapper Plus (ETM+) satellite image taken on July 2, 2000 |
|
| Fig. 5: |
Topographic attributes: elevation (meters), slope (degree)
aspect (degree), planform curvature, topographic wetness and stream
power |
Planform curvature is calculated using a 3×3 cell neighborhood, so in
our case it represents the curvature in a 60x60 m area. Planform curvature
values were reclassed to concave, linear and convex based on the following
criteria: if n‘0.1, the planform curvature is convex and flowing water
will tend to diverge; if -0.1‘n<0.1, the curvature is linear and
if n<-0.1, the curvature is concave and flowing water will tend to
converge.
Topographic wetness and stream power indices were used to quantify flow
intensity and accumulation potential. Topographic wetness (also known
as Compound Topographic Index [CTI] or topographic moisture index) at
a particular point on the landscape is the ratio between the catchment
area contributing to that point and the slope at that point (Wilson and
Gallant, 2000). Higher positive values are wetter and lower negative values
are drier and values are calculated as:
The Stream Power Index (SPI) is closely related to the topographic wetness
index and is used to estimate the erosive power of the terrain. Areas
with large stream power indices have a great potential for erosion. If
total stream power is greater than that required to transport the sediment
available (supply limited), then there will be a net loss in sediment
and the stream will erode. If stream power is less (transport limited)
than that required, then there will be a net gain in sediment and the
stream will aggrade. Values are calculated as:
| Stream power index |
= |
Catchment areaxtan β (Moore et al., 1993) |
| where β |
= |
Slope in degrees |
In order to remove spurious features, the resulting wetness map and stream
power map were filtered using the “majority filter” routine
(3×3 scanning window).
Topographic Position Index (TPI): Topographic Position Index (TPI)
is the difference between the elevation at a cell and the average elevation
in a neighborhood surrounding that cell. Positive values indicate that
the cell is higher than its neighbors while negative values indicate the
cell is lower. TPI is a simplification of the Landscape Position Index
described by Fels and Zobel (1995) and was developed in detail by Weiss
(2001). TPI values provide a simple and powerful means to classify the
landscape into morphological classes (Jenness, 2005).
|
| Fig. 6: |
TPI grids using 6 different neighborhood sizes |
The neighborhood size and shape is critical to the analysis and should
be based on the scale of landscape feature being analyzed. To classify
very small features like small streams or drainages, a small circular
neighborhood was used. To identify large canyons or mountains, a large
circular neighborhood was used. Choosing the correct neighborhood is generally
an iterative process in which several options are tried before the most
useful sizes are identified. In this study, TPI grids generated from 50,
100, 150, 200, 250 and 450 m neighborhoods was presented (Fig.
6).
Slope position classification: Next, TPI values were used to classify
the landscape into slope position classes. This classification is based
on how extreme the TPI values are and by the slope at each point. Logically,
high TPI values would be found near the tops of hills while low TPI values
would be found in valley bottoms. TPI values near 0 would be found on
either flat ground or somewhere mid-slope and slope values are used to
distinguish between these two possibilities. In this study, a 4-category
slope position grid from each of the 6 TPI grids was generated (Fig.
7). A TPI threshold value of ±1 SD was used to identify hilltops
and valley bottoms, where the standard deviation value was calculated
from all elevation values in the watershed analysis area. A slope threshold
of ±6° was used to distinguish between flat areas and mid-slope
areas (Table 1).
Landform classification: TPI values calculated from two neighborhood
sizes provide more information about the general shape of the landscape
than TPI values from a single neighborhood and therefore more complex
landscape features can be identified by combining TPI grids generated
at different scales. A point on the landscape with a negative small-neighborhood
TPI value and a positive large-neighborhood TPI value is likely to represent
a small valley on a larger hilltop. Such a feature may reasonably be classified
as an upland drainage. Conversely, a point with a positive small-neighborhood
TPI value and a negative large-neighborhood TPI value likely represents
a small hill or ridge in a larger valley.
|
| Fig. 7: |
Slope classes using TPI grids from 6 neighborhood sizes |
|
| Fig. 8: |
Landforms using Weiss (2001) classes based on 50 and
450 m TPI |
In our case, 50 and 450 m TPI grid, in combination with slope were used
and classified landforms using criteria described by Weiss (2001) (Fig.
8; Table 1). As with the slope position classifications,
high and low TPI values were distinguished by setting a threshold of ±1
SD. In cases where TPI values from both neighborhood sizes were between
-1 and 1, small plains and midslope areas were distinguished by using
a threshold slope value of 6°. A full description of each morphological
classification can be found in Weiss (2001) and Jenness (2005).
Relationships between landscape classifications (slope position and landforms)
and topographic attributes (elevation, slope, planform curvature, topographic
wetness and stream power) were determined by using the standard ArcView
Zonal Statistics function to calculate statistics for topographic attributes
within each landscape class. Relationships between landscape classifications
and geology and landcover classes were analaysed by using the standard
ArcView Tabulate Table function to calculate the proportions of landscape
classes that corresponded with each geology and landcover classes.
RESULTS AND DISCUSSION
Slope position classification: From the 6 neighborhood sizes were
tested, it was found that the 100 m neighborhood did the best job at extracting
the terraces and small karstic depressions features interested in identifying
within the watershed. These features were not large enough to be extracted
by larger neighborhoods and the smaller 50 m neighborhood tended to extract
only the edges of the features rather than the features themselves.
The slope position classification results for all 6 neighborhood sizes
are shown in Fig. 7. Within the 100 m neighborhood classification,
66,6% of the watershed was classified as Mid-slope, 14.4% Valley, 14.2%
Hilltop and 10.8% was classified as Flat Surface.
Topographic attributes of landform classes: Figure
8 provides an overview of the area with the primary landforms. More
than 50% of the area is classified as open slope, which is not surprising
because is the area is intersected with several rivers. Slope, topographic
wetness, stream power, planform curvature and elevation characteristics
of the landforms are shown in Table 2.
It is interesting to note that the classes “Canyons, Deeply
Incised Streams”, “Midslope Drainages, Shallow Valleys” and
“Upland Drainages, Headwaters” all tended to have strongly
negative planform curvature values, while “Local Ridges/Hills
in Valleys”, “Midslope Ridges, Small Hills in Plains”and
“Mountain Tops, High Ridges”all tended to have strongly positive
planform curvature values. The classes with negative curvature values
all corresponded to negative small-neighborhood TPI values, while the
classes with positive curvature all corresponded to positive small-neighborhood
TPI values. Curvature is always calculated from a 3×3 cell neighborhood,
so curvature in our example was calculated using a square 60 m neighborhood.
This neighborhood size is very close to our 50 m radius circular TPI neighborhood,
so it is not surprising that planform curvature tends to be appears to
be correlated with our small-neighborhood TPI grid.
Stream power is directly related to both slope and catchment area, so
it is also not surprising that stream power was strongest where either
the small-neighborhood or large-neighborhood values were negative. Negative
TPI values indicate that a cell is lower than its neighbors and such cells
would be expected to have a larger catchment area than cells that are
higher than their neighbors.
Geologic and landcover characteristics of landform classes: As
seen in Table 3, 92.8% of alluvial surfaces were classified
as U-Shaped Valleys, Plains or Open Slopes (i.e., the three highest values:
17.6+ 58.9+ 16.3%), 59.5% of the agglomerates were classified as open
slopes and upper slopes; 88% of the sandstones, mudstones and limestones
were classified as open slopes, midslope ridges, small hills in plains
and plain small and 48% of limestones were open slopes. This is evidence
that geology has great importance on landforms of the watershed.
| Table 1: |
Descriptions of landform classes and slope position
classes |
 |
| Table 2: |
Zonal statistics as table which shows values of topographic
attributes raster (slope, topographic wetness, stream power, planform
curvature and elevation) within the zones of landscape (SD: Standard
Deviation) |
 |
| Table 3: |
Relationships between landform classes and geology,
landcover and slope classes* |
 |
Geology: G1: Alluvium, G2: Aglomerate G3: Sandstone,
mudstone and limestone G4: Limestone; Landcover: LC1: Forest, LC2:
shrub and brush rangeland, LC3: Bare exposed bare exposed soil and
rock, *: Each cell contains two values, where the top value represents
the proportion of the row (Landform) class calculated to be in the
column class and the bottom value (in bold) represents the proportion
of the column class that was calculated to be in the row class |
Landcover classes also demonstrate interesting relationships with landform
classes (Table 3). Canyons and deeply incised streams
appear to have a lot of forest; plains have a lot of bare rock and exposed
soil, upper slopes appear to have relatively little bare rock and exposed
soil; local ridges/hills in valleys appear to have a lot of forest and
relatively little bare rock and exposed soil and mountain tops/ridges
appear to have a lot of forest and relatively little bare rock and exposed
soil. The hugely disproportional amount of bare rock/exposed soil found
in plains may be accounted for by the fact that these areas are primarily
agricultural and the bare rock/exposed soil class includes agricultural
areas.
Three different erosional or denudational surfaces can be identified
using TPI. These denudational surfaces were identified as “Upper
Slopes, Mesas” in the Weiss landform classification scheme. The lowest
level denudational surface was observed at 100-150 m relative height.
The mid-level denudational surface is found between 250 and 300 m relative
height and cuts the Miocene - Pliocene volcanic formations. The highest
level denudational surface is found at 400 m relative height, at the border
of the watershed (Fig. 9). According to Erol (1983),
the lowest surfaces date to the Lowest Pleistocene period; the mid-level
surfaces are from the Pliocene period and the highest surfaces date to
the Upper Miocene period.
The depression formation in the centre of the region is known as Yazoren
Polje, a primarily agricultural area. This flat surface was identified
as a plain in the Weiss landform classification scheme. The plain transitions
rapidly to open slopes along the northern and southern boundaries This
sudden slope change, located along the normal fault lines illustrated
in Fig. 3, suggests that normal faulting may contribute
to the depression formation. The plateau along the northern border of
the polje rises 100-150 m relative to the depression and the southern
border of the polje is bordered by a plateau rising nearly 200 m relative
height (Fig. 2, Fig. 9a-b). This 100
m difference is likely the result of uplifting along the normal fault.
It is hypothesized that Yazoren Polje and its watershed have been drained
by the gorge at the elbow of capture. According to Ardel (1958) and Kazanci
and Gorur (1997), the Pleistocene is a erosional period in the Southern
Marmara region of Turkey and closed depressions are open with epigenetic
gorge or superposed valley in the Southern Marmara region of Turkey. According
to Erol (1992) and Atalay (1998), Pleistocene climatic change led to different
terrace systems and down cutting began in the lowest Pleistocene and continued
through the low, middle and late Pleistocene. This capture process, occurring
because of the ongoing uplift and compression of the landscape, has been
historically common in the Marmara region of Turkey (as well as throughout
Turkey in general) as Neotectonic activities and Quaternary sea level
fluctuation largely control the development of this region. The river
incised its valley during the Quaternary period and thus played a major
role in the development of the landforms present in the area today (Atalay,
1998).
|
| Fig. 9: |
Cross-sections of watershed of Yazore Creek |
Two shoulders or narrow terraces along the slopes of the gorge can be
identified with TPI (Fig. 8). Terrace surfaces are found
along local ridges or mid-slope ridges. It appears that the gorge was
incised over a very long time. The terrace around the gorge observed by
myself at 60-70 m above the floodplain; lowest one is observed at 20-30
m above the floodplain (Fig. 9c-e). These same terrace
surfaces were identified by Tagil (2004) near the Balikesir plain, the
nearest plain located 28 km in NE direction. The lowest terrace could
not be determined in the gorge itself, but it is apparent at the entrance
and exit of the gorge and also in the southern part of Yazoren Polje.
Because the capture gorge cut into the Yuntdag volcanics of the Late Miocene
to Pliocene age, the capture therefore occurred after the Pliocene. This
suggests that the capture and gorge were formed during and after the Pleistocene.
The landform application can help us to detect not only open depressions
but also closed depressions, river channels and less obvious features
such as vales. Figure 8 shows several closed karst depressions (Kizilgol
pit and lots of unnamed depressions) that were misclassified as deeply
incised streams and U-shaped valleys. Sharp edges mark the boundaries
of these karst depressions. While the interior lower level area is classified
as flat or plain, the areas around these small depressions are classified
as open slopes due to its elevated position compared to the depression
bottom. The smallest of these karstic depressions are nearly 500-1500
m in diameter and 150-500 m in length. They vary in size.
CONCLUSIONS
Landscape surveying and mapping techniques based on geomorphologic analysis,
combined with GIS and remote sensing techniques, are useful tools for
natural resource management. In this study, TPI was used to generate morphological
types for a semi-automated derivation of landform elements. With the presented
methodology landform elements have been generated according to Weiss (2001)
and the results reflect his definitions. Digital elevation models offer
many potential habitat descriptors other than simply a set of elevation
values. DEMs can yield a wide variety of landscape morphological characteristics
which may be important to wildlife, land managers and researchers. This
article describes some of the analytical approaches and topographic concepts
and shows how they are useful to explain geomorphological processes.
There are occasionally problems with this method. In this semi-automatic
landform model, smaller creeks got lost or were at least poorly identified.
For example, narrow creeks running across wide flats without forming a
broader valley were not identified, although they might be identifiable
using more sensitive classification criteria. Some areas simply cannot
be handled correctly by the computer if they extend beyond the boundary
of the DEM. Except cases such as these: TPI provides a powerful tool to
describe topographic attributes of a study area.
Subscribe Today
