An attempt for automatic detection and visualization of talus
cones from digital elevation data
Balázs Székely1,2 and Tomaž Podobnikar1,3
Dr Balázs Székely and Dr Tomaž Podobnikar
1
Christian Doppler Laboratory Spatial Data from Laser Scanning and Remote Sensing,
Institute of Photogrammetry and Remote Sensing, Vienna University of Technology
Gußhausstraße 27-29, A-1040 Vienna, Austria
Tel: +43 1 58801 12210, Fax: +43 1 58801 12299
2
Space Research Group, Department of Geophysics and Space Science, Eötvös
University
Pázmány Péter sétány 1/a, HU-1117 Budapest, Hungary, bz@ipf.tuwien.ac.at
3
Scientific Research Centre of the Slovenian Academy of Sciences and Arts
Novi trg 2, SI-1000 Ljubljana, Slovenia, tp@ipf.tuwien.ac.at
Abstract
The importance of methods for automated feature extraction from digital elevation data is increasing as
the amount of data to be processed multiplies in size. Landscape elements can be extracted from the data
if the common characteristic parameters can be identified and an effective searching algorithm can be
found to identify the feature.
In our approach we focus on actively evolving slopes in alpine areas, where sediment redeposition takes
place in response to postglacial processes. Scree-covered valley slopes, oversteepened at places, may
become unstable in long-term due to the active geomorphic evolution. Mass movements, partly induced
by seismic activity, may cause significant damages, therefore automated delineation of talus surfaces is
importance for assessment purposes.
In our pilot study we developed tools for automatic identification of scree slopes based on sectorial
analysis of slope histograms. The preliminary results are promising; however, the parameter selection
can only be done yet manually.
1. INTRODUCTION
1.1 Significance of the automated topographic feature extraction
The development of current technologies of data acquisition multiplied the amount of available accurate
digital elevation data. Photogrammetric, LiDAR-based (WAGNER et al. 2006, RONCAT et al. 2007, in press)
and other techniques provide hundreds of millions of measured surface points; the horizontal resolution of
such datasets reached now the sub-meter range. Consequently, the size of the digital database of a project
exceeds the order of magnitude of terabytes. Of course the data acquisition is carried out to achieve a
certain goal of the project; on the other hand, the re-use and re-evaluation of the expensively obtained data
is more and more common, as new aspects of analysis appear (e.g., PIKE 2000, TIMÁR et al. 2005).
Although the manual feature extraction remains important for the next decades as well, these vast datasets
generate imminent need for fast and automated processing techniques, especially in the field of feature
identification. Feature extraction are typically intended to get vectorized in formation of the geomorphic
features like sinuous parts of a river (e.g., TIMÁR et al. 2005, ZÁMOLYI et al. 2007) or peaks (e.g, SZÉKELY
and KARÁTSON 2004, CHANG and SINHA 2007).
Feature extraction in this sense can be split into to phases: an identification phase when the possible feature
candidates are marked by an automated process, and a verification and extraction phase in which the
preselected feature set is manually or semi-automatically reduced to the features that we are looking for.
Feature extraction methods are applied widely in geomorphic, natural hazard assessment, archaeological
and other studies, since they provide preprocessed feature sets that allow focused research work.
One of the most frequently applied and even more rapidly expanding use of the digital elevation data is the
assessment of the stability of slopes. This analysis is especially crucial in mountainous environment, where
rapid evolution of topography is due to the on-going orogeny, the effects of the deglaciation, climate
change (Fig. 1). In alpine settings where the topographic evolution is mostly driven by climate change, the
analysis of digital elevation data has attracted much interest, because unstable valley slopes, often covered
by post-glacial talus cones, may suffer various-scale mass movements including rock avalanches, rock and
landslides. In these areas the scree slopes are often unvegetated, however in some settings young forests or
shrubs may cover them. Since these slopes are mostly of loose, poorly sorted material, decadal
meteorological events, or excess discharge from glacial or snow meltwater may lead to sudden incision.
Figure 1 A glacially overdeepened valley in Vorarlberg (Austria) in the Eastern Alps. The oversteepened
valley slopes start to be reshaped by rock falls, rock avalanches and other mass movements forming a talus
coverage of various thickness. It is important to note that the age of the talus surface is different: the age of
the vegetation gives hints for the sequence of events, from bare rock surface (very young) to forest covered
(several decades old).
Such suddenly incised valleys (that actually cause local base level drops) then may pave the way for
medium-sized or major mass movements. The slopes slightly steeper than the angle of repose, under certain
circumstances, may remain temporarily metastable, especially if they are vegetated. Removal of vegetation
(deforestation or other change in the land use) may change their stability situation. If such slopes are also
connected with outlets of hanging valleys that are common in post-glacial geomorphic settings, the surplus
in sediment discharge from the upstream area is completely deposited at the slopes of the talus cones. In
long run the slopes become even more oversteepened, therefore increasingly unstable.
Thus, the identification and monitoring of such talus cones on decadal scale are of major importance. Hill
foot areas often host man-made structures as well, so the potential damage of mass movements are typically
high (Fig. 2). In this contribution we intend to tackle the problem of the automated extraction of such
surfaces in order to be able to create maps and analyse the properties of these slopes. With unconventional
processing techniques applied on high-resolution, high-accuracy digital terrain models (DTM) we intend to
detect automatically well developed talus cones. In this paper the results of the first successful experiments
are presented.
Figure 2 A densely populated postglacial talus cone in Gaschurn, Vorarlberg (Austria). The smooth and
wavy surface abruptly changes on the right side of the image to a steeper slope angle. This is due to the
rapid incision of the river Ill. In long term this incision will lead to the metastable status of the talus cone
that may endanger the build-up structures. Automated recognition may help to identify such places that can
be affected by incision-induced mass movements.
1.2 Theoretical approach
The most common spatial analyses in geographical information systems (GIS) include topologic and
cartographic modelling, modelling of networks, automated cartography, map algebra and spatial statistics.
A GIS is considered as a suitable tool for the environmental modelling, designed for creating, storing,
updating, analysing, displaying, and managing (manipulating) spatial data and associated attributes (ESRI
1997). GIS applications are common in business, government, research, the internet, etc. While it has been
truly operational since mid 1960s, its usage has grown significantly since the 1980s. The strength of GIS
lies in its ability to visualise and analyse spatial patterns (TOMLINSON 2003), as well as to allow exploration
of more complex interactions between social and natural spaces. Spatial analysis that builds the models,
could be classified as descriptive, explanatory, predictive and normative (CHOU 1997). In our case, a
descriptive analysis evaluates the data and their suitability for explanatory models that are goal of this study.
The other approach to spatial modelling (ANSELIN 2005) starts with exploratory analysis to find interesting
patterns, continue with visualisation for showing the patterns and then with spatial modelling for explaining
the patterns that is also part of the study procedure. As part of the modelling procedure other type of
models could be involved, e.g. regression, Boolean, empirical ones. Therefore, spatial modelling combine
different spatial analyses regarding of decision-making process and even more, the data (variables) and
models could be part of environmental decision support system (KANEVSKI and MAIGNAN 2004) for
estimation of selected environmental phenomena, e.g. erosion, potential mass movements.
In this study the primarily interest is to apply the raster-based methods on continuous surfaces – digital
terrain models (DTMs). Spatial analyses are mostly analytical operations of continuous surfaces
(classifying and reclassifying, overlaying, geometrical operations, neighbourhood operations) and spatial
interpolations (deterministic and stochastic, global and local methods, etc.). We apply here classifying, an
operation for merging attributes to selected classes. Continuous values have to be classified to form
categories, and then reclassification is often needed, an operation where more classes are reduced to fewer.
It is usually used for converting nominal or ordinal values to a Boolean binary form.
Since we intend to have one single class (the slope features to be found) and another non-classified group at
the end of the processing reclassification has to be applied. To achieve that, however, previous data
reduction steps are needed. With overlaying, that is an analytical operation that combines two or more data
sets to a new one on a same area, the data sets are compared by logical operations (Boolean algebra: AND,
OR, NOT, etc.) on binary data sets, or processed by arithmetical operators +, –, ·, /, etc. and relational
operators (>, =, etc.). In our case the numerous indicator functions can be merged this way.
Last, but not least, analytically important processing in GIS that is related with classification and
overlaying operations is map algebra (TOMLIN 1990). Sophisticated geometrical operations offer
calculations of distances, areas, connections, directions, etc., for example Euclidean distances and surfaces,
buffer zones, cost distances, Thiessen polygons, etc. Instead of that, for simplicity and for a better
performance neighbourhood operations have been used that operate with windows of different sizes and
forms (e.g. square window 3 x 3 cells) for calculation of slope, aspect, hill shading, visibility analysis, etc.
This analysis is usually carried out on a DTM. In a later stage, if the behaviour of the data is better known
in various environments (e.g., crystalline, calcareous rocks), more sophisticated and computer-exhaustive
processing techniques can be applied to minimize the manual input and heuristic parameter settings.
2. DATA AND METHODS
2.1 The study area
Our primary selection criterion for the study area was that, from geomorphic point of view, the digital
elevation data of the area should be as accurate as it can be. Another important quality requirement was the
homogeneity of the data set, especially concerning the error level. As it was mentioned above, due to the
various resolutions of the DTMs, the calculated slope angle may vary and – depending on the actual relief
of the area – the DTM may turn to be less suitable for the slope angle calculation. For test purposes,
therefore, a small part of the upper valley of the Sava Dolinka river was chosen in the NW part of Slovenia
near to the border with Austria and Italy (Fig. 3). The high geomorphological quality Digital Terrain Model
of Slovenia (DTM of Slovenia; © 2005, Surveying and Mapping Authority of Republic of Slovenia) has
been compiled from various data sets using high quality control techniques (PODOBNIKAR 2005). DTM of
Slovenia is available in resolution of 12.5, 25 and 100 m. Of these resolutions the higher ones have been
used. For slope angle applications the optimal resolution can be determined (HUTCHINSON and GALLANT
2000). It is found that not always the highest resolution is the optimal for this purpose (e.g., ZÁMOLYI
2006).
Figure 3
A study area of the upper valley of the Sava Dolinka river in the NW part of Slovenia with
dimensions of 39 km by 15 km.
The extent of the study area is approximately 39 km by 15 km. The Sava river drains there the mountain
range of Karavanke (in the north) and Julian Alps (in the south). Due to the different lithological properties
of these two regions, the drainage pattern, the denudation history, the sediment production, and all related
phenomena can be very different. As a consequence the river finds its thalweg according to these properties.
However, the karstification of the calcareous lithological units considerably modifies the drainage pattern.
2.2 DTM derivatives
The detection of the potential areas of the talus cones bases on overlays different independent variables that
produce a potential surface of the talus cones. The variables and their connection with the talus cones are
described as follows.
Visibility simulation
The aim of this variable was to distinguish between the areas that are at the bottom of the relative relief and
those that are exposed to the simulated sun-motion (PODOBNIKAR 2008). We suppose that the talus cones
are located near to the bottom of the relative relief. This variable was calculated from the DTM 12.5.
Calculating the exposedness of the terrain in different scales is the aim of this group of methods. The output
may be the visualisation of the terrain surface that shows areas which are differently exposed to a virtual
light source thus allowing the distinction between more or less prominent topographic features (Fig. 4a).
Curvature
The talus cone has a curvature over all of area, but it is very low. A positive curvature indicates that the
surface is convex and a negative curvature that the surface is concave. The talus cones have a slightly
convex curvature. This variable was calculated from the DTM 25 as a geomorphologically precise and very
reliable data set (Fig. 4b).
Slope
Talus cones have particular ranges of inclination angles along the gradients. This variable was calculated as
well as from the DTM 25 (Fig. 4c). The importance of the local slope histogram is very high in the
geomorphometric studies, since it is typically invariant in case of tectonic movements, uplift, subsidence,
etc. (SZÉKELY et al. 2002). The slope histogram can be considered as “topographic memory” for longer
periods of time.
Variables based on a local window
Additional variables that base on a local window calculation were computed. The local window has a shape
of an annular wedge. The wedge was constructed on such way that imitated the shape of the typical talus
cones with certain parameters: radius, wedge, and azimuths in four sectors:
east:
45º to 135º
south: 135º to 225º
west:
225º to 315º
north: 315º to 360º and 0º to 45º
Different operations on the local windows were calculated. If the values of the local windows fit well to the
talus cones, then low range and standard deviation between the windows’ values is expected. Of course, the
standard deviation should not be too low, that could mean completely flat areas (Fig. 4d).
Aspect
The talus cones are directed towards a particular direction, unless their horizontal radius is low. With
applying four sectors – azimuths toward east, south, west and north direction, the particular talus cones are
more precisely classified considering of the all listed variables (Fig. 4e).
2.3 Modelling
All of the described variables are classified by tuning with some of selected talus cones in the study area.
The main criteria for typical talus zones are: talus zones are relatively in the bottom of the local relief, the
curvature is low and convex, the slope is regular, near to constant (not too low or too high), the azimuth is
mainly constant and the wedge is typical, too (less than 180º).
All of the listed variables were classified to particular range according the typical talus zone description.
With classifying were produced binary raster sets. They were then overlaid by logical operations “AND” of
Boolean algebra. The areas that survived operations with all of variables considered as talus cones (Fig. 4f).
As the talus cones areas were produced for four sectors, the final results were mosaiced to a final potential
of talus cones (Fig 5).
a
b
c
d
e
f
Figure 4 Variables of visibility simulation (a), curvature (b), slope (c), variable based on a local window
for the south sector (d), aspect for the south sector (e) and potential talus cones for the south sector (f).
More blue (than red) areas mean lower relative relief (as presented with variable a). Presented is the area
around Mojstrana with dimensions of 7.9 km by 5.8 km.
3. RESULTS
The preliminary results of calculations are promising. Map algebra operations of sectorial parameters
yielded the best results. Both in the wider valley of Sava Dolinka river, and also in the tributary valleys
there are contiguous patches of classed that can be considered as talus slopes (Fig 5). The vast majority of
the talus area has been categorized by the method to the correct category. The misclassifications are
relatively rare, and they appear rather as single points, than in extensive patches. It seems that later they can
be removed applying a simple filter.
Figure 5 A final potential of talus cones mosaiced for four sectors. Yellow is talus cones for the south
sector, blue for north, green for east and pink for west. Presented is the area around Mojstrana with
dimensions of 7.9 km by 5.8 km.
4. DISCUSSION
Although we present here the results of a relative small test area, the method works also in larger extents as
well. On the other hand the processing of larger area needs more preparation time of the operator, since
misclassifications may occur. Especially if the various lithological classes are present in the area to be
processed, the results cannot remain always stable, because the modelling may turn to be selective in the
area of one lithology, and in the area of another rock type the classification selects to many or too few
points. This behaviour should be eliminated in operational environment.
5. CONCLUSIONS
Sectorial summation and statistical analysis of slope angle histograms seem to be feasible techniques to
enhance even smaller talus cones. The outlined extent of the cone then can be analyzed for slope angles and
can be compared to angle of repose of the (stable) talus cones made of the same material in the vicinity.
This technique is found to be extendable partly to scree slopes with some restrictions. However, in some
cases yet manual control is needed on the parameter set to achieve the optimal detection. Furthermore,
some sorts of ranking filtering seem to be also need later as a final step of the processing.
The automated extraction of such features will also be used for visualization of geomorphic elements of the
landscape. Combination of numerical extraction with sophisticated visualization techniques is not only
important in the geomorphic evaluation, but also in the communication of natural hazards to increase public
awareness.
ACKNOWLEDGEMENTS
In this study a portion of the Digital Terrain Model of Slovenia (© 2005, Surveying and Mapping Authority
of Republic of Slovenia) have been used. The techniques for the detection of features have been carried out
in collaboration with the Mars related research project TMIS.plus.II (as part of “HRSC on Mars Express”)
which is funded by the Austrian Research Promotion Agency in the frame of the ASAP program and the
Scientific Research Centre of the Slovenian Academy of Sciences and Arts.
REFERENCES
ANSELIN, L. (2005): Mapping and geovisualization (ACE 592SA – Spatial Analysis). University of Illinois, Urbana–
Campaign.
CHANG, Y.C., SINHA, G. (2007): A visual basic program for ridge axis picking on DEM data using the profilerecognition and polygon-breaking algorithm. Computers and Geosciences 33 (2), 229-237.
CHOU, Y.-H. (1997): Exploring Spatial Analysis in Geographic Information Systems. OnWord Press, Santa Fe.
ESRI (1997): Understanding GIS, The ARC/INFO method. John Wiley & Sons, Redlands.
HUTCHINSON, M. F., GALLANT, J. C. (2000): Digital Elevation Models and representation of terrain shape. In: WILSON J.
P., GALLANT J. C. (eds.): Terrain Analysis: Principles and application. John Wiley & Sons, USA, pp. 29–50.
KANEVSKI, M., MAIGNAN, M. (2004): Analysis and modelling of spatial environmental data. EPFL Press, Lausanne.
PIKE, R. J. (2000): Geomorphometry – diversity in quantitative surface analysis. Progress in Physical Geography,
24(1), 1–20.
PODOBNIKAR, T. (2005): Production of integrated digital terrain model from multiple datasets of different quality.
International Journal of Geographical Information Science 19 (1): 69-89.
PODOBNIKAR, T. (2008): Enhancing terrain features for improved cartographic visualization. International conference
on cartography and GIS, Borovets, Bulgaria.
RONCAT, A., WAGNER, W., MELZER, TH., ULLRICH, A. (2007): Echo detection and localization in full-waveform
airborne laser scanner data using the average difference function estimator. Photogr. J. Finland (in press).
SZÉKELY, B., KARÁTSON, D. (2004): DEM-based morphometry as a tool for reconstructing primary volcanic landforms:
examples from the Börzsöny Mountains, Hungary. Geomorphology 63, 25–37.
SZÉKELY, B., REINECKER, J., DUNKL, I., FRISCH, W., KUHLEMANN, J. (2002): Neotectonic movements and their
geomorphic response as reflected in surface parameters and stress patterns in the Eastern Alps. EGU Stephan
Mueller Special Publication Series, 3:149-166.
TIMÁR, G., SÜMEGI, P., HORVÁTH, F. (2005): Late Quaternary dynamics of the Tisza River: Evidence of climatic and
tectonic controls. Tectonophysics, 410: 97-110.
TOMLIN, C.D. (1990): Geographic information systems and cartographic modeling. Prentice Hall, Englewood Cliffs,
New Jersey.
TOMLINSON, R.F. (2003): Thinking about GIS: Geographic Information System planning for managers, ESRI Press,
Redlands.
WAGNER, W., ULLRICH, A, DUČIĆ, V, MELZER, TH., STUDNICKA, N. (2006): Gaussian decomposition and calibration of a
novel small-footprint full-waveform digitizing airborne laser scanner. ISPRS J. Photogr. Rem. Sens. 60:100112.
ZÁMOLYI, A. (2006): Nagy és kis reliefenergiájú digitális domborzati modellek esettanulmánya (Case studies of digital
elevation models with high and low relief energy; in Hungarian with English abstract). Geodézia és
Kartográfia 58(11):24-30.
ZÁMOLYI, A., SZÉKELY, B., TIMÁR, G., DRAGANITS, E. (2007): Quantitative river channel analysis based on
georeferenced historical maps – documenting vertical movements in the Little Hungarian Plain. Geophysical
Research Abstracts 9: 06624.
BIOGRAPHY OF THE AUTHORS
Balázs SZÉKELY, Dr. rer. nat. (PhD)
Balázs has received his MSc in Geophysics and MSc in Astronomy at Eötvös University, Budapest,
Hungary (1989). He joined to the Space Research Group at the Department of Geophysics at Eötvös. He
studied solar-terrestrial relationships, worked in projects of crop yield estimation based on satellite
imagery; later he started to analyse of digital elevation data. In 1998 he joined to a project at the Institute of
Geology and Palaeontology of the University of Tübingen (Germany), where he carried out morphometric
analysis of the Eastern Alps. After having received his PhD in Geology in Tübingen (2001), he continued
to apply Surface Processes Modelling. In 2004 he returned to the Space Research Group at Eötvös to study
the neotectonics of the Pannonian basin. Recently he is focussing on tectonic geomorphology and interested
in archaeometry. He currently analyses LiDAR DTMs at the Institute of Photogrammetry and Remote
Sensing, Vienna University of Technology (Austria).
Tomaž PODOBNIKAR, PhD
Tomaž holds a BSc in geographical information systems and its applications in environment and
archaeology, MSc in Monte Carlo methods and its applications in geographical information systems, PhD
in digital terrain modelling from various data sources of different quality, all the University of Ljubljana.
He has authored several papers on GIS, DTM, and its applications in quality assessment, archaeology and
environment and numerous applications like production a DTM of Slovenia that is currently widely
available from the government. He was research fellow at the Technical University Delft (The
Netherlands), Vienna University of Technology (Austria) and University of Franche-Comté (France). He is
currently research on geomorphological analysis at Institute of Photogrammetry and Remote Sensing,
Vienna University of Technology. He is involved to Scientific Research Centre of the Slovenian Academy
of Sciences and Arts and to University of Nova Gorica as well.