CHAPTER 5

Digital terrain modeling and glacier

topographic characterization
Duncan J. Quincey, Michael P. Bishop, Andreas Kääb, Etienne Berthier, Boris Flach,
Tobias Bolch, Manfred Buchroithner, Ulrich Kamp, Siri Jodha S. Khalsa, Thierry Toutin,
Umesh K. Haritashya, Adina Racoviteanu, John F. Shroder, and
Bruce H. Raup

ABSTRACT 5.1 INTRODUCTION

The Earth’s topography results from dynamic inter-
actions involving climate, tectonics, and surface
processes. In this chapter our main interest is in
describing and illustrating how satellite-derived
DEMs (and other DEMs) can be used to derive
information about glacier dynamical changes.
Along with other data that document changes in
glacier area, these approaches can provide useful
measurements of, or constraints on glacier volume
balance and—with a little more uncertainty related
to the density of lost or gained volume—mass bal-
ance. Topics covered include: basics on DEM gen-
eration using stereo image data (whether airborne
or spaceborne), the use of ground control points
and available software packages, postprocessing,
and DEM dataset fusion; DEM uncertainties and
errors, including random errors and biases; various
glacier applications including derivation of relevant
geomorphometric parameters and modeling of
topographic controls on radiation fields; and the
important matters of glacier mapping, elevation
change, and mass balance assessment. Altimetric
data are increasingly important in glacier studies,
yet challenges remain with availability of high-qual-
ity data, the current lack of standardization for
methods for acquiring, processing, and representing
digital elevation data, and the identification and
quantification of DEM error and uncertainty.
The Earth’s topography is the result of dynamic
interactions involving climate, tectonics, and sur-
face processes (Molnar and England 1990, Bishop
et al. 2003). Numerous feedback mechanisms oper-
ate over unique spatiotemporal scales that govern
hypsometry and terrain parameters. Similarly, the
topography governs a variety of physical param-
eters and processes including stress fields, erosion
and mass wasting, precipitation, sediment deposi-
tion, surface and groundwater flow direction and
ponding, surface energy budget, and glaciation in a
complex topographic feedback loop (Roe et al.
2002, Reiners et al. 2003, Arnold et al. 2006a).
Scientists have long recognized the significance of
topographic parameters for studying and modeling
process mechanics, mapping the landscape, and
detecting spatiotemporal change. Consequently,
digital terrain modeling and the generation and
analysis of terrain surface representations,
commonly referred to as digital elevation models
(DEMs) and geomorphometry, have become an
important research topic for studying and solving
problems in hydrology, geomorphology, and
glaciology (Wilson and Gallant 2000).
Glaciological research indicates that glaciers in
most areas around the world are receding (Barry
2006, Gardner et al. 2013). Internationally, scien-
tists are attempting to inventory, monitor, and
114 Digital terrain modeling and glacier topographic characterization
understand current ice mass distributions and vol-
ume, glacier–climate interactions, glacier sensitivity
to climate change, as well as assess the impact of
glacier change related to water resources, natural
hazards, and sea level fluctuations (Meier 1984,
Dyurgerov and Meier 2000, Haeberli et al. 2000,
Kaser 2001, Arendt et al. 2002, Huggel et al.
2002, Meier et al. 2003, Bishop et al. 2004, Kargel
et al. 2005). With many high-altitude glacial lakes
now forming in alpine areas (Huggel et al. 2002,
Gardelle et al. 2011), with the threat of devastating
landslides ever present and perhaps shifting (Kargel
et al. 2010), with human habitation of risky places
near glaciers rising (Kargel et al. 2011), and con-
struction of valuable infrastructure increasing in
alpine environments that never before had much
human presence (Kargel et al. 2012), accurate
DEM data are crucial for hazard assessments. This
is especially true for glacier hazard-prone regions
where field access is difficult due to severe terrain,
political sensitivity, or financial constraints.
Glaciological research is increasingly utilizing
remote-sensing studies and topographic informa-
tion generated from airborne and satellite imaging,
radar and LiDAR (light detection and ranging)
systems, and GNSS (global navigation satellite sys-
tem) surveying. Furthermore, DEMs are required
for image orthorectification and radiometric cali-
bration (Bishop et al. 2004), surface energy balance
studies (Arnold et al. 2006a), debris-covered glacier
mapping (Paul et al. 2004b), glacier ice volume
loss and mass balance estimates (Berthier et al.
2007), and equilibrium line altitude estimation
(Furbish and Andrews 1984, Leonard and Fountain
2003).
The mere availability of satellite imagery and
DEMs, however, does not necessarily equate to
accurate thematic information production and
glacier parameter estimation (Raup et al. 2007).
Consequently, researchers must account for a
variety of issues in digital terrain modeling and
geomorphometric analysis. Digital terrain modeling
is a complex process that focuses on terrain surface
representation schemes, acquisition of source data,
terrain descriptors and sampling strategies, inter-
polation techniques and surface modeling, quality
control and accuracy assessment, data manage-
ment, and interpretation and applications (Wilson
and Gallant 2000, Li et al. 2005, Fisher and Tate
2006). Accurate glacier assessment using topo-
graphic information is frequently an issue of
DEM quality and methodological approach (Raup
et al. 2007).
The purpose of this chapter is to address the topic
of digital terrain modeling and geomorphometry
for glacier assessment and mapping. Specifically,
we consider numerous issues that must be effec-
tively dealt with in order to estimate selected glacier
parameters and delineate glacier boundaries. We
first provide background information on digital
terrain modeling to set the stage for DEM produc-
tion using satellite imagery and methodological
approaches. We then provide examples of analytical
techniques that can be used to study glaciers.
5.2 BACKGROUND
Digital elevation models are digital representations
of the Earth’s relief and are fundamental to many
applications. The majority of currently available
digital elevation datasets are the product of photo-
grammetric software, which exploits the technique
of stereoscopy using overlapping image data,
although increasing volumes of elevation data are
also derived from radar interferometry and from
laser altimetry. Additional elevation datasets are
available from digitized map sheet topography
and, although usually spatially limited, ground sur-
veys. As with many remote-sensing exercises, the
generation of digital elevation data represents a
complex process involving image acquisition, com-
putation and modeling, data management, and
analysis. The terrain itself is spatially complex,
and the need to represent it digitally and describe
it quantitatively has led to various mathematical
approaches for different scale and data constraints.
Examples include the analysis of systematic high
or low-frequency periodicity in altitude using
Fourier analysis, and the identification of self-
similar, spatially autocorrelated, and scale-
dependent topography using fractal geometry and
semivariograms.
Digital elevation data are normally represented in
one of four ways, including use of: (1) a regular
grid; (2) a triangulated irregular network (TIN);
(3) contour lines; or (4) point profiles. Regularly
gridded DEMs comprise square pixels of constant
spatial resolution (the effective ground dimension)
or constant latitude/longitude resolution, with each
pixel assigned an elevation value, usually in meters
and with respect to some datum (sea level or a
higher order representation of the geoid). They
are favored by geoscientists because they are
directly comparable with remote-sensing imagery
of equal spatial resolution, simple to analyze statis-

tically, computationally easy to represent, and can
be saved in a range of formats (e.g., GeoTIFF,
HDF). Conversely, they are poor at representing
abrupt changes in elevation and heterogeneous
topography, particularly at coarse resolutions, can
generate large file sizes at fine resolution, and may
have a large amount of data redundancy across flat
areas. TINs represent topography using a mesh of
adjacent, nonoverlapping triangles with each vertex
assigned an elevation value. The TIN model stores
topological information allowing proximity and
adjacency analyses across a range of spatial scales.
TINs usually require fewer data points than a DEM
grid and are particularly useful for accurately repre-
senting extreme topographic features (e.g., channels
and ridges) as well as for surface modeling (e.g., the
calculation of slope, aspect, surface area and length,
etc.). Contour maps, some digitized from paper
topographic maps, provide more generalized repre-
sentations of the terrain, but they are not so com-
monly used in glacial applications given the
availability of contemporary DEMs for the entire
globe; however, for the derivation of ice thickness
changes over multiple decades, such digitized maps
can be crucial. Lastly, point profile data normally
represent elevations collected by laser altimetry
(e.g., GLAS on board the ICESat), which illumi-
nate a circular sampling area (footprint), the size
and spacing of which is generally dependent on the
altitude of the sensor. These data therefore tend to
provide sparse coverage when acquired by satellite,
and repeat pass data are not necessarily spatially
coincident.
The terms DEM, DSM (digital surface model),
and DTM (digital terrain model), are often used
interchangeably to describe any digital representa-
tion of the Earth’s topography, but differ funda-
mentally in their inclusion or exclusion of surface
cover in the data. DSMs can be thought of as
including all elevation data recorded regardless of
surficial cover; thus they represent tree canopy
height across a forested area and building heights
in the urban environment, for example. DTMs
‘‘strip out’’ such surface features and attempt to
represent only the solid Earth, which can be a chal-
lenging task (sometimes requiring large amounts of
interpolation) where dense ground cover(s) exist.
The term DEM is used more generally to describe
an elevation dataset without specifying the eleva-
tion reference (e.g., solid Earth, surface cover,
subsurface topography). Many satellite-derived ele-
vation datasets are of insufficient vertical resolution
to require an explicit declaration of whether they
Background 115
include surface cover or not, and are therefore gen-
erally referred to simply as DEMs.
The use of elevation data derived from remotely
sensed imagery has a number of advantages over
alternative sources. First, the range of sensors avail-
able for producing DEMs offers flexibility in scale.
Many medium-resolution (10–30 m) satellite sen-
sors (e.g., ASTER, ERS-1/2) are able to provide
stereoscopic imagery for DEMs of wide areal cover-
age, while local to catchment-scale DEMs can be
derived from an increasing number of fine resolu-
tion (1–10 m) sensors (e.g., SPOT 5 HRS, Quick-
bird, GeoEye) or aerial imagery. Normally, the
scale of the topographic variability to be modeled
informs the selection of the source imagery (see
Section 5.3.1). Second, many glacierized areas of
the world are inaccessible for logistical or political
reasons, so for the inclusion of three-dimensional
data in global databases such as GLIMS, remotely
sensed elevation information provides the only
option. Third, as historical archives of these data
are accumulating, the opportunity to conduct
multitemporal (and thus dynamical) studies
becomes possible. Fourth, and finally, the increas-
ing availability of no-cost data derived from
remotely sensed sources opens up the discipline of
cryospheric remote sensing to all interested parties,
which can only ultimately advance scientific under-
standing at a greater pace.
In relation to this last point, however, it is impor-
tant that standards and procedures for DEM
generation and analysis are available for scientists
worldwide, an issue that is discussed later in this
chapter (p. 137). In brief, the utility of a DEM is
dependent upon a certain level of expertise from the
operator to ensure the correct scale of representa-
tion, accurate characterization of terrain param-
eters, and overall planimetric and vertical
accuracy. For example, a major difficulty in using
digital photogrammetry to generate elevation data
over glacier surfaces is that accurate DEMs can
only be derived if the glacier surface shows sufficient
topographic and/or radiometric heterogeneity to
correlate image pairs (Favey et al. 1999), which
otherwise can result in grossly inaccurate elevation
estimates over some large expanses of bare ice and/
or snow. Image saturation or low dynamic numbers
in deeply shadowed areas also result in either null
values or erroneous elevations. Consequently, great
care is required to ensure that the data selected for
an application are appropriate, processing is carried
out with a high level of expertise, and errors in any
derived data are accurately reported, so that real
116 Digital terrain modeling and glacier topographic characterization

geophysical patterns and features can be differen-
tiated from image and processing artifacts.
5.3 DIGITAL ELEVATION
MODEL GENERATION
5.3.1 Source data
Topographic maps exist for every country, and
these can be utilized to generate a DEM, provided
the map is of sufficient quality for the intended
application. In developing countries, issues of
cartographic scale, map coverage, quality, and
availability may be problematic. In some countries
the maps are military secrets, or have been degraded
in quality given the scale, contour interval, or intro-
duction of error (e.g., Soviet topographic 1:50,000
map series of Afghanistan). If suitable maps exist
DEMs can be generated by cartographic digitiza-
tion using techniques such as line-following digiti-
zation or raster scanning.
High-resolution aerial and satellite images are
also effective ways to generate and update topo-
graphic information. Digital elevation models can
be derived using stereo (overlapping) pairs and
photogrammetric techniques, which depend on
knowledge of the exact image and terrain geom-
etries at the time of acquisition. Numerous investi-
gators have reported on the generation of DEMs
from aerial photography and a multitude of
satellite-imaging systems including ASTER, IRS,
and SPOT for glaciological applications (Kääb
2002, Berthier and Toutin 2008). Although satellite
imagery is generally available, aerial photography is
more difficult to acquire for many glacierized
regions of the world (e.g., Karakoram) because of
political sensitivities and poor archiving, and some
air photos are expensive (e.g., Alaska), which can
limit historical analyses. However, the demilitariza-
tion of Corona satellite imagery provides historical
data for many regions of the world at relatively fine
(<10 m) spatial resolution; Corona also provides
stereo coverage facilitating photogrammetric
analyses (Surazakov et al. 2007, Bolch et al. 2008).
For cryospheric applications, and especially
useful for three-dimensional information, ASTER
is one of the most appropriate and accessible sen-
sors available for several reasons. First, and most
importantly, the coupled nadir and backward-
looking sensor systems in the near-infrared enables
fine-resolution and routine along-track stereoscopic
vision (Fig. 5.1). Second, the spatial resolution of
these stereoscopic data (15 m) provides the perfect
balance between image detail and wide swath width
(60 km) that is required for regional-scale glacier

Figure 5.1. Imaging geometry of the ASTER sensor, on board the Terra satellite, launched in December 1999. The
satellite is in a 705 km, Sun-synchronous orbit that results in a 16-day revisit period for any given location on Earth.
The ASTER sensor is equipped with two bands in the near-infrared: one at nadir (3N) and one backward looking
(3B). It takes the sensor approximately one minute to cover the same area of ground with both sensors, thus yielding
a stereo pair of 60 km swath from which DEMs can be extracted (adapted from Hirano et al., 2003).

studies. Third, the adjustable sensor gain settings
provide increased contrast over bare ice and snow,
which is critical for the derivation of accurate
elevation data using photogrammetric techniques.
Fourth, the relatively short revisit period of the
sensor (16 days for nadir viewing or 2 days for
off-nadir pointing) provides high-intensity coverage
in cases of natural disaster emergencies or other
special needs; in more routine cases, the short revisit
time provides many chances to acquire a cloud-free
image of an area of interest, even in notoriously
cloudy areas. ASTER data are routinely used to
provide DEMs for the derivation of three-
dimensional glacier parameters for all glacierized
regions, and are available at no cost to regional
centers involved in the GLIMS project. However,
the actual performance of ASTER has resulted in
far less frequent data acquisitions and often sub-
optimal sensor gain settings than the observation
plan calls for, thus reducing the system’s value from
what it could be.
Images acquired by synthetic aperture radar
(SAR) contain information relating to both the
magnitude and phase of the backscatter signal, both
of which are useful for deriving DEMs. Radargram-
metry exploits the magnitude element of the return
signal. Similar to photogrammetry, radargram-
metry makes use of two spatially separated back-
scatter images to form a stereo model of the terrain
(Toutin et al. 2013), with the exact image geometry
being supplied by increasingly accurate orbital data
records (Scharroo and Visser 1998). Interferometric
SAR (InSAR) makes use of the phase element of
radar return. The phase difference between two
independent return signals is very sensitive to topo-
graphic variations (as well as any surface displace-
ments between image acquisitions) and can be
‘‘unwrapped’’ to derive a digital terrain surface,
often with very high accuracy. DEMs over glacier-
ized terrain have been successfully derived using a
range of SAR image sources including ERS-1/2
(Eldhuset et al. 2003), Radarsat (Peng et al.
2005), and sensors on board fixed wing aircraft
(Muskett et al. 2003). The preceding chapter by
Kääb et al. has a detailed treatment of various
methodologies and glacier applications using radar
data.
Airborne laser scanning systems are an important
operational tool for generating high-resolution
topographic information. Numerous researchers
have utilized and evaluated DEMs generated from
LiDAR data for glaciological applications (Rees
and Arnold 2007). Pulses of electromagnetic energy
Digital elevation model generation 117
interact with surface materials to regulate the inten-
sity of the returning signal. A LiDAR system pro-
duces data that may be characterized as a 3D point
cloud, which must be processed to remove erron-
eous measurements relating to reflectance from
clouds, smoke, or birds, for example (Arnold et
al. 2006b). Consequently, filtering, classification,
and modeling are required to generate a DEM
from laser-based measurements. The high quality
of LiDAR-based DEMs permits improved geo-
morphometric characterization of the terrain and
glacier surfaces, which translates into improved
assessment and modeling of a variety of
processes.
Finally, Global Navigation Satellite Systems
(GNSS) can be used for direct measurement of
the Earth’s surface topography, and are increas-
ingly replacing the use of traditional theodolites
and total stations. Earth scientists are routinely
utilizing GPS receivers on a variety of platforms
to collect point and profile measurements for ter-
rain and glacier surface representation (Miller and
Pelto 2003). Differential GPS (dGPS) refers to cal-
culation of an accurate roving receiver position with
respect to a reference station and can yield accura-
cies as fine as 10 mm when the differential phase of
multiple satellite signals is exploited. This method is
particularly relevant for the collection of ground
control data as input to DEM generation.
5.3.2 Aerial and satellite
image stereoscopy
The process of deriving elevation data by photo-
grammetric techniques is well established and
widely documented (Kääb et al. 1997, Kääb and
Funk 1999). Deriving elevation data from stereo
photographs relies on recreating the geometry
between sensor and terrain at the exact time of
image acquisition. The user supplies approximate
data relating to the aircraft location and the average
terrain height above sea level in addition to accurate
statistics on the camera geometry and image dimen-
sions. The software applies least-squares regression
to reduce errors in the approximations and arrives
at a stereo model that realistically represents the
real-life situation at the time of image acquisition.
The least-squares adjustment algorithm is used to
estimate the unknown parameters associated with a
solution while also minimizing error within the
solution. This is regarded as one of the most accu-
rate techniques to:
118 Digital terrain modeling and glacier topographic characterization

Figure 5.2. A typical stereo photogrammetric model (or block), illustrating the monotemporal generation of height
information from overlapping images (imagery of time 1 only). Where multitemporal imagery is available (imagery of
time 1 and imagery of time 2), calculations of surface elevation change become possible. By processing multi-
temporal images in a single block, potential errors, particularly those relating to external ground control, can be
minimized.

. estimate or adjust the values associated with
exterior orientation (sensor positioning and
view direction at time of image acquisition);
. estimate the X, Y, and Z coordinates associated
with tie points;
. estimate or adjust the values associated with inter-
ior orientation (sensor geometry, including focal
length); and
. minimize and distribute data error—introduced
by ground control point (GCP) coordinates, tie
point locations, and camera information—
through the network of observations
(list modified from Leica Helava 2004).
The least-squares algorithm processes the data
iteratively until the errors associated with the input
data are minimized. These errors are measured in
terms of residuals (the degree to which the observa-
tions deviate from the functional model). Residual
values (the distance between calculated statistics
and user input values) are summarized by the root
mean square error (RMSE), which should ideally be
less than twice the pixel size of the imagery for the
resultant orthoimages and elevation information to
be considered accurate when compared with the
actual terrain properties.
Once the error is within thresholds considered
satisfactory by the user, the DEM is extracted by
the generation of spatial rays from the first image
that intersect in three dimensions with the corre-
sponding points on the second image (identified
by image-matching algorithms), thus giving a
height value for each modeled pixel (Fig. 5.2). Ele-
vation is determined by measuring shifts in the x-
direction (x parallax) of the rectified images. The
exact process followed in DEM generation may
differ slightly depending on the software employed
(see Section 5.3.4) but typically follows a series of
key stages (Fig. 5.3).
5.3.3 Ground control points
Ground control data may be derived by traditional
surveying techniques (such as triangulation, tri-
lateration, traversing, and leveling), contemporary
survey techniques (such as dGPS measurement and
 Digital elevation model generation 119

Figure 5.3. The major steps required for the extraction of DEMs from satellite imagery. Exact processing sequences
vary according to the software used, with some offering total user control over all sensor/model parameters and
processing algorithms (white box), and others offering no control over internal implementation (black box). In both
cases, users should be careful to understand associated uncertainties in derived elevation data.

laser ranging), and/or the identification of clearly
defined features (e.g., mountain peaks and road
intersections) on large-scale map sheets. They
may give information on horizontal positioning,
height, or both. Collectively, ground control data
inform the stereo model of the spatial position and
orientation of a photograph or satellite image rela-
tive to the ground at the time of exposure (or scan-
ning), and their quality is important to the accuracy
of derived elevation data. As a minimum, the accu-
racy and precision of any ground control data
should be twice that of the expected DEM to avoid
adversely affecting the derived data. Research has
shown that the accuracy, number, and distribution
of ground control points used in the iterative least-
squares adjustment to refine the geometric model
can also impact DEM accuracy significantly
(Toutin 2008). While three GCPs are theoretically
sufficient to compute a stereo model from frame
imagery (Gonçalves and Oliveira 2004), the mini-
mum number for other sensors depends on the
sensor model employed and the number of
unknowns in it. A larger number than the mini-
mally required is usually employed to ensure redun-
dancy in least-squares adjustment, to reduce the
impact of map and plotting errors, and to facilitate
post-derivation accuracy tests (Toutin 2002). On
the other hand, where GCPs are known to have
low accuracy it is beneficial to limit the number used
in the adjustment to avoid errors propagating
through to the output DEM.
In addition to the quality of any ground control
data used in a stereo model, the theoretical accuracy
of a DEM is also dependent on the base-to-height
ratio of the system and the quality of the image
matching used to refine the stereo model. If no
ground control data are available, only relative
DEMs may be derived (where horizontal and ver-
tical values are only known relative to an arbitrary
datum). Their accuracy is thus dependent only on
sensor characteristics and the success of image
matching. Further parameters that can influence
the accuracy of any derived elevation data include
the quality of geometric and radiometric image cali-
brations (which also impact matching success and
ground control point identification) and the quality
of sensor ephemeris and attitude data that is entered
into the stereo model. Check points, which are most
often surplus ground control data, can be used to
quantify DEM errors but, similarly, the quality of
accuracy assessment is only as good as that of the
checkpoints used.
120 Digital terrain modeling and glacier topographic characterization
5.3.4 Software packages
In addition to the software used by NASA and the
Japanese Aerospace Exploration Agency (JAXA)
for the operational generation of ASTER DEMs,
there are four university or private software
packages, as well as five commercial off-the-shelf
(COTS) software modules for processing stereo
ASTER data to generate DEMs (Toutin 2008).
We now give a brief description of some of the main
software packages available.
The Geomatica 2 OrthoEngine SE of PCI Geo-
matics (www.pcigeomatics.com), adapted to ASTER
stereo data, was developed in 1999 under USGS
contract with the collaboration of the Canada Cen-
tre for Remote Sensing, Natural Resources Canada,
for mathematical, 3D modeling, and algorithmic
aspects. It may be the most-used software for
performing 3D sensor orientation and DEM and
orthoimage generation. Both relative DEMs with-
out GCPs and absolute DEMs with GCPs can be
generated while an existing DEM can be added in
the processing. The accuracies obtained in different
research studies were similar to those obtained at
USGS: horizontal and vertical accuracy of 15 m
and 20 m (1) respectively, depending on GCPs,
study site, and relief.
The LPS photogrammetric software suite from
ERDAS (http://www.erdas.com) includes ASTER-
specific data import, radiometric correction, and
sensor model functions. It uses a feature-based,
automatic terrain-adaptive, hierarchical image-
matching scheme for automatic DEM extraction
and can output grid, TIN, and point cloud data.
Of the software described here, LPS provides the
most comprehensive control over the DEM extrac-
tion and editing process. A number of automatic
spike and blunder detection algorithms aid the
detection of local extraction errors, and automatic
blending and smoothing around the perimeter of
area edits leads to seamless data continuation
through interpolated or replaced values. The most
promising of these algorithms fuses independent
datasets to achieve a complete DEM; in one
approach, Crippen (2010 and unpublished results)
used parallax–elevation cross-correlation as a
means of detecting artifacts in the ASTER DEM
so as to determine where replacement with other
data (e.g., from SRTM) should be done.
The AsterDTM module, developed by SulSoft,
the exclusive ENVI distributor in Brazil, was added
to ENVI as an add-on module in 2004 (http://
www.ittvis.com). The software can create DEMs
from ASTER Level 1A and 1B stereo pairs with
the aid of orbit/sensor modeling, quasi-epipolar
image generation, cross-correlation matching, and
automatic detection of water bodies. An existing
DEM can be added to fill in low-correlation areas.
The claimed accuracy by ENVI for a relative DEM
without GCPs is better than 20 m and for an
absolute DEM with GCPs 30 m in planimetry
and 15 m in elevation (all with 90% confidence
interval). These results were confirmed by end users
generating DEMs from Level 1A stereo pairs both
with and without GCPs, and from Level 1B stereo
pairs without GCPs.
The Desktop Mapping System Softcopy 2 ,
Version 5.0 is designed to run under the Microsoft 1
Windows 1 XP Professional operating system and
provides a complete two-dimensional and three-
dimensional photogrammetric mapping capability
using scanned aerial photographs or digital
images recorded by airborne or satellite sensor sys-
tems. Experiments with stereo ASTER data over
different mountainous study sites achieved RMS
errors of about 15–25 m (1), showing equivalent
results to those derived by USGS (see PCI).
DMS Softcopy 5.0 retains the speed, functionality,
and ease of use of earlier versions of the DMS
software, but has been streamlined to allow for
even faster, more efficient implementation of
operations.
SilcAst, produced by Sensor Information
Laboratory Corp. Japan (www.silc.co.jp) and exclu-
sively developed for ASTER, is written in IDLR6.1
and can be executed with IDL VM without an IDL
license. The main functions are digital elevation
extraction either from Level 1A and Level 1B
imagery, water body identification, orthorectifica-
tion, and Level 1B data generation, but the software
does not accept GCPs. ASTER 3D data products
are provided as Standard (fully automatic) with
potential miscalculated elevations, and High
(semi-automatic) with interactively corrected mis-
calculated elevations. In 2009, SILC with SilcAst
produced a high-quality global DEM (G-DEM1;
see Section 5.3.6) from ASTER data acquired
within an 83 latitudinal limit. Covering all the land
on Earth, it is freely available to all users in 1  1
grid tiles in latitude and longitude. While the
ASTER G-DEM1 contains 30 m grid spacing with
7 m accuracy (1), recent scientific studies demon-
strated some high-frequency errors (appearing
within tile grids as pits, bumps, ‘‘mole runs’’, and
other residuals and artifacts), requiring down-
sampling to 90 m spacing.

5.3.5 Postprocessing (interpolation and
smoothing)
DEM generation often requires the interpolation of
elevation values between available sample points,
particularly when contour maps or satellite imagery
are the source data. Consequently, various ap-
proaches for how to interpolate this calculated
(not measured) elevation in the DEM production
process have been developed. The choice of
approach depends mainly on relief within the
imagery, the type of information that will be
derived from the DEM, and the proposed applica-
tion (Rasemann et al. 2004). Interpolation accuracy
depends on interpolation algorithm type, surface
characteristics, and the distance between sample
points (Kubik and Botman 1976). There is a paucity
of quantitative evaluation of interpolation accuracy
over glacierized terrain and, of the few studies that
do exist, only a handful have employed ground
data to make the assessment (Cogley and Jung-
Rothenhausler 2004). At present, uncertainties in
DEM accuracy do exist depending on the choice
of the interpolation algorithm employed; therefore,
all qualitative and, in particular, quantitative results
from glaciological analyses that include DEMs
must be interpreted with caution.
Some of the interpolation methods that have
been used in glacier research include (but are not
limited to) Triangular Irregular Network (TIN;
Gratton et al. 1990); Inverse Distance Weighted
(IDW; Etzelmüller and Björnsson 2000); spline
(Mennis and Fountain 2001, Bolch et al. 2005);
and kriging (Burrough and McDonnell 1998).
Triangulation: since this method produces indi-
vidual triangles with constant exposition and aspect
rather than a continuous relief, the final DEM is not
differentiable. To ensure that information on cur-
vature can be extracted, the DEM has to be con-
verted to a grid. Problems occur at sharp edges,
deep valleys, and cliffs. Often, valleys and ridges
show a typical ‘‘terracing’’ effect. Such artifacts
can be eliminated by manually adding elevation
information, by using structural information with
elevation values (Heitzinger and Kager 1999,
Rickenbacher 1998), or by subdividing the triangles
(Schneider 1998).
Inverse Distance Weighted: a relatively simple algo-
rithm that weights the sample point value according
to its distance from the pixel of interest. The weight-
ing and number of sample points have a strong
Digital elevation model generation 121
influence on DEM quality. For example, a strong
weighting based on only few sample locations pro-
duces a rough surface in which the sample points
themselves are heavily emphasized.
Spline: this interpolation method applies a special
function that is defined step by step by piecewise
polynomials that are continuously differentiable.
Most common in DEM generation are low-order
cubic (r ¼ 3) and bicubic (r ¼ 2) splines. In general,
the surface of the final DEM is smooth and even.
However, the main problem is the occurrence of
overshoots (i.e., unnatural ‘‘holes’’ and ‘‘spikes’’).
To prevent such overshoots, a regularized spline-
with-tension algorithm is used (Mitasova and Mitas
1993). This method uses a function that specifies
how fast the collection of sample points decreases
with increasing distance thereby attenuating the
magnitude of overshoots.
Kriging: a geostatistical method, based on a
stochastic model, which predicts the elevation at a
specific location and is particularly used when a
trend (here: the spatial correlation of elevation
values) exists. One disadvantage of this method is
that adjustment of the semivariogram model to the
existing data structure requires experience.
Various studies have considered the relative
accuracy and appropriateness of each of the above
approaches. For example, Kamp et al. (2005) tested
three interpolation methods—IDW, spline, and
TIN—when generating a DEM from 1:50,000 topo-
graphical maps of Cerro Sillajhuay in the Andes of
Bolivia and Chile using Esri ArcGIS 3.2 software.
In this study, both IDW and spline methods pro-
duced very good results in high relief areas, but a
large number of artifacts in generally flat terrain.
The authors concluded that both algorithms had
problems handling larger distances between con-
tour lines. Instead, the most accurate DEM was
generated using the TIN method, although a first
raw DEM contained many artifacts—mainly over
flat terrain, caused by the triangles that are used by
the TIN method for interpolation.
Tests similar to those made by Kamp et al. (2005)
were described by Racoviteanu et al. (2007) for the
Cordillera Ampato (southern Peru). In this study,
the examination of RMSEz (vertical error) values
for DEMs derived from topographic data (1:50,000
maps constructed from 1955 aerial photography)
revealed that no interpolation method performed
122 Digital terrain modeling and glacier topographic characterization
perfectly. While the IDW DEM and TIN DEM
showed accuracies of only 21 m and 24 m, respec-
tively, the spline (here: TOPOGRID) DEM pro-
duced a vertical accuracy of 15 m,
outperforming the other available algorithms.
However, the authors noted that all three DEMs
produced ‘‘terracing’’ effects, an artifact of prefer-
ential sampling along the contour lines, with points
closer to the contour lines being interpolated using
the same elevation values. The terracing effect was
most severe using the IDW interpolator. Similar
terracing artifacts were described by Etzelmüller
and Björnsson (2000) using the same interpolator,
and by Mennis and Fountain (2001) using the
spline-with-tension interpolator. Wilson and Gallant
(2000) showed that such terracing artifacts affect
calculations of topographic characteristics such as
slope, aspect, and profile curvature.
DEMs constructed over largely featureless ter-
rain often fail to characterize subtle (or even
abrupt) changes in surface topography, which can
be rectified by the use of breaklines. Breaklines are
linear features that describe interruptions in surface
smoothness and are typically used to define chan-
nels or ridges, for example, where there is a clear
change in surface topography. However, the devel-
opment of real 3D breaklines is still a problem, and
Bolch and Schröder (2001) actually concluded that
the integration of simple breaklines does not sig-
nificantly improve the overall quality of the DEM.
As a solution, in their DEM generation process
Kamp et al. (2005) manually added some contour
lines using elevation information from stereoscopic
analysis of aerial photographs. Eventually, the TIN
DEM was converted into a grid-based DEM using
Esri ArcInfo 7.2 software, and two nearest neighbor
filters helped smooth the final model, albeit at the
expense of some topographic detail (e.g., the loss of
some edges, frost cliffs, and gorges). The authors
concluded that the final DEM was of sufficient
quality for macro and mesoscale geomorphological
analysis (e.g., of rock glaciers); however, processing
the DEM from contour lines was relatively time
consuming.
5.3.6 Data fusion
While small gaps in DEM data can be filled by
interpolation, large voids stemming from shadow-
ing, correlation failures, and other effects require a
different approach. It is sometimes desirable to
combine, or fuse, DEMs from different sources in
order to obtain the most complete coverage of an
area. As well as filling in any gaps in the respective
DEMs, such fusion can also help to identify any
severe vertical and horizontal errors in common
areas (Kääb et al. 2005). DEM-merging techniques
range from replacement of data (Kääb 2005a) and
weighted fusion, resulting in smooth transitions
between DEMs (Weidmann 2004), to merging in
support of DEM processing itself (e.g., DEM
approximation in order to geometrically constrain
stereo parallax matching or interferometric phase
unwrapping) (Honikel 2002). One successful
approach involves identification of artifacts in a
single-scene ASTER DEM and then infilling with
a global dataset such as SRTM or the ASTER G-
DEM. Artifact delineation is almost perfect (almost
no errors of omission or commission), so the infilled
product is usually of very high quality (Crippen
2010).
The ASTER G-DEM1 is the result of the fusion
of suitable individual ASTER-derived DEMs by
averaging all elevations in the vertical stack. While
DEM fusion using this approach is helpful for
deriving a spatially more complete and accurate
set of geomorphometric parameters, it should be
avoided or handled with care for DEM-based stu-
dies of glacier thickness changes. The number of
stereo images used to produce a given G-DEM1
grid point elevation varies from one to several tens.
Whereas the number of scenes used for each grid
point is known globally, G-DEM1 ancillary data do
not support knowledge of when the images were
acquired. The mean date is about 2004/2005, but
for a given locale it could be earlier or later by a few
years. Variance of the mean from 2004/2005 is less
than a year in most areas, but it can be much greater
for some places where few ASTER images have
been acquired. G-DEM1 carries some of the arti-
facts present in individual DEMs, and is generally
very noisy, often rendering it unsuitable for detailed
studies requiring high resolution, although for
studies of basin-scale hypsometry, G-DEM1 can
be convenient. A new G-DEM2, of similar con-
struction but lacking the severity of noise and arti-
facts of the earlier version, was released in October
2011. G-DEM2 benefits from 260,000 additional
ASTER scenes, and preliminary tests indicate
improved accuracy in both horizontal and vertical
dimensions, and a substantial reduction in the num-
ber of voids (Tachikawa et al. 2011). Still remaining
for the future is development of a new GDEM that
would allow better tracking of input images.

5.4 DEM ERROR AND UNCERTAINTY
5.4.1 Representation of DEM error
and uncertainty
Given the number of stages involved in generating a
DEM, errors are inevitable in both the vertical and
planimetric coordinates. In general, error can be
classified as: (1) gross errors that are associated with
equipment failure; (2) systematic error that repre-
sents the results of a deterministic system that can
be accounted for by using functional relationships;
and (3) random errors that occur within (or because
of ) the multitude of operational tasks that are
involved in computation. In this context, error is
typically defined as systematic and/or random var-
iations about a ‘‘true’’ reference value (Fisher and
Tate 2006). The magnitude of vertical error can be
an important consideration, as changes in glacier
surface altitude over shorter time frames may
become undetectable, depending upon the nature
of the source data. Consequently, the usefulness
of a DEM for estimating mass balance must be
carefully evaluated (Bamber and Payne 2004).
A common descriptor of error is root mean
square error (RMSE):
sP ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P 2
i j ðzi; j  zi; j;r Þ
RMSE ¼
n and PCI Orthoengine all provide a suite of post-
where zi; j is the elevation at location (i; j ) in the
DEM, and zi; j;r is a reference elevation at the same
location. The metric has been used when comparing
elevations off-glacier in a variety of glaciological
studies, although it does not describe very well
the statistical distribution of vertical error (Fisher
and Tate 2006). It has, however, become a standard
measure of map accuracy. Other researchers have
used additional error metrics such as mean error
(ME) and standard deviation (S) (Cuartero et al.
2004):
PP
ðzi; j  zi; j;r Þ
ME ¼
n temporal DEMs are required for the measurement
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
PP
½ðzi; j  zi; j;r Þ  ME 2
S¼
n1
Systematic under or overestimation (biases) are
depicted as positive or negative ME values.
Such global summary (single-value) metrics are
quick to calculate and easy to report, but they fail to
characterize the spatial pattern of error, which is
critical to environmental applications. DEM errors
can vary spatially, and research indicates that the
DEM error and uncertainty 123
spatial structure of error in DEMs remains poorly
understood (Carlisle 2005). Numerous researchers
have attempted to study and describe the nature of
spatial error and have found that patterns of error
can be anisotropic, scale dependent, spatially vari-
able, and spatially correlated, but most importantly
DEM error is closely related to the complexity and
characteristics of the terrain (Kyriakidis et al. 1999).
Work in alpine environments often reveals patterns
of error over mountain terrain and glacier surfaces
that are highly variable in spatial extent and loca-
tion, and are related to acquisition of source data,
spatial interpolation, and terrain geometry.
5.4.2 Type and origin of errors
Once a DEM has been generated by determining
the parallax between pixels matched by cross-
correlation there is always a need for postprocessing
due to mistakes in pixel matching, correlation fail-
ures, and, if present, the impact of clouds in the
scenes. Further, shadowing in regions of high relief,
low image contrast over clean snow and ice, and
self-similar surface features on glaciers will lead to
voids and artifacts in the resulting DEM. Some
software packages, such as SilcAst, will try to
remove such features automatically, with no user
ð5:1Þ control over the process. Alternatively, ENVI, LPS,
processing tools for manual editing. However,
Eckert et al. (2005) found that in some cases the
postprocessing tools provided by software packages
were not sufficient and further morphologically
based interpolation algorithms were required to
produce a suitable DEM. Such matching mistakes
and voids are relatively easy to identify and correct
for, but where inaccurate GCPs and poorly chosen
tie points are used, distortions, biases, and plani-
metric shifts may also be introduced into the
derived DEM, which can be difficult to model
without the use of a reference elevation dataset.
ð5:2Þ This is particularly problematic where multi-
of elevation changes (see Section 5.5.5).
ð5:3Þ Two types of error affect the measurement of
glacier elevation changes: horizontal (planimetric)
and elevation (altimetric). Planimetric error refers
to horizontal shift of often correct elevation values
to erroneous easting/northing locations. This can
be either a systematic or a variable shift, and can
be exhibited in one or more directions. Such shifts
are easily quantified by correlating corresponding
orthoimages and quantifying the orthophoto paral-
124 Digital terrain modeling and glacier topographic characterization
Figure 5.4. Scatterplot of mean difference altitude
generated from a SRTM DEM (2000) and an
ASTER-derived DEM (2004). Mean values were com-
puted off-glacier over the Nanga Parbat Massif in
Pakistan. Variations in viewing geometry and/or space-
craft orbital parameters result in a nonlinear bias that
significantly deviates from zero. Consequently, linear
and nonlinear systematic biases must be removed when
comparing DEMs generated from different sensor prod-
ucts.
lax (Li et al. 1996, Georgopoulos and Skarlatos
2003), since there should be no relief displacement
in either image if the DEMs are accurate. Altimetric
error refers to systematic biases that may artificially
increase or decrease elevation in a given direction
across part of, or the whole, DEM, as well as more
locally variable slope or aspect-related biases (Fig.
5.4). The correction of both types of error usually
requires some form of mathematical modeling. The
most promising approach has focused on the use
of ordinary least-squares regression modeling to
create spatially nonstationary, spatially correlated
and heteroskedastic error surfaces from only a
small number of sample locations (Kyriakidis et
al. 1999, Carlisle 2005). These error surfaces can
then be used either to create higher accuracy
DEMs, or to inform analyses of error propagation
in the computation of surface derivatives (Oksanen
and Sarjakoski 2005). It is crucial that both plani-
metric and altimetric errors are identified, particu-
larly in the case of glacier monitoring as they can
lead to erroneous volume changes (Schiefer et al.
2007) or even change the sign of the resulting mass
balance calculation (Berthier et al. 2006).
Systematic errors are common in some of the
most widely available topographic datasets, making
it essential for users to make appropriate error
assessments, even if they believe biases are removed
prior to analysis. Particularly relevant in this
respect is the widely used continuous DEM pro-
duced by the Shuttle Radar Topography Mission,
which was flown in February 2000 and provides a
topography covering continental areas from 60 N
to 56 S (Rabus et al. 2003) with a 1 and 3 arcsec
spatial resolution (about 30 and 90 m, respectively).
It has yielded significant advances in the monitoring
of glacier wastage in Alaska and neighboring
Canada (Muskett et al. 2003, Larsen et al. 2007,
Schiefer et al. 2007, Berthier et al. 2010), Patagonia
(Rignot et al. 2003, Rivera et al. 2005), and central
Asia (Surazakov and Aizen 2006). However, since
its release, significant biases in SRTM data have
been identified by a number of authors, and a clear
examination of the bias still needs to be performed.
For example, Kääb (2005a) found a systematic
mean error of about 7 m for the Gruben area in
Switzerland. Still in the Alps, Berthier et al. (2006)
found a small mean error (<3 m) but detected a bias
that increased linearly with elevation. Such a bias
related to elevation is still not fully understood but
has since been confirmed for Patagonia (Möller et
al. 2007), the Himalayas (Berthier et al. 2007), and
the Canadian province of British Columbia (Schie-
fer et al. 2007). The latter authors used a piecewise
model to correct elevation bias (constant bias at low
elevation and linear bias above a threshold alti-
tude). The regional pattern of the biases in SRTM
data is also complex. Surazakov and Aizen (2006)
found a localized ice-free region with systematic
error in the SRTM data ranging from 20 to 12
m on a 30 km spatial scale. Berthier and Toutin
(2008) found a 20  20 km region where SRTM
data are systematically 5–10 m higher than SPOT5
or ICESat elevation data. Note here that, although
sensor platform orbits are too widely spaced for
mountain glacier elevation change mapping, ICE-
Sat elevation profiles have been shown to be a good
source of data to detect and understand errors in
SRTM data for ice-free regions where no elevation
changes are expected (Carabajal and Harding 2005,
Berthier and Toutin 2008).
5.5 GEOMORPHOMETRY
Glacier geomorphometry refers to the three-
dimensional topographic characteristics of a
glacier surface including its size, shape, hypsometry,
orientation, and position. Geomorphometric
analyses have historically pursued fine-resolution
digital elevation models to improve and refine land

surface parameters and objects (Drăgut et al. 2009).
However, with the increasing availability of detailed
DEMs it is becoming clear that the relationship
between DEM scale and the ability to extract geo-
morphometric information is not straightforward.
Indeed, for many applications, increased DEM
resolution only serves to introduce noise into the
analysis. Accordingly, a number of methods have
been applied to account for scale-related problems
through DEM generalization, although further
work along these lines is needed if accurate topo-
graphic characterization of glacierized areas is to be
achieved (Drăgut et al. 2009). Despite this current
imperfection, many quantitative analyses of land-
scape and landscape evolution remain heavily
dependent on the derivation of geomorphometric
data in their methods.
5.5.1 Geomorphometric land
surface parameters
When deriving geomorphometric information from
DEM data it should be noted that, for every order
of differentiation, the effective resolution of the
product declines and noise increases. The first
derivative (based on calculations between two
points) normally defines slope, whereas the second
derivative (based on calculations between three
points) normally defines curvature. For every order
of differentiation, small errors in the primary data-
set become exaggerated. Thus, a DEM having a
30 m posting can report the first and second deriva-
tive values still with 30 m postings, but each grid cell
of the derivatives includes information from neigh-
boring cells and so the effective resolution is
reduced for the derivatives, even though a value is
computed for each grid cell. Effectively this means
that derivative information for very small glaciers is
useless, especially if the derivatives, such as curva-
ture, are used to define lateral moraines, computed
drainages, or other landforms or hydrological units.
Thus, differential geomorphometric parameters
carry both great power to discriminate distinct
terrains and landforms, and great risk that noise
or systematic bias is transmitted into geomorpho-
metric mapping products.
Depending on the tolerance for noise or low
spatial resolution, geomorphometric analysis can
emphasize nondifferential parameters (based on
hypsometry), first-order differential analysis (slope,
slope aspect), or second-order differentials (curva-
ture, tangential curvature) (Fig. 5.5). Most off-the-
shelf image-processing software is capable of pro-
Geomorphometry 125
ducing such metrics using standard algorithms, with
the quality of the derived geomorphometric data
being largely dependent on the accuracy of the
input elevation data. In an applied context, land
surface parameters are a useful means of comparing
the topographic characteristics of glacierized areas
and, in addition, are increasingly being used in
combination with satellite imagery to derive more
robust approaches to accurate glacier delineation
(see Chapter 4, and Section 5.6 for further details).
They are also often employed in process modeling,
with elevation providing useful information on
gravity-driven processes. For example, land surface
parameters such as slope gradient can indicate the
gravitational potential available for geomorphic
work (i.e., erosion), and slope aspect the direction
of the work carried out (Smith and Pain 2009).
It is important to note that geomorphometric
data may vary dramatically between adjacent sub-
catchments and, less obviously, for the same glacier
or glacierized area through time. Changes in glacier
topography, for example, may be caused by a sus-
tained change in mass balance, or by a change in ice
dynamics (or both). Basic elevation data reflecting
these changes may be derived by taking a topo-
graphic profile along the flow line of the glacier
from multitemporal DEMs or by computing the
altitude–distance function, which effectively com-
pares every pixel representing the surface of the
glacier to a terminus position by plotting the aver-
age altitude in a distance bin. More complex eleva-
tion change data can be derived by hypsometric
analysis, essentially a frequency analysis, character-
izing the altitude–area distribution of a glacier
surface. However, the generation of more complex
data such as these is not always desirable; for
example, small changes in surface elevation may
be easily detected in simple multitemporal topo-
graphic profiles, but not so in a multitemporal
hypsometric analysis where large volumes of
(potentially noisy) data can mask subtle surface
changes. Whichever geomorphometric approach is
adopted, it is essential that systematic biases in the
DEM are removed prior to analysis and interpreta-
tion of the results (see Section 5.4).
5.5.2 Scale-dependent analysis
A range of interacting geomorphic processes shape
the glacial environment, leading to landforms that
are rarely strictly spatially delimited (Schneevoigt
and Schrott 2006). Such landforms are diverse in
size and shape, even within a single land cover
126 Digital terrain modeling and glacier topographic characterization

Figure 5.5. Geomorphometric analysis of the Batura Glacier, Hunza Valley, Pakistan: (a) SPOT HRVIR
panchromatic satellite imagery (acquired October 17, 2008), copyright CNES 2008/Distribution Spot Image.
(b) SPOT HRS-derived composite DEM (image dates 2002–2004), from which (c) slope, (d) aspect, (e) shaded
relief, and (f ) plan convexity, can all be calculated. Figure can also be viewed as Online Supplement 5.1.

classification, and microscale landforms may com-
pose local-scale landforms, which may themselves
compose catchment-scale landforms. In modeling
such dynamic and diverse environments, linear
process-form relations and those that are spatially
restrictive are therefore unrepresentative. There is
thus increasing focus on the hierarchical organiza-
tion of topography in studies on mountain geo-
morphology, and the subsequent adoption of
scale-dependent analyses. These can be simply
defined as methods that examine the variability of
the three-dimensional spatial structure of morpho-
logical features on multiple levels.
With the increasing availability of DEMs from
airborne and spaceborne platforms, automated
approaches have been developed to identify specific
geomorphometric features (e.g., glacier extent, loca-
tion and density of crevasses) based on their shape
and textural properties. One such method is to
measure surface roughness on multiple scales and
in multiple directions, by variogram analysis. The
variogram is effectively an index of dissimilarity
between pairs of points that are separated by a
predefined vector and within a specified search win-
dow. Multiple variograms can be used to character-
ize terrain morphological features, such as debris
flow deposits, channels, and rockfall deposits, and
to calibrate secondary morphological indices such
as anisotropy, periodicity, and the spatial vari-
ability calculated at specific lags. These indices,
while providing useful quantitative landscape
characterization data in their own right, can then

be used as input into further, perhaps more complex
(e.g., pattern recognition) landscape classification
techniques (van Asselen and Seijmonsbergen 2006).
A further derivative of the variogram is the frac-
tal dimension (D), which provides a measure of
whether topographic surfaces and spectral reflec-
tances of land cover units are self-similar over a
range of spatial scales. Many landscape features
and other environmental data have been shown
to exhibit self-similarity, at least statistically (Good-
child and Mark 1987). The limited number of
studies that have evaluated the utility of fractals
for these purposes indicate that the majority of
geomorphic surfaces exhibit scale-dependent fractal
dimensions (i.e., limited to certain spatial ranges;
Bishop et al. 1998), with the range in scales often
determined by the controlling geophysical processes
that have shaped, or continue to shape, the land-
scape (Lathrop and Peterson 1992). Fractal dimen-
sions have also been employed to improve
classification accuracy (de Jong and Burrough,
1995) and to provide insight into operational-scale
and process–structure relationships. This approach
appears to have great potential in this respect,
although to investigate self-similarity in surface
roughness characteristics over the full range of
spatial scales requires the use of high-quality
DEMs, which may not always be available in
glacierized areas.
5.5.3 Topographic radiation modeling
Digital elevation models are fundamental to the
three-dimensional modeling of solar radiation var-
iation across a landscape. Insolation modeling can
either be point specific or area based. Point-specific
modeling considers the geometry of surface
orientation, the visible sky, and the effect of local
topography. Local topographical effects can be
determined using empirical relations, visual estima-
tions, or, where field data are logistically possible,
by upward-looking hemispherical photography.
Such calculations can give locally very accurate
results, but are clearly limited in spatial coverage.
Area-based insolation modeling relies solely on the
use of a DEM to calculate surface orientation and
shadowing across a subcatchment, for example, and
as a result has much greater application for under-
standing large-scale landscape processes. Along
these lines, a number of studies have made surface
irradiance and ablation gradient calculations to
identify locations most suitable for past and present
glaciation (Carr and Coleman 2007, Allen 1998) by
Geomorphometry 127
identifying site shelter and exposure, and to be
applied as input to energy balance melt models,
which are clearly fundamental to understanding
the relationship between glacier behavior and
climate.
Net radiation (also known as net flux) is defined
as the balance between incoming and outgoing
energy and can be computed as:
Qn ¼ Eð1  Þ þ L #s þ L #t þ L "
where E is total surface irradiance,  is surface
albedo, L #s is longwave sky radiation, L #t is long-
wave radiation from the surrounding terrain, and
L " is emitted longwave radiation. It is regulated by
numerous processes and can be highly variable
depending on atmospheric, topographic, and sur-
face biophysical conditions. Indeed, in mountain-
ous areas radiative fluxes are particularly complex,
fluctuating considerably in space and time because
of the effects of slope, aspect and the effective hor-
izon, and the input of reflected and emitted radia-
tion from surrounding slopes (Hock 2005). As such,
there is a movement towards the use of increasingly
fine temporal (hourly) and spatial (25 m) resolu-
tion modeling, with the latter again dependent on
the availability of suitably fine-resolution and
accurate digital elevation data. Where such data
do exist, studies have shown that the melt rate of
snow and ice and spatial and temporal patterns in
energy balance can be calculated accurately across a
glacier, albeit with dependence on accurate field and
meteorological data (Brock et al. 2000).
5.5.4 Altitude functions
Most glacier parameters vary as a function of
altitude, and these relationships can give insight
into glacier dynamics and processes that simple
two-dimensional analyses would not reveal. A
glacier’s most fundamental altitude parameter is
the equilibrium line altitude (ELA) because it
divides the glacier into accumulation and ablation
areas (Braithwaite and Raper 2010). This can be
simply estimated using multispectral satellite
imagery in combination with a DEM, by assuming
that the ELA is equal to the snow line position at
the end of the melting season (Bamber and Rivera
2007). Such information is fundamental to mass
balance calculations, which are useful indicators
of climatic variations where no actual ELA data
or better alternative proxies exist (Yadav et al.
2004). In the absence of calculated mass balance
data or transient end-of-season snow line data,
128 Digital terrain modeling and glacier topographic characterization
the topographic profile of a glacier can be used as a
relative proxy for the dynamic state of glaciers, with
those glaciers in rapid recession tending to exhibit
concave-up or flattened (planar) surface profiles in
the ablation zone, and those accumulating or main-
taining mass characterized by convex-up surface
profiles (transverse profiles especially) in the abla-
tion zone (Quincey et al. 2009).
The altitude–area profile (or the hypsometric
curve) for each glacier represents a distinct 3D
spatial signature that is the end result of thousands
or millions of years of erosion and reflects interac-
tion of the regional climate with subglacial bedrock
(Tangborn 1999). It has been used in many cases to
infer the state of geomorphic development within a
glacial system (Katsube and Oguchi 1999). Glacier
erosion processes are determined most significantly
by glacier velocity, ice thickness, subsurface geol-
ogy, and basal thermal conditions. Differential
erosion therefore occurs with altitude as each of
these parameters changes accordingly. Topographic
analyses of the Nanga Parbat Massif in Pakistan
revealed a spatially variable relief structure that
correlated to geomorphic events and dominant
surface processes (Bishop et al. 2001). Glaciation
was found to generate the greatest mesoscale relief
at high altitudes and warm-based glaciation was
found to reduce relief at intermediate altitudes.
More recently, numerical modeling has shown that
mountain heights can be explained by a combina-
tion of glacier erosion above the snow line and
isostatic uplift caused by erosional unloading, with
maximum elevations being constrained to an alti-
tude window just above the snow line (Egholm et al.
2009).
More generally, glacier hypsometry determines
how responsive glacier-wide mass balance is to
climatic perturbations (Furbish and Andrews
1984). For example, a rise in the snow line of
200 m in a given region (in response to local atmo-
spheric warming, for example) would have a much
greater impact on a low-gradient valley glacier than
a steeply sloping one, as in the former case a
greater proportion of the overall ice mass would
become subject to net annual mass loss. Thus, as
an example, if there were a rise in local atmospheric
temperature in the Hunza Valley of the Karakoram,
the mass balance of the Batura Glacier would be
affected to a greater extent than that of the
Ghulkin Glacier, and the mass balances of the Pasu
and Ghulmit Glaciers would change somewhere
between those two end members (Fig. 5.6). Simi-
larly, glacier clinometry (characterization of the
Figure 5.6. Hypsometric curves for the Batura,
Ghulkin, Ghulmit, and Pasu Glaciers located in the
Hunza Valley, Pakistan.
slope/altitude function) expresses glacier relief,
and is thus also useful for determining erosion rates
and mass balance fluctuations.
The relationship between altitude and glacier
velocity is not straightforward. Glaciers with low
relief tend to exhibit flow that is relatively uniform
across the altitudinal range, whereas those with
high relief (and are thus fed predominantly by snow
and avalanche material) tend to have very high flow
rates at altitude and much reduced flow towards
their termini. Altitude–velocity profiles can be use-
ful for proxy assessments of mass balance (Luck-
man et al. 2007), for identifying areas susceptible to
glacial lake development (Quincey et al. 2007), and
for determining predominant glacier flow regimes
(Copland et al. 2009). In the case of the latter,
generally flat transverse velocity profiles with a
rapid reduction in flow at the margins indicate
‘‘block flow’’, suggesting down-glacier movement
of ice en masse by basal sliding, whereas convex
velocity profiles that display greatest displacements
towards the centerline of the glacier indicate a
flow regime dominated by ice deformation. Large
seasonal variability in flow can further highlight the
importance of meltwater availability for a given
glacier, and may even help to identify different flow
mechanisms that exist at different altitudes within a
single glacial system (Fig. 5.7; Quincey et al. 2009).
5.5.5 Glacier elevation changes and
mass balance calculations
Stereo photogrammetry has been employed to cal-
culate glacier surface elevation changes for many
 Geomorphometry 129

Figure 5.7. Selected seasonal and annual centerline velocity profiles for the Baltoro Glacier, Pakistan Karakoram.
Date format is year, month, day. Note varying flow rates across summer and winter seasons, and varying patterns
with altitude. In particular, note the similarity of flow profiles in the lowermost 400 m of the glacier, and contrasting
profiles with greater altitude, which may be related to the availability of meltwater at the glacier bed.

years, using images acquired by both airborne and
satellite sensors. The most reliable source of DEMs
for volume change estimates is still aerial photos.
Given their very fine resolution, if precise GCPs are
available, DEMs with RMS accuracies the size of
approximately one pixel (few tens of centimeters)
can be obtained (Kääb 2002). Consequently, aerial
DEMs are still frequently employed to study vol-
ume changes at the scale of a single glacier and
detect systematic errors in cumulative mass balance
estimates obtained from the traditional (pit and
stake) glaciological method (Østrem and Haaken-
sen 1999). However, aerial photographs cover a
limited surface on the ground (generally one glacier)
and it is difficult to fly a plane in remote and high-
relief areas. Consequently, they cannot be regarded
as a means of obtaining regular coverage of glacier
topography at the regional scale. Given the large
footprint of satellite images (typically 60  60 km),
it is tempting to apply the geodetic method to space-
borne DEMs. The principle is the same as for aerial
photography: two sequential glacier surface models
are derived from pairs of stereoscopic images and
subtracted to yield a map of elevation changes.
Satellite-derived DEMs are less accurate than those
computed from aerial photographs so caution must
be taken when interpreting the signal and compre-
hensive error analysis is mandatory.
Changes in surface elevation can be measured
most accurately by processing all imagery within
a single block or, put more simply, using a single
mathematical model (Kääb 2000; Fig. 5.2). The
quality of derived surface elevation data using this
technique is mainly dependent on the interior
orientation of the stereo model (Baltsavias 1999)
rather than the accuracy of any ground control
used, which makes the technique particularly
appropriate for assessing remote terrain. Factors
such as scanned resolution of the images, accuracy
of fiducial mark locations, and accuracy of camera
calibration parameters are particularly important
in such analyses. Model GCPs introduce a second-
order error that is small (of the order of a few
percent) in relation to final vertical displacements,
which is consistent between the two DEMs, provid-
ing they are bundle-adjusted as a single image
block. Therefore, the technique can, in theory, be
used without any ground control at all, while still
130 Digital terrain modeling and glacier topographic characterization
producing accurate vertical displacement results
from the relative image orientation alone (Kääb
2002).
A major source of error in vertical differences
between multitemporal DEMs is their three-
dimensional co-registration. In cases where no com-
mon GCPs and tie points have been applied, severe
biases might exist in particular between space-
derived DEMs. Errors in elevation differences from
such a lack of precise co-registration might easily
exceed real elevation changes, and should therefore
be investigated as part of every DEM-differencing
work (cf. Fig. 5.4). The most efficient method to
investigate and possibly correct DEM biases is
the cosine method that relates elevation differences
found for stable terrain to the combination of ter-
rain slope and aspect (Kääb 2005a, Nuth and Kääb
2011).
While a 10–20 m elevation error may be accept-
able for the orthorectification of images, it is not
for glacier elevation change mapping. Typically, a
glaciologist aims at measuring the rate of change in
surface elevation within 0.5 m/yr (this is the upper
bound for the error). Larger errors are acceptable
only when massive (tens of meters), but unusual
elevation changes are monitored (Stearns and
Hamilton 2007). A glaciologically relevant accuracy
can be obtained by (1) increasing the time period
between the two datasets and/or (2) improving the
quality and limiting the biases of the DEMs. The
first strategy is generally explored by comparing
recent spaceborne DEMs (ASTER, SRTM, SPOT)
with maps from the 1950s, 1960s, or 1970s. Apply-
ing such historic topographic datasets with more
recently derived DEMs may permit determination
of glacier elevation changes across several time-
steps. For example, Kargel et al. in this book’s
Figure 15.4F show a technique that captures three
sequential DEM elevation-differencing analyses
(over a 40-year total period) into one RGB compo-
site image. Each color channel (RGB) represents
the earliest, middle, and the latest elevation-differ-
encing analysis, respectively; the composited image
provides immediate visual perception of the accel-
eration or deceleration of elevation changes coin-
cident with the color-scaled time period. In this
way, spatial, dynamic, and multitemporal informa-
tion on elevation changes are uniquely viewed
within one image. The second strategy relies on
higher resolution satellite images with stereo cap-
abilities such as SPOT5 HRS or ALOS PRISM.
Even in the latter case a minimal image time span
of about 3 to 5 years is required.
The accumulation zone is the most critical part of
the glacier when mapping elevation changes. If
DEMs are derived from optical images or aerial
photographs, the high reflectance of the year-round
snow-covered accumulation zone may lead to lim-
ited features in the photographs/images or, in the
worst case, saturation of the sensor (Kääb 2002). In
the case of old maps derived from aerial photo-
graphs, errors may have been introduced by inter-
polation of contours (Schiefer et al. 2007). For
satellite images, featureless terrain can lead to data
gaps or erroneous pits and spikes in DEMs of the
accumulation zone. Enhanced elevation change
errors may also affect the accumulation zone
because the ice-free regions, which are used for
detecting and modelling any elevation biases, are
located at lower altitude. Only a few nunataks
may be visible to check the presence of relative
elevation biases at higher altitude, but, in practice,
even this is difficult because they are generally steep
and small features. Consequently, any elevation
biases need to be extrapolated from the lowest to
the highest elevations. Other reasons for increased
uncertainties in the accumulation zone relate to
difficulty in choosing density values to convert
volume-to-mass changes and, when DEMs derived
from SAR images are used, the necessity to take
into account penetration of the radar signal
(depending on its frequency) into the snow/firn
(Rignot et al. 2001). Consequently, good practice
is to compute separate estimates of glacier volume
changes for the ablation and accumulation zones,
even if only a rough estimate of the position of the
equilibrium line altitude (ELA) is available. Differ-
ent uncertainties can thus be attributed to each of
the two regions.
A number of strategies exist to actually compute
glacier volume changes from a set of glacier eleva-
tion changes (Kääb 2008). (1) If the two (or more)
DEMs to be differenced are complete, the summa-
tion of vertical differences gives the volume differ-
ence. Since this is often not the case (see above),
either (2) the hypsographic method can be used
where mean thickness change for each elevation
bin is multiplied by its area and all bins summed,
or (3) polynomial variation of elevation changes
with height can be fitted to measured elevation
differences and the elevation differences for each
glacier point then computed according to its height
using the polynomial parameters. In cases where
one of the multitemporal elevation datasets is not
a grid, or elevation information is distributed
unevenly, such as in contour lines or altimetry

footprints, calculations must be made carefully. For
example, it might be advantageous in terms of mini-
mizing error propagation not to interpolate point
elevation data to a DEM and then difference two
datasets, but rather to first calculate differences
between the elevation points and the DEM, and
then to interpolate elevation differences across the
wider glacier surface. In a number of cases, eleva-
tion differences will show less spatial variability
than elevation itself, and interpolation routines will
thus produce less serious artifacts.
Data from nonoptical sensors can also provide
accurate measurement of glacier surface eleva-
tions and, with multitemporal datasets, elevation
changes. Spaceborne radar altimeters (on board
ERS-1, ERS-2, or Envisat, for example) are well
suited to flat and large ice sheets or ice caps but
are not suitable to survey mountain glaciers
targeted by GLIMS because of their kilometric
footprints. The smaller (60 m) footprint of the
Geoscience Laser Altimeter System (GLAS) on
board ICESat makes it a potential source of data
for large mountain glaciers and ice caps (Zwally et
al. 2002). However, the large spacing of the orbits,
the fact that they are not repetitive, and the diffi-
culty of removing the influence of clouds limit the
usefulness of ICESat data. Still, ICESat data can
be used to evaluate the quality of other DEMs
(Berthier and Toutin 2008) or to estimate glacier
elevation changes when compared with other
multitemporal elevation data (Sauber et al. 2005,
Surazakov and Aizen 2006, Kääb 2008). Repetitive
surveys by airborne laser altimetry have been shown
to be extremely accurate. However, central line sur-
face profiling may lead to errors due to the difficulty
of extrapolating to the whole glacier surface
(Arendt et al. 2006).
It is clear then that DEMs derived from remote
sensing can provide glacier elevation change data
on a range of temporal and spatial scales. While
useful in their own right, elevation change data
are also a fundamental component of glacier mass
balance calculations. If two DEMs cover the whole
glacier, their difference (i.e., volume change) can be
converted to mass balance assuming a constant
density of 900 kg m 3 in the ablation area, although
the true density value to use in the accumulation
zone is still the topic of ongoing research. Most
authors use the same density as in the ablation zone,
as using a density different from 900 kg m 3 implies
a change in the rate of densification. Ideally the
calculation should include quantification describing
any shift in the underlying ground (e.g., due to
Glacier mapping 131
isostatic rebound or tectonic movement) although
in reality this is normally small compared with
glacier surface elevation change. The ultimate result
is mass balance quantification expressed in meters
water equivalent (MWE) over the imaged time
period.
A number of studies have used this approach to
assess the mass balance of alpine glaciers, using
combinations of historical topographic maps (Riv-
era et al. 2007) and DEMs derived from SPOT
imagery (Berthier et al. 2007), SRTM (Racoviteanu
et al. 2007), ASTER (Khalsa et al. 2004), Corona
(Bolch et al. 2008), and laser altimetry (Sauber et al.
2005). Very few have compared remotely sensed
mass balance calculations with those derived by
field investigation, but those that have (e.g., Hagg
et al. 2004, Rabatel et al. 2005, Zemp et al. 2010)
have found good agreement between dataset
results. As noted above, because of the limited accu-
racy of elevation data derived by remote-sensing
methods, many studies have strived to obtain mass
balance estimates derived over longer, preferentially
decadal time-scales, to ensure the change in
observed data exceeds the uncertainty in the
approach.
5.6 GLACIER MAPPING
Snow and ice exhibit distinct spectral properties
that make accurate delineation of debris-free
glaciers possible using a variety of remote-sensing
techniques (e.g., thresholded ratio images), but the
automatic classification of debris-covered ice is
more complex, because of the inherent difficulty
of spectrally distinguishing supraglacial sediment
and landslide material from that present in the
surrounding terrain (Paul et al. 2004b). For this
reason, DEMs are widely used for terrain visualiza-
tions and to provide additional information on
landscape morphometry, land surface context,
and spatial topology, all of which facilitate a more
accurate and reliable classification result for debris-
covered glaciers.
At the most basic level, DEM data can be used to
generate terrain visualizations to aid the manual
delineation of glacier boundaries (Fig. 5.8). Glacier
termini are often difficult to identify using plani-
metric imagery alone because, particularly when
heavily debris-covered, they appear contiguous
with the proglacial zone, but small elevation varia-
tions visible in three-dimensional visualizations can
132 Digital terrain modeling and glacier topographic characterization
Figure 5.8. Simple 3D visualization (SPOT) HRVIR
panchromatic imagery acquired October 17, 2008,
overlain on SPOT HRS-derived DEM from imagery
spanning 2002–2004) of the Batura Glacier, Pakistan,
which could be used to aid glacier boundary or catch-
ment headwall delineation, where they are unclear
using planimetric data in isolation, for example. For
scale, the glacier is approximately 2 km in width across
the main trunk. Copyright CNES 2008/Distribution
Spot Image. Figure can also be viewed as Online Sup-
plement 5.2.
sometimes remove this ambiguity. Similarly, the
exact location of a catchment divide may be
unclear, especially where fresh snowfall masks the
headwall ridge; three-dimensional information is
particularly useful in this respect. For the glacier
itself, DEM data are useful for identifying areas
of low slope angle (see Fig. 5.5b), which are char-
acteristic of debris-covered glacier tongues in most
parts of the world. When combined with comple-
mentary data such as land cover classifications, and
with results from neighborhood and change detec-
tion analyses, it has been shown that slope angle
data can be used as the primary morphological
characteristic to delineate glacierized terrain (Paul
et al. 2004b, Bolch et al. 2007). However, specific
slope thresholds used to delineate glacier facies vary
from region to region, with ice masses in the Euro-
pean Alps generally characterized by short, steep
tongues relative to the long (>10 km) flat tongues
of parts of the Himalaya, for example. Further, this
approach depends on slope angle being calculated
across a sufficiently broad spatial kernel to avoid
highlighting local slope variability (e.g., natural
peaks and troughs across the glacier surface), while
retaining sufficient detail so that small-scale features
such as inner moraine flanks are not excluded from
the analysis.
Additional first-order information that can be
calculated from DEM data include landscape plan
and profile curvatures, which can be used to define
both glacial and nonglacial landform elements
(Bolch and Kamp 2006; see Fig. 5.5c–f ). By cluster-
ing surfaces with similar curvature characteristics it
is possible to differentiate between glacier surfaces
and valley bottoms (low convexity), ridges and
lateral moraines (high convexity), medial moraines
(moderate convexity), and transitions between
glacier margins and lateral moraines (high concav-
ity). A similar but more complex approach is to use
hierarchical object-oriented modeling, which
employs various DEM-derived terrain object prop-
erties such as slope angle, slope azimuth, curvature,
and relief to identify locally contiguous portions of
the landscape, which are iteratively aggregated to
form higher order landform objects at smaller and
smaller scales (Bishop et al. 2001). Glacier bound-
aries can thus be delineated at a landform level, and
objects and the relationships between objects can
be used for further landscape analyses, such as
studying the role of surface processes (e.g., ero-
sion/uplift) in topographic evolution. Morpho-
metric approaches such as these show great
potential for glacier mapping across the world,
but are heavily dependent on the input of high-
quality DEM data for accurate results.
A range of satellite sensors already have stereo-
scopic imaging capability and some recent missions
(e.g., Cartosat-1, ALOS PRISM, Terra-SAR-X,
and the TanDEM-X-mission) have significantly
increased the potential for extracting DEM data
at very high resolution for use in glacier mapping.
For many remote high-mountain regions of the
world, however, gaining access to high-quality
DEMs can be problematic, with ASTER and
SRTM data often providing the best (or only) avail-
able sources. Errors common in ASTER-derived
DEMs of steep, high-mountain relief can sometimes

produce artifacts in classification results, but may
still provide a quality of output that exceeds classi-
fication methods not employing a DEM of any sort
(Paul et al. 2004b). DEM resolution can be an
equally important consideration, with aerial
photography DEMs (such as those provided by
the Swiss Topo Survey) outperforming satellite sen-
sor–derived data in almost all cases (Bolch and
Kamp, 2006), particularly in the delineation of
small ice masses. In areas where accurate and
fine-resolution DEMs are not available, manual
digitization of small samples may still therefore
be necessary to achieve a satisfactory classification
result. Therefore, attempts have been made to intro-
duce additional information into the classification
process, such as metrics relating to geomorph-
ometry, texture, and context (Bishop and Shroder
2000, Bishop et al. 2001, Taschner and Ranzi 2002,
Paul et al. 2004b, Ranzi et al. 2004, Buchroithner
and Bolch 2007, Bolch et al 2008). Sophisticated
classification approaches use both first and sec-
ond-order topographic derivatives to segment land-
scape units accordingly, and make use of statistical,
artificial intelligence and hierarchical–structural
methods.
5.6.1 Pattern recognition
Pattern recognition techniques perform two key
tasks: description and classification. Given an
object (or set of objects—pixels in a satellite image,
for example), pattern recognition systems first seek
to describe those object(s), often in three dimen-
sions, and subsequently classify the objects based
on those descriptions. The rigor with which the
prior task is performed largely governs the quality
of the latter. The description and classification of
input data in this way can be handled by statistical
or structural methods or, more frequently, a com-
bination of both (the hybrid approach). Statistical
methods make use of decision theory concepts to
segment the landscape based on their quantitative
features, whereas structural methods make use of
syntactic grammars to segment the landscape based
on the arrangement of their morphological features
(see Section 5.6.3). Either way, DEM data are a
fundamental input to the classification procedure.
Statistical pattern recognition techniques are
largely dependent on the choice of model param-
eters, the quality of the expert knowledge used in
the prior part of the model, and the resolution of
the image and DEM data. For example, and with
specific reference to debris-covered glacier classifi-
Glacier mapping 133
cation, let R denote the pixel domain of a set of
co-registered data layers (e.g., satellite imagery,
elevation and geomorphometric parameters). Com-
bining all these layers into vector space results in a
feature vector per pixel. Similar vectors can be
derived to describe a shading field (to handle
across-image slope and exposition variability) and
a segmentation field, such that:
x : R ! R 000
~ vector field of features (observable)
~: R ! R
 000
shading field (unknown)
s:R!K segmentation field (unknown)
where K denotes the set of labels used for segmen-
tation. In the simplest case there are only two poss-
ible labels—one for a debris-covered glacier and
another for ‘‘background’’, but more are usually
employed to gain a full description of the environ-
ment. The statistical model relates the three quan-
tities (above) by a probability distribution:
pð~ ~; sÞ ¼ pð~
x;  xj ÞpðsÞ
~; sÞpð~
where it is assumed that shading and segmentation
are a priori independent.
If the parameters of the normal distribution are
already known, then the recognition task simply
requires the following: given the observation field
x we have to estimate the best shading field 
~ ~ and
then estimate the best segmentation field s. Routine
algorithms such as the Gibbs sampler (Geman et al.
1990) and the Expectation Maximization algorithm
(Schlesinger 1968, Dempster et al. 1977) are nor-
mally employed to make the estimations.
If the parameters of multivariate normal distri-
butions are unknown (first factor in the equation
above) they can be estimated by maximum likeli-
hood learning given an expert’s segmentation of the
scene. In this case the Expectation Maximization
algorithm is again applied but with partial super-
vision. The expert’s segmentation is relaxed by
defining only three regions: background, fore-
ground, and a zone between, leaving the system
to learn the optimal boundary. The resulting itera-
tive learning scheme comprises repetition of the
following subtasks:
1. Given the actual parameters of the normal dis-
tributions and the actual estimation of the shad-
ing field, sample the segmentations and estimate
the marginal probabilities pr ðsr ¼ k j  xÞ for
~; ~
each pixel r.
134 Digital terrain modeling and glacier topographic characterization

Figure 5.9. (a) Optical satellite image of the classification area (ASTER image acquired October 23, 2003);
(b) relaxed expert segmentation (see text); (c) first component of estimated shading; and (d) final segmentation
obtained after learning. Black, green and red encode background labels, blue and violet encode foreground labels,
yellow encodes the boundary label, and cyan encodes lakes. Figure can also be viewed as Online Supplement 5.3.

2. Use these marginals to improve the parameters
of the normal distributions (i.e., the covariance
matrices and mean vectors for all segments).
3. Re-estimate the shading field  ~ based on the
current marginal probabilities and parameters
of the normal distributions.
When the optimal boundary is reached, the itera-
tion is stopped and the pixels are assigned their
segment label (Fig. 5.9).
5.6.2 Artificial intelligence techniques
Artificial intelligence techniques (often referred to
as knowledge-based systems) essentially comprise a
set of logical rules that provide an analog for the
reasoning used in anthropogenic decision making.
These can be as simple as a series of IF–THEN–
ELSE-type statements that test input data (e.g., a
pixel) for certain characteristics (e.g., to be within a
range of spectral values) and then follow one of two
paths based on the result. Such ‘‘expert systems’’
systematically arrive at a single classification for
each pixel by ruling out alternatives based on input
data values. This mimics the decision-making pro-
cesses employed by a field geomorphologist, for
example; indeed it is important to note that such
systems can also exploit three-dimensional contex-
tual and shape information in the same way a human
can. DEMs are therefore a fundamental component
of a comprehensively designed (and truly three-
dimensional) expert system.
Artificial neural networks (ANNs) aim to process
information in a similar way to the human brain, by
using a large number of highly interconnected pro-
cessing elements (neurons) working in parallel to
solve a specific problem. The key difference between
ANNs and other artificial intelligence techniques
is that ANNs learn by example and cannot be
designed to perform a specific task in the first
instance. ANNs are particularly useful for handling

diverse and nonparametric datasets and for model-
ling nonlinear relationships, making them a particu-
larly attractive method for mapping glacierized
terrain. Nevertheless, the value of the output classi-
fication is only as good as the quality of the input
data; for ANN techniques to be truly robust they
must learn to recognize patterns associated with
geomorphometric land surface parameters, such
as texture, context, and spatial topology, which
can be derived (at least partly) from first-order
and second-order DEM analyses (Bishop et al.
2001, 2004).
Fuzzy classification algorithms are able to incor-
porate inaccurate sensor measurements, vague class
descriptions, and imprecise modeling into the anal-
ysis (Binaghi et al. 1997, Benz et al. 2003), and can
result in accurate portrayals of subpixel mixtures of
classes. Fuzzy classification procedures assign
values to each pixel representing the pixel’s degree
or probability of membership in each class, rather
than clustering similar pixels together into a
thematic set. The fuzzy approach is analogous to
human perception, in which there can be uncer-
tainty even between experts analyzing the same en-
vironment. Fuzzy sets are a classification tool in
themselves; they can be used alone or incorporated
into higher level artificial intelligence techniques
such as expert systems; either way it is assumed that
there is a priori expert knowledge about the en-
vironment under investigation to be able to design
fuzzy membership functions. In a similar way to
expert systems and ANNs, the addition of a topo-
graphic dataset into the classification is fundamen-
tal to the accurate identification of glacial
landforms through shape and contextual analyses
(Schneevoigt et al. 2008).
5.6.3 Object-oriented mapping
Object-oriented mapping describes the segmenta-
tion and partitioning of the landscape into discrete
spatial entities based upon their geomorphometric
characteristics (e.g., slope angle, slope azimuth, cur-
vature and relief; Bishop et al., 2001). These entities
represent the lowermost level of a hierarchical land-
scape structure, which in total describes the sup-
position of surface processes that comprise the
topography of a glacierized landscape (Bishop
and Shroder 2000). Therefore, by combining spatial
entities at the lowermost level, specific terrain feat-
ures can be identified; by combining terrain features
at the second level, landforms can be identified, and
so on. High-quality DEMs are clearly fundamental
Discussion 135
to this approach, if the spatial entities at the lower
end of the hierarchical structure are to be accurately
quantified and error propagation through the sys-
tem is to be avoided. This approach has been
successfully employed to accurately delineate the
ablation zone of the heavily debris-covered Raikot
Glacier in the Nanga Parbat Himalaya, for example
(Bishop et al. 2001). Terrain object properties were
found to be diagnostic of glacier processes repre-
senting glacierization, thereby differentiating glacier
topography from other surfaces governed by differ-
ent surface processes.
Object-oriented mapping can be used as a classi-
fication technique in its own right, but it has been
shown to be particularly effective when used in
combination with ANNs (see above) to classify
alpine glacial environments. In this specific context,
the procedure may follow five key stages (Raup et
al. 2007):
1. Classification of land cover using spectral data
and topography.
2. Spatial analysis of imagery to generate geo-
metric, shape, and topological information.
3. Generation of geomorphometric parameters
from DEMs.
4. Fusion of data into an object-oriented param-
eterization scheme.
5. Classification of supraglacial features and
glaciers by neural networks.
This hybrid approach appears to reduce error and
increase consistency in results when compared with
other approaches. Indeed, it is true of all the pre-
viously described techniques for glacier mapping
that they may yield useful terrain classification in
their own right, but they have even greater potential
when they are integrated together. Fundamental to
their success is the provision of a three-dimensional
analysis, however. Topological, contextual, shape,
and geomorphometric analyses are limited without
the use of a DEM, and numerous studies have
demonstrated the need for such three-dimensional
properties if fully automated glacier mapping is to
be realized (Bishop et al. 2004, Paul et al. 2004b,
Bolch and Kamp 2006).
5.7 DISCUSSION
It is clear that digital elevation data derived from
stereoscopic airborne and satellite imagery are fun-
damental to many essential cryospheric applica-
136 Digital terrain modeling and glacier topographic characterization
tions, and in particular to deriving glacier statistics
for inclusion in global databases. Square-gridded
(raster) DEMs are most widely employed by
researchers because they provide more realistic ter-
rain representations than data interpolated from
topographic maps and are easier to derive and to
handle than DEMs derived by radar interferometry
or airborne laser scanning. However, they are far
from perfect, particularly for detecting the abrupt
changes in topography that characterize glacial
environments, and the challenge of producing
fine-resolution DEM data with low data storage
requirements remains an issue. Further, the genera-
tion of reliable elevation data at any resolution finer
than that offered by the SRTM (90 m) remains
problematic for many glacierized areas of the
world. This is partly because of persistent cloud
cover, particularly in mountain regions, and a lack
of ground control data, particularly in areas of
political sensitivity. Additionally, SRTM data are
static and do not allow for multitemporal analyses.
While the recently released G-DEM2 may represent
a possible breakthrough in terms of spatial detail
(30 m pixel size) and accuracy, elevation values
represent the combined period of ASTER opera-
tions (2000–2011), so offer no better temporal reso-
lution than the SRTM DEM.
Indeed, it is this variability in data quality that
remains the major limiting factor in global glacier
mapping and characterization. It has consequences
for the type, number, and quality of glacier param-
eters that can be derived for any given area.
Furthermore, it influences the effectiveness of dif-
ferent methods and approaches, and makes the
standardization of procedures a challenging task.
Ideally, elevation data at resolutions and accuracies
comparable with those offered by airborne laser
altimetry are required for all glacierized regions
of the world. Unfortunately, without a wide-swath
spaceborne altimeter, we are many years (if not
decades) from being able to achieve such a dataset.
Even then, it will raise numerous issues regarding
high-density point clouds and preprocessing.
Nevertheless, there remains great potential in the
use of various approaches, such as the integration
of multiple multiscale elevation datasets for deriv-
ing multiscale information about a single glacier-
ized catchment, and the use of LiDAR data as
accurate ground control for photogrammetric
DEM extraction, among others. While novel
approaches to deriving glacier parameters have
been successful, some are yet to be effectively and
comprehensively employed on a regional basis. For
example, the geodetic method has been proven a
useful tool to assess glacier surface lowering and,
subsequently, to produce estimates of mass balance.
This work, however, has yet to be systematically
conducted in many regions of the world. Further,
where independent studies have been carried out
using such approaches, they are not always directly
comparable because of methodological incon-
sistencies, or inconsistency in the reporting of data
error.
Although remote-sensing sensor technology
has advanced significantly in recent years, the
challenges of representing topography in high-
mountain high-relief regions, and accurately quan-
tifying the error within those topographic data,
remain significant. For example, a range of studies
have shown that for many glacierized regions of the
world (mostly those of moderate relief ) the topog-
raphy can be characterized by ASTER DEMs with
an accuracy of 15–30 m (68% confidence level)
after some significant postprocessing, but only to
60 m (68% confidence) in areas with steep rock
headwalls and large low-contrast accumulation
areas (Kääb 2002, Toutin 2008). The issue is com-
plicated further in high-relief mountain areas by
slope–aspect error dependence (Bolch 2004), which
tends to result in better quality elevation data being
derived on slopes facing south (in the northern
hemisphere) because of advantageous illumination
and sensor–terrain viewing angle. Illumination
variations can be compensated for (to a certain
degree) using topographic normalization methods,
but this itself requires a DEM, which is not always
available at this processing step. In terms of
error quantification, some standardization of
methods is required. Inconsistencies currently exist
between studies in the manner in which errors
are reported: some studies quote errors to two
standard deviations whereas some only quote the
RMSE. A limited number do not even quantify the
expected error (or uncertainty) in the presented data
and, of those that do, the accuracy of data can
sometimes be highly dependent on the number
and positioning of chosen check points (Toutin
2008). It is often difficult, therefore, for an indepen-
dent researcher to replicate results, or to establish
the exact error analysis that has taken place and
how reliable it is.
Where reliable topographic data do exist they can
be used very effectively for mapping and change
detection studies. First-order geomorphometric
parameters have been successfully employed to
enhance glacier boundary delineation, calculate sur-

face lowering rates, and inform numerical models
(e.g., surface uplift/erosion). This latter effort,
specifically the integration of remote sensing and
modeling for the analysis of landform–process
coupling, remains in its early stages, along with
several other interesting possibilities that are bene-
fiting from better and more widely available digital
elevation data. For example, space-based methods
now exist capable of generating a mass balance
index to improve ELA estimation, and glacier flow
and erosion modeling are being enhanced by
improved and accurate boundary conditions. With
such ventures come further challenges, of course,
such as accounting for spatial and temporal fluctua-
tions that were previously beyond modeling resolu-
tion; many glacier simulations have until recently
neglected to account for daily and seasonal abla-
tion, ice velocity, and erosion fluctuations, for
example (MacGregor et al. 2000).
The accurate modeling of glacierized terrain is
essential because it plays such a fundamental role
in the modulation of atmospheric, Earth surface,
and glaciological processes (as well as being shaped
by these same processes). The glacial environment,
and in particular the glacier surface, is extremely
labile; this is especially so at present with major
adjustments occurring due to climatic forcing. On
a catchment scale, topography governs sediment
transport and ice fluxes, collectively influencing
glacier erosion. On a more local scale, glacier sur-
face topography can influence debris cover distri-
butions, debris depth, meltwater routing, and
supraglacial ponding. Conversely, processes and
parameters provide feedback that affects glacier
topography—thick debris covers can insulate
glacier ice, for example, and meltwater ponds ther-
mally erode ice at their margins and their bases. At
the most local level glacial features such as ogives,
moraines, seracs, and crevasse fields define the
topography of glacier surfaces. It is therefore clear
that there is complex two-way interaction between
topography and surface processes across a range of
spatial and temporal scales that needs to be better
understood to inform ideas relating to geomorpho-
logical landscape evolution.
The interaction between topography and most
geophysical processes occurs over a range of spatial
scales, which cannot currently be truly represented
within DEM data. Therefore, while the resolution
of the generated DEM is limited by the resolution
of input source data, scientific interpretations
should also be mindful to restrict analyses to the
natural scale of the terrain-dependent application.
Discussion 137
As a guide, the resolution of the derived DEM can
provide a practical indication of the scale of infor-
mation content; analyses of processes or features
that occur on a finer scale than this should be made
with caution. For example, relatively small-scale
features (e.g., moraine ridges) can be entirely missed
by medium-resolution datasets, yet may change an
interpretation of a process or set of features with
their presence. Conversely, with a fine-resolution
DEM, analyses made on a coarser scale may be
hampered by noise, and indeed a coarser resolution
DEM may be more appropriate for use. With con-
stant computational developments, methods for
representing fine-scale shape and structure are con-
tinually improving and so is the incorporation of
terrain structure into considerations of spatial scale.
A classic example of this is the development of
scale-dependent classification methods, which are
increasingly considering terrain hierarchies with
the realization that landforms are often not
spatially delimited and that one landform may par-
tially comprise another.
Consequently, research on glacier mapping and
the extraction of glacier parameters indicates that
the integration of topographic information on a
variety of spatial scales is important in order to
produce reliable results. However, methodologies
significantly vary, and some problems remain for
high-mountain areas, particularly in change detec-
tion studies. For example, surface features must be
maintained across the imaged period for accurate
elevation datasets to be derived. This depends on
there being sufficient optical contrast between the
feature and its surroundings as well as the feature
being large enough to be resolved given the sensor
resolution; and it is also dependent on there being
no (or minimal) modification of the feature shape,
size, and contrast during the change, or the masking
of the feature by snow cover, for example. Further,
the magnitude of the change(s) to be detected may
not exceed the uncertainty in the method employed.
Errors in satellite-derived DEMs of high-relief
terrain are particularly problematic; consequently
it is unlikely that researchers will be able to reliably
detect surface lowering on a timescale of less than a
decade. The solution is to either increase the image
resolution (and thus accuracy) or to extend the
observation period (thus increasing the magnitude
of change). The latter solution is becoming increas-
ingly possible as data archives extend with time and
new sources of historical data (e.g., Corona; Bolch
et al. 2008) emerge. The former issue is aided by the
launch of new and improved sensors.
138 Digital terrain modeling and glacier topographic characterization
Remote sensing is a constantly advancing disci-
pline, so the launch of a new and/or improved
sensor is often imminent and the promise of more
accurate or more detailed imagery is never far away.
Some of the most promising developments for the
generation of new elevation data come from non-
optical sources. For example, the recently successful
launch of TanDEM-X, the interferometric partner
of TerraSAR-X, promises to provide a global DEM
to HRTI-3 specifications (Weber et al. 2006; i.e., 12
m spatial resolution, <2 m relative vertical accuracy
and <10 m absolute vertical accuracy). These
specifications can only currently be achieved by
aerial photogrammetry and very fine–resolution
optical satellite imagery such as WorldView, but
with associated costs. Spaceborne laser altimeters
(e.g., GLAS, on board ICESat-1) provide compar-
able accuracy, but with very limited swath width
and low point density, and there is no real prospect
of routine collection of such data again until the
proposed launch of ICESat-2 in 2015.
If topographic and spectral data are to be truly
integrated into surface process modeling, and
results compared between independent studies,
there is a requirement for a representational frame-
work that scientists can work towards. Currently,
scientists employ a range of analytical tools, algo-
rithms, processing approaches, and software for the
generation, manipulation, and interpretation of
topographic data. Methods are highly empirical,
thus the type and quality of the derived data are
dependent, to a large extent, on the analyst. Con-
sequently, replication of existing results can be
difficult (even more so the application of one pub-
lished technique to a new area). Standardization
and protocols for information extraction and inte-
gration are therefore required if data quality, result
accuracy, and the validity of comparing measure-
ments across different studies (and study areas) are
to be assured.
5.8 CONCLUSIONS
Digital elevation models are fundamental to the
extraction of three-dimensional glacier parameters
for inclusion in global databases (e.g., WGI,
GLIMS, GlobGlacier) as well as for calculating
mass balance data, automatically delineating
glacier boundaries, modeling surface energy bal-
ance and radiation fluxes, characterizing geomor-
phometry, and analyzing altitudinal-dependent
processes. A range of sensors exists that offers data
appropriate for the extraction of elevation informa-
tion, but ASTER remains the most widely used data
source because of its stereoscopic capability, wide
spectral range, medium-to-fine spatial resolution
and, importantly, low cost. The extraction of eleva-
tion information from ASTER data performs well
in areas with low relief and gently sloping topog-
raphy, but errors can be large in areas with steep
rock headwalls and in areas with low contrast (e.g.,
fresh snow cover). A range of postprocessing tools
can be used to identify and reduce such errors
before any secondary analyses are undertaken.
Digital elevation models from nonoptical sources
(e.g., LiDAR, InSAR) are becoming increasingly
common and can often improve on, or be fused
with, existing elevation datasets to enhance cryo-
spheric studies. Such sources have tremendous
potential for the extraction of three-dimensional
data over glacierized terrain in the future, and in
particular for increasing the temporal resolution of
change detection analyses, which is currently of the
order of a decade or more. However, they may
also bring new challenges in terms of handling data
volume (particularly in the case of LiDAR point
clouds) and in upscaling acquired information,
where such detail is not required. For this reason,
the development and publication of standard meth-
ods (or protocols) for acquiring, processing, and
representing digital elevation data will become
increasingly important. In the short term, chal-
lenges remain in identifying and quantifying alti-
metric errors, particularly when comparing DEMs
from different sources, and in simply gaining access
to reliable data for some of the most politically
sensitive regions of the world.
5.9 ACKNOWLEDGMENTS
Quincey was funded by a Research Council U.K.
Fellowship; Bishop was funded by the National
Aeronautics and Space Administration under the
NASA OES-02 program (Award NNG04GL84G).
ASTER data courtesy of NASA/GSFC/METI/
Japan Space Systems, the U.S./Japan ASTER
Science Team, and the GLIMS project.
5.10 REFERENCES
Allen, T.R. (1998) Topographic context of glaciers and
perennial snowfields, Glacier National Park, Montana.
Geomorphology, 21, 207–216.

Arendt, A., Echelmeyer, K., Harrison, W.D., Lingle, G.,
and Valentine, V. (2002) Rapid wastage of Alaska
glaciers and their contribution to rising sea level.
Science, 297, 382–386.
Arendt, A.A., Echelmeyer, K., Harrison, W., Lingle, C.,
Zirnheld, S., Valentine, V., Ritchie, B. and Drucken-
miller, M. (2006) Updated estimates of glacier volume
changes in the western Chugach Mountains, Alaska,
and a comparison of regional extrapolation methods.
Journal of Geophysical Research, 111, F03019.
Arnold, N.S., Rees, W.G., Hodson, A.J., and Kohler, J.
(2006a) Topographic controls on the surface energy
balance of a high Arctic valley glacier. Journal of Geo-
physical Research, 111, F02011.
Arnold, N.S., Rees, W.G., Devereux, B.J., and Amable,
G.S. (2006b) Evaluating the potential of high-
resolution airborne LiDAR data in glaciology. Inter-
national Journal of Remote Sensing, 27(6), 1233–1251.
Baltsavias, E.P. (1999) A comparison between photo-
grammetry and laser scanning. ISPRS Journal of
Photogrammetry and Remote Sensing, 54, 83–94.
Bamber, J.L., and Payne, A.J. (2004) Mass Balance of the
Cryosphere: Observations and Modelling of Contempor-
ary and Future Changes. Cambridge University Press,
Cambridge, U.K.
Bamber, J.L., and Rivera, A. (2007) A review of remote
sensing methods for glacier mass balance determina-
tion. Global and Planetary Change, 59, 138–148.
Barry, R.G. (2006) The status of research on glaciers and
global glacier recession: A review. Progress in Physical
Geography, 30(3), 285–306.
Benz, U.C., Hofmann, P., Willhauck, G., Lingenfelder,
I., and Heynen, M. (2003) Multi-resolution, object-
oriented fuzzy analysis of remote sensing data for
GIS-ready information. ISPRS Journal of Photogram-
metry and Remote Sensing, 58(3/4), 239–258.
Berthier, E., and Toutin, T. (2008) SPOT5-HRS digital
elevation models and the monitoring of glacier eleva-
tion changes in North-West Canada and South-East
Alaska. Remote Sensing of Environment, 112, 2443–
2454.
Berthier, E., Arnaud, Y., Vincent, C., and Remy, F.
(2006) Biases of SRTM in high-mountain areas: Impli-
cations for the monitoring of glacier volume changes.
Geophysical Research Letters, 33, L08502.
Berthier, E., Arnaud, Y., Rajesh, K., Sarfaraz, A.,
Wagnon, P., and Chevallier, P. (2007) Remote sensing
estimates of glacier mass balances in the Himachal
Pradesh (Western Himalaya, India). Remote Sensing
of Environment, 108(3), 327–338.
Berthier, E., Schiefer, E., Clarke, G.K., Menounos, B.,
and Rémy, F. (2010) Contribution of Alaskan glaciers
to sea-level rise derived from satellite imagery. Nature
Geosci., 3, 92–95.
Binaghi, E., Madella, P., Montesano, M.G., and
Rampini, A. (1997) Fuzzy contextual classification of
References 139
multisource remote sensing images. IEEE Transactions
on Geoscience and Remote Sensing, 35(2), 326–340.
Bishop, M.P. and Shroder, J.F., Jr. (2000) Remote
sensing and geomorphometric assessment of topo-
graphic complexity and erosion dynamics in the Nanga
Parbat massif. Geological Society, London, Special
Publications, 170, 181–200.
Bishop, M.P., Shroder, J.F., Jr., Hickman, B.L., and
Copland, L. (1998) Scale dependent analysis of satellite
imagery for characterization of glacier surfaces in the
Karakoram Himalaya. Geomorphology, 21, 217–232.
Bishop, M.P., Bonk, R., Kamp, U., and Shroder, J.F., Jr.
(2001) Terrain analysis and data modelling for alpine
glacier mapping. Polar Geography, 25(3) 182–201.
Bishop, M.P., Shroder, J.F., Jr., and Colby, J.D. (2003)
Remote sensing and geomorphometry for studying
relief production in high mountains. Geomorphology,
55, 345–361.
Bishop, M.P., Barry, R.G., Bush, A.B.G., Copland, L.,
Dwyer, J.L., Fountain, A.G., Haeberli, W., Hall,
D.K., Kääb, A., Kargel, J.S. et al. (2004) Global
land-ice measurements from space (GLIMS): Remote
sensing and GIS investigations of the Earth’s cryo-
sphere. Geocarto International, 19(2), 57–84.
Bolch, T. (2004) Using ASTER and SRTM DEMs for
studying glaciers and rockglaciers in northern Tien
Shan. Paper presented at Proceedings of Theoretical
and Applied Problems of Geography on a Boundary of
Centuries, Almaty, Kazakhstan, June 8–9, 2004, Part I,
pp. 254–258.
Bolch, T., and Kamp, U. (2006) Glacier mapping in high
mountains using DEMs, Landsat and ASTER data.
Paper presented at Proceedings Eighth International
Symposium on High Mountain Remote Sensing Carto-
graphy, March 20–27, 2005, La Paz, Bolivia (Grazer
Schriften der Geographie und Raumforschung, 41),
Institut für Geographie und Raumforschung, Univer-
sität Graz, Austria, pp. 13–24.
Bolch, T., and Schröder, H. (2001) Geomorphologische
Kartierung und Diversitätsbestimmung der Periglazial-
formen am Cerro Sillajhuay (Chile/Bolivien) (Erlanger
Geographische Arbeiten, Sonderband 28), Institut für
Geographie, Universität Erlangen, Germany, 141 pp.
[in German].
Bolch, T., Kamp, U., and Olsenholler, J. (2005) Using
ASTER and SRTM DEMs for studying geomorphol-
ogy and glaciers in high mountain areas. In: M. Oluic
(Ed.), New Strategies for European Remote Sensing:
Proceedings 24th Annual Symposium EARSeL, May
25–27, 2004, Dubrovnik, Croatia, Millpress, Rotter-
dam, The Netherlands, pp. 119–127.
Bolch, T., Buchroithner, M., Kunert, A., and Kamp, U.
(2007) Automated delineation of debris-covered
glaciers based on ASTER data. Paper presented at
Geoinformation in Europe: Proceedings of 27th
EARSeL Symposium, June 4–7, 2007, Bolzano, Italy,
pp. 403–410.
140 Digital terrain modeling and glacier topographic characterization
Bolch, T., Buchroithner, M., Pieczonka, T., and Kunert,
A. (2008) Planimetric and volumetric glacier changes in
the Khumbu Himal, Nepal, since 1962 using Corona,
Landsat TM and ASTER data. Journal of Glaciology,
54(187), 592–600.
Braithwaite, R.J. and Raper, S.C.B. (2010) Estimating
equilibrium-line altitude (ELA) from glacier inventory
data. Annals of Glaciology, 50(53), 127–132.
Brock, B., Willis, I.C., Sharp, M.J., and Arnold, N.S.
(2000) Modelling seasonal and spatial variations in
the surface energy balance of Haut Glacier d’Arolla,
Switzerland. Annals of Glaciology, 31, 53–62.
Buchroithner, M., and Bolch, T. (2007) An automated
method to delineate the ice extension of the debris-
covered glaciers at Mt. Everest based on ASTER
imagery. Paper presented at Proceedings Ninth Inter-
national Symposium on High Mountain Remote Sensing
Cartography (Grazer Schriften der Geographie und
Raumforschung, 43), Institut für Geographie und
Raumforschung, Universität Graz, Austria, pp. 71–78.
Burrough, P.A., and McDonnell, R.A. (1998) Principles
of Geographic Information Systems. Oxford University
Press, Oxford, U.K.
Carabajal, C.C., and Harding, D.J. (2005) ICESat valida-
tion of SRTM C-band digital elevation models. Geo-
physical Research Letters, 32, L22S01.
Carlisle, B.H. (2005) Modelling the spatial distribution of
DEM error. Transactions in GIS, 9(4), 521–540.
Carr, S., and Coleman, C. (2007) An improved technique
for the reconstruction of former glacier mass-balance
and dynamics. Geomorphology, 92, 76–90.
Cogley, J.G., and Jung-Rothenhausler, F. (2004). Uncer-
tainty in digital elevation models of Axel Heiber
Island, Arctic Canada. Arctic, Antarctic and Alpine
Research, 36, 249–260.
Copland, L., Pope, S., Bishop, M.P., Shroder, J.F., Jr.,
Clendon, P., Bush, A., Kamp, U., Seong, Y.B., and
Owen, L.A. (2009) Glacier velocities across the central
Karakoram. Annals of Glaciology, 50(52) 41–49.
Crippen, R.E. (2010) Global topographical exploration
and analysis with the SRTM and ASTER elevation
models. Geological Society Special Publications, 345,
5–15.
Cuartero, A., Felicı́simo, A.M., and Ariza, F.J. (2004)
Accuracy of DEM generation from TERRA-ASTER
stereo data. International Archives of Photogrammetry
and Remote Sensing, 35(B2), 559–563.
de Jong, S.M., and Burrough, P.A. (1995) A fractal
approach to the classification of Mediterranean vegeta-
tion types in remotely sensed images. Photogrammetric
Engineering and Remote Sensing, 61, 1041–1053.
Dempster, A.P., Laird, N.M., and Rubin, D.B. (1977)
Maximum likelihood from incomplete data via the
EM algorithm. Journal of the Royal Statistical Society,
Series B (Methodological), 39(1), 1–38.
Drăgut L., Eisank, C., Strasser, T., and Blaschke, T.
(2009) A comparison of methods to incorporate scale
in geomorphometry. Paper presented at Proceedings of
Geomorphometry 2009, Zurich, Switzerland, August
31–September 2, pp. 133–139.
Dyurgerov, M.B., and Meier, M.F. (2000) Twentieth
century climate change: Evidence from small glaciers.
Proceedings of the National Academy of Science, 97(4),
1406–1411.
Eckert, S., Kellenberger, T., and Itten, K. (2005) Accu-
racy assessment of automatically derived digital eleva-
tion models from ASTER data in mountainous terrain.
International Journal of Remote Sensing, 26(9), 1943–
1957.
Egholm, D.L., Nielsen, S.B., Pedersen V.K., and
Lesemann, J.-E. (2009) Glacial effects limiting moun-
tain height. Nature, 460, 884–887.
Eldhuset, K., Andersen, P.H., Hauge, S., Isaksson, E.,
and Weydahl, D.J. (2003) ERS tandem InSAR proces-
sing for DEM generation, glacier motion estimation
and coherence analysis on Svalbard. International
Journal of Remote Sensing, 24(7), 1415–1437.
Etzelmüller, B., and Björnsson, H. (2000) Map analysis
techniques for glaciological applications. International
Journal of Geographical Information Science, 14, 567–
581.
Favey, E., Geiger, A., Gudmundsson, G.H., and Wehr,
A. (1999) Evaluating the potential of an airborne laser-
scanning system for measuring volume changes of
glaciers. Geografiska Annaler, 81A, 555–561.
Fisher, P.F., and Tate, N.J. (2006) Causes and conse-
quences of error in digital elevation models. Progress
in Physical Geography, 30(4), 467–489.
Flach, B., Kask, E., Schlesinger, D., and Skulish, A.
(2002) Unifying registration and segmentation for
multi-sensor images. In: L. Van Gool (Ed.), Pattern
Recognition (Lecture Notes in Computer Science,
2449), Springer-Verlag, Berlin, pp. 190–197.
Furbish, D.J., and Andrews, J.T. (1984) The use of hyp-
sometry to indicate long-term stability and response of
valley glaciers to changes in mass transfer. Journal of
Glaciology, 30(105), 199–211.
Gardelle, J., Arnaud, Y., and Berthier, E. (2011) Con-
trasted evolution of glacial lakes along the Hindu Kush
Himalaya mountain range between 1990 and 2009.
Global Planet. Change, 75, 47–55.
Gardner, A.S., Moholdt, G., Cogley, J.G., Wouters, B.,
Arendt, A.A., Wahr, J., Berthier, E., Hock, R., Pfeffer,
W.T., Kaser, G. et al. (2013) A reconciled estimate of
glacier contribuions to sea-level rise: 2003 to 2009.
Science, 340, 852–857.
Geman, D., Geman, S., Graffigne, C., and Dong, P.
(1990) Boundary detection by constrained optimiza-
tion. IEEE Transactions on Pattern Analysis and
Machine Intelligence, 12(7), 609–628.
Georgopoulos, A., and Skarlatos, D. (2003) A novel
method for automating the checking and correction
of digital elevation models using orthophotographs.
The Photogrammetric Record, 18(102), 156–163.

Gonçalves, J.A., and Oliveira, A.M. (2004) Accuracy
analysis of DEMs derived from ASTER imagery.
International Archives of Photogrammetry and Remote
Sensing, 35, 168–172.
Goodchild, M.F., and Mark, D.M. (1987) The fractal
nature of geographic phenomena. Annals of the Asso-
ciation of American Geographers, 77(2), 265–278.
Gratton, D.J., Howarth, P.J., and Marceau, D.J. (1990)
Combining DEM parameters with Landsat MSS and
TM imagery in a GIS for mountain glacier character-
ization. IEEE Transactions on Geosciences and Remote
Sensing, 28, 766–769.
Haeberli, W., Barry, R., and Cihlar, J. (2000) Glacier
monitoring within the Global Climate Observing Sys-
tem. Annals of Glaciology, 31, 241–246.
Hagg, W., Braun, L., Uvarov, V.N., and Makarevich,
K.G. (2004) A comparison of three methods of mass
balance determination in the Tuyuksu Glacier Region,
Tien Shan. Journal of Glaciology, 50(171), 505–510.
Heitzinger, D. and Kager, H. (1999) Hochwertige Ge-
ländemodelle aus Höhenlinien durch wissensbasierte
Klassifikation von Problemgebieten. Photogrammetrie
Fernerkundung Geoinformation, 1, 29–40 [in German].
Hirano, A., Welch, R., and Lang, H. (2003) Mapping
from ASTER stereo image data: DEM validation
and accuracy assessment. ISPRS Journal of Photo-
grammetry and Remote Sensing, 57(5/6), 356–370.
Hock, R. (2005) Glacier melt: A review of processes and
their modeling. Progress in Physical Geography, 29(3),
362–391.
Honikel, M. (2002) Fusion of spaceborne stereo-optical
and interferometric SAR data for digital terrain model
generation. Mitteilungen des Institutes für Geodäsie und
Photogrammetrie, 76.
Huggel, C., Kääb, A., Haeberli, W., Teysseire, P., and
Paul, F. (2002) Remote sensing based assessment of
hazards from glacier lake outbursts: A case study in
the Swiss Alps. Canadian Geotechnical Journal, 39(2),
316–330.
Kääb, A. (2000) Photogrammetric reconstruction of
glacier mass balance using a kinematic ice-flow model:
A 20 year time series on Grubengletscher, Swiss Alps.
Annals of Glaciology, 31, 45–52.
Kääb, A. (2002) Monitoring high-mountain terrain
deformation from repeated air- and spaceborne optical
data: Examples using digital aerial imagery and
ASTER data. ISPRS Journal of Photogrammetry and
Remote Sensing, 57, 39–52.
Kääb, A. (2005a) Remote Sensing of Mountain Glaciers
and Permafrost Creep (Schriftenreihe Physische Geo-
graphie, 48), University of Zurich, Switzerland, 266 pp.
Kääb, A. (2005b) Combination of SRTM3 and repeat
ASTER data for deriving alpine glacier flow velocities
in the Bhutan Himalaya. Remote Sensing of Environ-
ment, 94(4), 463–474.
Kääb A. (2008) Glacier volume changes using ASTER
satellite stereo and ICESat GLAS laser altimetry: A
References 141
test study on Edgeøya, Eastern Svalbard. IEEE Trans-
actions on Geoscience and Remote Sensing, 46(10),
2823–2830.
Kääb, A., and Funk, M. (1999) Modelling mass balance
using photogrammetric and geophysical data: A pilot
study at Griesgletscher, Swiss Alps. Journal of Glaciol-
ogy, 45, 575–583.
Kääb, A., Haeberli, W., and Gudmundsson, G.H. (1997)
Analysing the creep of mountain permafrost using high
precision aerial photogrammetry: 25 years of monitor-
ing Gruben rock glacier, Swiss Alps. Permafrost and
Periglacial Processes, 8, 409–426.
Kääb, A., Huggel, C., Fischer, L., Guex, S., Paul, F.,
Roer, I., Salzmann, N., Schlaefli, S., Schmutz, K.,
Schneider, D. et al. (2005) Remote sensing of glacier-
and permafrost related hazards in high mountains: An
overview. Natural Hazards and Earth System Science,
5(4), 527–554.
Kamp, U., Bolch, T., and Olsenholler, J. (2005) Geomor-
phometry of Cerro Sillajhuay (Andes, Chile/Bolivia):
Comparison of digital elevation models (DEMs) from
ASTER remote sensing data and contour maps. Geo-
carto International, 20, 23–34.
Kargel, J.S., Abrams, M.J., Bishop, M.P., Bush, A.,
Hamilton, G., Jiskoot, H., Kääb, A., Kieffer, H.H.,
Lee, E.M., Paul, F. et al. (2005) Multispectral imaging
contributions to global land ice measurements from
space. Remote Sensing of the Environment, 99, 187–219.
Kargel, J.S., Leonard, G., Crippen, R.E., Delaney, K.B.,
Evans, S.G., and Schneider J. (2010) Satellite monitor-
ing of Pakistan’s rockslide-dammed Lake Gojal. EOS
Trans. Am. Geophys. Union, 91(43), doi: 10.1029/
2010EO430002.
Kargel, J., Furfaro, R., Kaser, G., Leonard, G., Fink, W.,
Huggel, C., Kääb, A., Raup, B., Reynolds, J., Wolfe,
D. et al. (2011) ASTER imaging and analysis of glacier
hazards. In: B. Ramachandran, C.O. Justice, and M.J.
Abrams (Eds.), Land Remote Sensing and Global En-
vironmental Change: NASA’s Earth Observing System
and the Science of Terra and Aqua, Springer-Verlag,
New York, pp. 325–373.
Kargel., J.S., Alho, P., Buytaert, W., Celleri, R., Cogley,
J.G., Dussaillant, A., Zambrano, G., Haeberli, W.,
Harrison, S., Leonard, G. et al. (2012) Glaciers in
Patagonia: Controversy and prospects. EOS Trans.
Am. Geophys. Union, 93, 212–213.
Kaser, G. (2001) Glacier–climate interaction at low lati-
tudes. Journal of Glaciology, 47, 195–204.
Katsube, K., and Oguchi, T. (1999) Altitudinal changes in
slope angle and profile curvature in the Japan Alps: A
hypothesis regarding a characteristic slope angle. Geo-
graphic Review of Japan, 72B, 63–72.
Khalsa, S.J.S., Dyurgerov, M.B., Khromova, T., Raup,
B.H., and Barry, R.G. (2004). Space-based mapping of
glacier changes using ASTER and GIS tools. IEEE
Transactions on Geoscience and Remote Sensing,
42(10), 2177–2183.
142 Digital terrain modeling and glacier topographic characterization
Kubik, K., and Botman, A.G. (1976) Interpolation accu-
racy for topographic and geological surfaces. ITC
Journal, 2, 236–274.
Kyriakidis, P.C., Shortridge, A.M., and Goodchild, M.F.
(1999) Geostatistics for conflation and accuracy assess-
ment of digital elevation models. International Journal
of Geographical Information Science, 13, 677–707.
Larsen, C.F., Motyka, R.J., Arendt, A.A., Echelmeyer,
K.A., and Geissler, P.E. (2007) Glacier changes in
southeast Alaska and northwest British Columbia
and contribution to sea level rise. Journal of Geophys-
ical Research, 112, F01007.
Lathrop, R.G., and Peterson, D.L. (1992) Identifying
structural self-similarity in mountainous landscapes.
Landscape Ecology, 6(4), 233–238.
Leica Helava (2004) ERDAS Imagine Version 8.7 and
Leica Photogrammetry Suite (LPS) User Guide, Leica
Helava, London.
Leonard, K.C. and Fountain, A.G. (2003) Map-based
methods for estimating glacier equilibrium line alti-
tudes. Journal of Glaciology, 49(166), 329–336.
Li, D., Wang, S., and Li, R. (1996) Automatic quality
diagnosis in DTM generation by digital image match-
ing techniques. Geomatica, 50(1), 65–73.
Li, Z., Zhu, Q., and Gold, C. (2005) Digital Terrain
Modeling: Principles and Methodology, CRC Press,
Boca Raton, FL.
Luckman, A., Quincey, D.J., and Bevan, S. (2007) The
potential of satellite radar interferometry and feature
tracking for monitoring flow rates of Himalayan
glaciers. Remote Sensing of Environment, 111, 172–181.
MacGregor, K.R., Anderson, R.S., Anderson, S.P., and
Waddington, E.D. (2000) Numerical simulations of
glacial-valley longitudinal profile evolution. Geology,
28, 1031–1034.
Meier, M.F. (1984) Contribution of small glaciers to
global sea level. Science, 226, 1418–1421.
Meier, M.F., Dyurgerov, M.B., and McCabe, G.J. (2003)
The health of glaciers: Recent changes in glacier
regime. Climatic Change, 59, 123–135.
Mennis, J.L., and Fountain, A.G. (2001) A spatio-
temporal GIS database for monitoring alpine glacier
change. Photogrammetric Engineering and Remote
Sensing, 67, 967–975.
Miller, M.M., and Pelto, M.S. (2003) Mass balance meas-
urements on the Lemon Creek Glacier, Juneau Icefield,
Alaska 1953–1998. Geografiska Annaler: Series A,
Physical Geography, 81(4), 671–681.
Mitasova, H., and Mitas, L. (1993) Interpolation by reg-
ularized spline with tension, I: Theory and Implemen-
tation. Mathematical Geology, 25, 641–655.
Möller, M., Schneider, C., and Kilian, R. (2007) Glacier
change and climate forcing in recent decades at Gran
Campo Nevado, southernmost Patagonia. Annals of
Glaciology, 46, 136–144.
Molnar, P., and England, P. (1990) Late Cenozoic uplift
of mountain ranges and global climate change:
Chicken or egg? Nature, 346, 29–34.
Muskett, R.R., Lingle, C.S., Tangborn, W.V., and
Rabus, B.T. (2003) Multi-decadal elevation changes
on Bagley Ice Valley and Malaspina Glacier, Alaska.
Geophysical Research Letters, 30(16), 1857.
Nuth, C., and Kääb, A. (2011) Co-registration and bias
corrections of satellite elevation data sets for quantify-
ing glacier thickness change. The Cryosphere, 5, 271–
290.
Oksanen, J., and Sarjakoski, T. (2005) Error propagation
of DEM-based surface derivatives. Computers and
Geosciences, 31, 1015–1027.
Østrem, G., and Haakensen, N. (1999) Map comparison
or traditional mass-balance measurements: Which
method is better? Geografiska Annaler, 81A(4), 703–
711.
Paul, F., Kääb, A, Maisch, M., Kellenberger, T., and
Haeberli, W. (2004a) Rapid disintegration of Alpine
glaciers observed with satellite data. Geophysical
Research Letters, 31, L21402.
Paul, F., Huggel, C., and Kääb, A. (2004b) Combining
satellite multispectral image data and a digital eleva-
tion model for mapping debris-covered glaciers.
Remote Sensing of Environment, 89, 510–518.
Peng, X., Wang, J., and Zhang, Q. (2005) Deriving terrain
and textural information from stereo RADARSAT
data for mountainous land cover mapping. Inter-
national Journal of Remote Sensing, 26(22), 5029–5049.
Quincey, D.J., Richardson, S.D., Luckman, A., Lucas,
R.M., Reynolds, J.M., Hambrey, M.J., and Glasser,
N.F. (2007) Early recognition of glacial lake hazards in
the Himalaya using remote sensing datasets. Global
and Planetary Change, 56, 137–152.
Quincey, D.J., Luckman, A., and Benn, D. (2009) Quan-
tification of Everest-region glacier velocities between
1992 and 2002, using satellite radar interferometry
and feature tracking. Journal of Glaciology, 55, 596–
606.
Rabatel, A., Dedieu, J.P., and Vincent, C. (2005) Using
remote-sensing data to determine equilibrium-line alti-
tude and mass-balance time series: Validation on three
French glaciers, 1994–2002. Journal of Glaciology,
51(175), 539–546.
Rabus, B., Eineder, M., Roth, A., and Bamler, R. (2003)
The shuttle radar topography mission: A new class of
digital elevation models acquired by spaceborne radar.
ISPRS Journal of Photogrammetry and Remote
Sensing, 57, 241–262.
Racoviteanu, A.E., Manley, W.F., Arnaud, Y., and Wil-
liams, M.W. (2007) Evaluating digital elevation models
for glaciologic applications: An example from Nevado
Coropuna, Peruvian Andes. Global and Planetary
Change, 59, 110–125.
Ranzi, R., Grossi, G., Iacovelli, L., and Taschner, S.
(2004) Use of multispectral ASTER images for map-

ping debris-covered glaciers within the GLIMS Pro-
ject. Paper presented at IEEE International Geoscience
and Remote Sensing Symposium, Vol. II, pp. 1144–
1147.
Rasemann, S., Schmidt, J., Schrott, L., and Dikau, R.
(2004) Geomorphometry in mountain terrain. In: M.
Bishop and J.F. Shroder (Eds.), Geographic Informa-
tion Science and Mountain Geomorpology, Springer-
Verlag, Berlin, pp. 101–137.
Raup, B., Kääb, A., Kargel, J.S., Bishop, M.P., Hamil-
ton, G., Lee, E.M., Paul, F., Rau, F., Soltesz, D.,
Khalsa, S.J. et al. (2007) Remote sensing and GIS
technology in the Global Land Ice Measurements from
Space (GLIMS) Project. Computers and Geosciences,
33, 104–125.
Rees, W.G., and Arnold, N.S. (2007) Mass balance
and dynamics of a valley glacier measured by high-
resolution LiDAR. Polar Record, 43, 311–319.
Reiners, P.W., Ehlers, T.A., Mitchell, S.G., and Mont-
gomery, D.R. (2003) Coupled spatial variations in pre-
cipitation and long-term erosion rates across the
Washington Cascades. Nature, 426, 645–647.
Rickenbacher, M. (1998) Die digitale Modellierung des
Hochgebirges im DGM25 des Bundesamtes fuer
Landestopographie. Wiener Schriften zur Geographie
und Kartographie, 11, 49–55 [in German].
Rignot, E., Echelmeyer, K., and Krabill, W. (2001) Pene-
tration depth of interferometric synthetic-aperture
radar signals in snow and ice. Geophysical Research
Letters, 28(18), 3501–3504.
Rignot, E., Rivera, A., and Casassa, G. (2003) Contribu-
tion of the Patagonia Icefields of South America to Sea
Level Rise. Science, 302(5644), 434–437.
Rivera, A., Bown, F., Casassa, G., Acuña, C., and
Clavero, J. (2005) Glacier shrinkage and negative mass
balance in the Chilean Lake District (40S). Hydrologi-
cal Sciences Journal, 50(6), 963–974.
Rivera, A., Benham, T., Casassa, G., Bamber, J., and
Dowdeswell, J.A. (2007) Ice elevation and areal
changes of glaciers from the Northern Patagonia Ice-
field, Chile. Global and Planetary Change, 59(1/4), 126–
137.
Roe, G.H., Montgomery, D.R., and Hallet, B. (2002)
Effects of orographic precipitation variations on the
concavity of steady-state river profiles. Geology, 30,
143–146.
Sauber, J., Molnia, B., Carabajal, C., Luthcke, S., and
Muskett, R. (2005) Ice elevations and surface change
on the Malaspina Glacier, Alaska. Geophysical
Research Letters, 32, L23S01.
Scharroo, R., and Visser, P. N. A. M. (1998) Precise orbit
determination and gravity field improvement for the
ERS satellites. Journal of Geophysical Research, 103,
8113–8127.
Schiefer, E., Menounos, B., and Wheate, R. (2007)
Recent volume loss of British Columbian glaciers,
Canada. Geophysical Research Letters, 34, L16503.
References 143
Schlesinger, M. (1968) The interaction of learning and
self-organization in pattern recognition. Cybernetics
and Systems Analysis, 4(2), 66–71.
Schneevoigt, N.J. and Schrott, L. (2006) Linking geo-
morphic systems theory and remote sensing: A con-
ceptual approach to Alpine landform detection
(Reintal, Bavarian Alps, Germany). Geographica Hel-
vetica, 3, 181–190.
Schneevoigt, N.J., van der Linden, S., Thamm, H.-P., and
Schrott, L. (2008) Detecting Alpine landforms from
remotely sensed imagery: A pilot study in the Bavarian
Alps. Geomorphology, 93, 104–119.
Schneider, K. (1998) Geomorphologisch plausible
Rekonstruktion der digitalen Repräsentation von
Geländeoberflächen aus Höhendaten. Unpublished
PhD thesis, University of Zurich, Switzerland, 226
pp. [in German].
Smith, M.J., and Pain, C.F. (2009) Applications of
remote sensing in geomorphology. Progress in Physical
Geography, 33(4), 568–582.
Stearns, L.A., and Hamilton, G.S. (2007) Rapid volume
loss from two East Greenland outlet glaciers quantified
using repeat stereo satellite imagery. Geophysical
Research Letters, 34, L05503.
Surazakov, A.B., and Aizen, V.B. (2006) Estimating vol-
ume change of mountain glaciers using SRTM and
map-based topographic data. IEEE Transactions on
Geoscience and Remote Sensing, 44(10), 2991–2995.
Surazakov, A.B., Aizen, V.B., Aizen, E.M., and Nikitin,
S.A. (2007) Glacier changes in the Siberian Altai
Mountains, Ob river basin, (1952–2006) estimated with
high resolution imagery. Environmental Research Let-
ters, 2, 045017.
Tachikawa, T., Kaku, M., Iwasaki, A., Gesch, D.,
Oimoen, M., Zhang, Z., Danielson, J., Krieger, T.,
Curtis, B., Haase, J. et al. (2011) ASTER Global Digi-
tal Elevation Model Version 2: Summary of Validation
Results. Available at http://www.ersdac.or.jp/GDEM/
ver2Validation/Summary_GDEM2_validation_report_
final.pdf
Tangborn, W. (1999) A mass balance model that uses
low-altitude meteorological observations and the
area-altitude distribution of a glacier. Geografiska
Annaler, Series A, Physical Geography, 81(4), 753–765.
Taschner, S., and Ranzi, R. (2002) Comparing the oppor-
tunities of Landsat TM and Aster data for monitoring
a debris covered glacier in the Italian Alps within the
GLIMS project. Paper presented at IEEE International
Geoscience and Remote Sensing Symposium, Vol. II,
pp. 1044–1046.
Toutin, T. (2002) Three-dimensional topographic map-
ping with ASTER stereo data in rugged topography.
IEEE Transactions on GeoScience and Remote Sensing,
40(10), 2241–2247.
Toutin, T. (2008) ASTER DEMs for geomatic and geo-
scientific applications: A review. International Journal
of Remote Sensing, 29(7), 1855–1875.
144 Digital terrain modeling and glacier topographic characterization
Toutin, Th., Blondel, E., Clavet, D., and Schmitt, C.V.
(2013) Stereo radargrammetry with Radarsat-2 in the
Canadian Arctic. IEEE Transactions on Geoscience and
Remote Sensing, 51(5), 2601–2609, doi: 10.1109/
TGRS.2012.2211605.
van Asselen, S., and Seijmonsbergen, A.C. (2006) Expert-
driven semi-automated geomorphological mapping for
a mountainous area using a laser DTM. Geomorphol-
ogy, 78(3/4), 309–320.
Weber, M., Herrmann, J., Hajnsek, I., and Moreira, A.
(2006) TerraSAR-X and TanDEM-X: Global mapping
in 3D using radar. Paper presented at Proceedings of
the Second International Workshop on ‘‘The Future of
Remote Sensing,’’ Antwerp, Belgium (ISPRS Intercom-
mission Working Group I/V Autonomous Navigation,
36–1/W44), International Society for Photogrammetry
and Remote Sensing, Hanover, Germany.
Weidmann, Y. (2004) Combination of ASTER- and
SRTM-DEMs for the assessment of natural hazards
in the Pamir Tadjikistan. Unpublished thesis, FHBB
Muttenz, Basle [in German].
Wilson, J.P., and Gallant, J.C. (Eds.) (2000) Terrain Anal-
ysis: Principles and Applications. John Wiley and Sons,
New York.
Yadav, R.R., Park, W.K., Singh, J., and Dubey, B. (2004)
Do the western Himalayas defy global warming? Geo-
physical Research Letters, 31, L17201.
Zemp, M., Jansson, P., Holmlund, P., Gärtner-Roer, I.,
Koblet, T., Thee, P., and Haeberli, W. (2010) Reana-
lysis of multitemporal aerial images of Storglaciären,
Sweden (1959–99), Part 2: Comparison of glaciological
and volumetric mass balances. Cryosphere, 4, 345–357.
Zwally, H.J., Schutz, B., Abdalati, W., Abshire, J.,
Bentley, C., Brenner, A., Bufton, J., Dezio, J.,
Hancock, D., Harding, D. et al. (2002) ICESat’s
laser measurements of polar ice, atmosphere, ocean,
and land. Journal of Geodynamics, 34, 405–445.