Content uploaded by Giuseppe Pulighe
Author content
All content in this area was uploaded by Giuseppe Pulighe
Content may be subject to copyright.
Content uploaded by Giuseppe Pulighe
Author content
All content in this area was uploaded by Giuseppe Pulighe
Content may be subject to copyright.
363
European J ournal of Remote Sensi ng - 2013, 46: 363-378
doi: 10.5721/EuJRS20134621
DEM extraction from archive aerial photos: accuracy
assessment in areas of complex topography
Giuseppe Pulighe1*and Francesco Fava2
1Istituto Nazionale di Economia Agraria, via Nomentana 41, 00161 Roma, Italy
2Remote Sensing of Environmental Dynamics Lab., Dep. of Earth and Environmental Sciences,
Università degli Studi di Milano-Bicocca, Piazza della Scienza 1, 20126 Milano, Italy
*Corresponding author, e-mail address: pulighe@inea.it
Abstract
The aim of this study is to analyze the accuracy of a Digital Elevation Model (DEM)
created with photogrammetric techniques from stereoscopic pairs of aerial photos in
areas with complex geomorphologic characteristics. The evaluation of DEM and derived
geomorphometric parameters was conducted by comparison with other standard DEM
products (i.e. TINITALY/01 and ASTER GDEM-V2) and by accuracy assessment based on
Check Points (CPs). The validation process includes the comparison of elevation proles,
the calculation of DEM accuracies, and the evaluation of the effect of slope and aspect on
the DEM accuracy.
The produced DEM accurately represent complex terrain (RMSE = 4.90 m), thus providing
information suitable for local-scale geomorphometric analysis. The obtained accuracy
resulted slightly worse than TINITALY/01 (RMSE = 2.53 m), but signicantly better than
ASTER GDEM (RMSE = 12.95 m). These results conrm that photo-based DEM extraction
can be a very competitive and precise methodology if other expensive high-resolution data
are not accessible.
Keywords: DEM, aerial photos, aerial triangulation, accuracy assessment.
Introduction
Nowadays, the Digital Elevation Model (DEM), which is a 3D digital representation of
an elevation surface over a specied area, is one of the most fundamental requirements
for a large variety of spatial analysis and modeling problems in environmental sciences. It
is used in analyses in ecology, hydrology, agriculture, geology, pedology, geomorphology
and many others, as a means both of explaining processes and of predicting them through
modeling [Schumann et al., 2008; Christoph et al., 2009; Marzolff and Poesen, 2009]. Our
capacity to understand and model these processes depends on the quality of the topographic
data that are available [Jarvis et al., 2004].
DEMs are also a necessary input parameters for: determining the extent of a watershed
Pulighe and Fava DEM extr action from archi ve aerial photos
364
and extracting a drainage network [Tucker et al., 2001], determining the slope and aspect
associated with a geographic region [Kuhni and Pffner, 2001], modeling and planning
for telecommunications [Sawada et al., 2006], orthorectication [Toutin, 2004], preparing
3D simulations, 3D perspectives and ight simulations [Lisle, 2006], agricultural
applications [Pilesjö et al., 2006], studies of landscape dynamics [Mitasova et al., 2005]
and morphometric characterization of volcanoes [Grosse et al., 2012]. As with any other
geospatial dataset, DEMs are produced at a number of spatial scales each of which has its
own cost-effective techniques for data acquisition [Oksanen and Sarjakoski, 2006]. With the
advent of satellite imagery covering the globe, various global datasets of topography have
been produced, of increasingly better resolution. The possible sources of these elevation
data can be Interferometric Synthetic Aperture Radar (InSAR), photogrammetric methods
with space and aerial images, laser scanning using airborne Light Detection and Ranging
(LiDAR) and classical ground survey [Nelson et al., 2009; Hirt et al., 2010]. Each of these
methods have advantages and disadvantages.
The classical eld survey is economic only for small areas (e.g. geoarchaeology). Aerial laser
scanning is detailed and accurate but very expensive, requiring specically designed ights
and intensive elaboration of the raw data [Grosse et al., 2012]. Satellite photogrammetry is
also weather depending and still quite expensive as the very high resolution satellites are
mainly operating in a single image mode. In fact, stereo pairs are frequently collected on
demand. Radar interferometry is weather independent (rainfall may cause some problems),
but time-consuming and quite complex to elaborate [Panagiotis et al., 2008].
Aerial photogrammetry is an accurate and powerful tool in surface model generation,
extracting high resolution DEMs by means of automated image matching procedures
[Fabris and Pesci, 2005]. Very high-resolution aerial imagery are currently available for
modeling more detailed earth surface processes [Jarvis et al., 2004; Marzolff and Poesen,
2009; Prokešová et al., 2010], especially in elds such as hydrology, pedology, landslide
dynamics or geomorphology. In particular, historical aerial photos collected in Italian
archives over the past 60 years [Fabris and Pesci, 2005; Fabris et al., 2011], as well as in
other countries, represent an extensive source of data that support environmental studies,
and in general for planning.
The performance of automated DEM generation were re-evaluated through the use of
professional photogrammetric workstations [Hohle, 2009], that allow important advantages,
for example, faster processing and low processing costs. The methodology of automatic
DEM extraction and orthophoto generation from digital stereo imagery is well established
and extensively described [e.g. Rivera et al., 2005; Pieczonka et al., 2011]. However,
an accuracy assessment of these elevation data is necessary. Inadequate and inaccurate
representations can lead to poor decisions that can negatively impact our environment and the
associated human, cultural, and physical landscape. This is particularly true in classication
or other cartographic modeling applications where elevation, slope and aspect are derived
from DEMs and used with other spatial data [Bolstad and Stowe, 1994]. Still, only a limited
number of studies have addressed the issue of accuracy evaluation of DEM produced by
photogrammetric methods from archive aerial photographs and derived geomorphometric
features in areas with complex topography. The purpose of this paper is to investigate the
quality of DEM created with photogrammetric methods using stereo aerial photos in areas
of complex topography, and its potential use for geomorphometic analysis at local scale.
365
European J ournal of Remote Sensi ng - 2013, 46: 363-378
Specically, the objective of this case study was to answer the following questions:
I) What is the accuracy of the DEM created from archive aerial photos in comparison
with two other DEMs generated with different methodologies?
II) How do slope and aspect inuence DEM accuracy?
Material and methods
Study area
The study area is located about 30 km south of Cagliari, in southern Sardinia, Italy, around
38º 59’ north latitude and 8º 54’ west longitude, and covers an area of 10 km x 5 km. The
elevations range from 4 m up to over 860 m a.s.l., the average slope is 22° and most of
the reliefs are facing south-east. Land cover consists mainly of a mixture of coniferous
forest and mediterranean maquis, with patches of pastures and agricultural areas. The site
is dominated by Quercus ilex L., Quercus suber L., Arbutus unedo L. and Phillyrea L. sp.
with a maximum height of 3 m. Paleozoic Granites (Complesso Granitoide del Sulcis-
Arburese) constitute the bedrock. They are deeply altered and friable, showing the typical
arenization at the surface. Limited Pleistocene and Holocene sediment covers are present
in the lower valley areas. This region is characterized by hilly and undulating terrain which
extends to the coastline, and represent a specic landscape characterized by steep ridges,
high peaks, deep valleys and gorges. Figure 1 depicts the study area and the locations of the
CPs collected from a topographic map with a scale of 1:10.000.
Figure 1 – Location of the study site and the distributions of CPs.
Pulighe and Fava DEM extr action from archi ve aerial photos
366
DEM generation from airphoto stereo pairs
The Digital Elevation Model, called DEM95, is a digital elevation model which we have
developed from 12 archive aerial photos along 2 strips at a mean scale of 1:34.000. The
ight was performed on July 29, 1995 (Tab. 1). The black-and-white photo frames, obtained
from the Italian Military Geographical Institute (IGM), were scanned using Wehrli Raster
Master RM2 photogrammetric scanner at 1200 dpi resolution, resulting in a ground pixel
resolution of 0.7 m. Each frame has associated information relating to the focal length,
lens type and acquisition date. The images were compressed in a TIFF format, without loss
of quality. The images orientation and point extraction procedures were carried out using
Leica Photogrammerty Suite (LPS 9.2) software package.
Table 1 - Main characteristics of the aerial photogrammetric survey.
Characteristics 1995 survey
Aerial photogrammetric camera Wild RC 20
Focal length (mm) 152.83
Flying height (m) 5000
Number of photograms 12
Number of strips 2
Overlap (%) > 60
Scan resolution (dpi) 1200
Ground pixel dimension (m) 0.7
Number of GCPs used 118
Residuals of exterior orientation
rX -0.253
rY 0.384
rZ -0.415
No camera calibration report was available, so the interior camera parameters were estimated
and subsequently rened from the mathematical model generated during aerial triangulation.
Aerial triangulation uses collinearity equations to establish a geometric relationship between the
image, camera and the ground. The project data le was created and the various parameters were
dened. They include, among others, camera model, ground control points, altitude information
such as ying height and average ground elevation, the type of imagery and camera parameters
(such as name, focal length and principle point coordinates), the order of images and strips and
the coordinate system for the control points.
The standard procedure to generate a DEM using automated stereo-correlation process
is based on fundamental steps that consist in interior orientation, exterior orientation
(registration into a dened reference system) and point extraction. The interior orientation
is performed to dene the frame position inside the camera.
The exterior orientation is obtained by two subsequent steps: relative and absolute orientation.
In the rst case the aim is to create the stereoscopic model in an arbitrary relative system
using points common to both images (tie points). A tie point is a point with unknown ground
coordinates, and is visually recognizable in the overlap area between multiple images. Tie
points are used to create a geometric relationship between the images in a project so they
are positioned correctly relative to one another. LPS software uses Automatic tie point
367
European J ournal of Remote Sensi ng - 2013, 46: 363-378
collection to automatically identify and measure tie points across multiple images and strips
of imagery. The absolute orientation is the transformation of the generated stereoscopic
model into an external reference frame dened by the coordinates of several ground control
points, recognized on the images [Fabris and Pesci, 2005]. Ground control points were
collected from orthoimages TerraItaly 2006 with 0.5 m spatial resolution, while elevation
coordinates necessary for height stabilization in the block bundle adjustment were collected
from DEM 10 m resolution. The coordinate reference system is Gauss-Boaga/Roma40.
Images and DEM are available at: http://www.sardegnageoportale.it/index.html. Ground
control points are distributed over the whole area to ensure an even coverage of the image
and to avoid clustering effects [Rocchini et al., 2012].
DEM95 was resampled to 5 m resolution, elevations are ellipsoidal heights referenced to
the reference systems Gauss-Boaga/Roma40. DEMs created by photogrammetric technique
occasionally contain blunders such as irregularly gridded data, mistagging of tops and
depression. These spikes, holes and other noises caused by automatic DEM extraction can
be corrected by editing the raster data values and replacing them with meaningful values.
The Spike/Well check tool in ERDAS imagine was applied to identify pixels that are out of
range with its surrounding neighbours. It also determines whether there are large spikes and
wells in the data. Visual interpretation showed no signicant gross errors in the maximum
and minimum values of elevations contained in the DEM95.
DEM from TINITALY/01
The most popular data sources for the creation of DEM are the digitized contours derived from
topographic maps. In this study we used a new digital elevation model, named TINITALY/01,
that is currently the most accurate DEM covering the whole Italian territory [Tarquini et al.,
2012]. The DEM was released by the Italian Ministry of the Environment and Territory (the
DIGITALIA project). The DEM was obtained from heterogeneous vector datasets, mostly
consisting in elevation contour lines and elevation points from several sources. The input vector
database was carefully cleaned up to obtain a seamless Triangulated Irregular Network (TIN)
derived by using the DEST algorithm (Determination of Earth Surface Structures) [Favalli
and Pareschi, 2004; Tarquini et al., 2007]. The adopted coordinate systems for TINITALY/01
is Universal Transverse Mercator/World Geodetic System 1984 (UTM-WGS84). The 32 zone
(for Western Italy) and the 33 zone (for Eastern Italy) is available in grid format as a 10 m cell
size grid. Tarquini et al. [2007] carried out a comprehensive assessment of the accuracy of the
TINITALY/01 DEM, nding a root mean square error in elevation (RMSE) between 0.8 and
6.0 m. The TINITALY/01 DEM is available for scientic purposes on the basis of a research
agreement at: http://tinitaly.pi.ingv.it/.
ASTER GDEM version 2
Advanced Spaceborne Thermal Emission and Reection Radiometer (ASTER, http://
asterweb.jpl.nasa.gov/) Global Digital Elevation Model (GDEM) version 2 was released on
October 2011 by the National Aeronautics and Space Administration (NASA) and Japan’s
Ministry of Economy, Trade and Industry (METI). ASTER GDEM version 2 (GDEM
V2) is obtained from data collected by the sensor Visible Near Infrared (VNIR), using
bands 3N (nadir-viewing) and 3B (backward-viewing) with a resolution of 15 m with a
swath 60 km wide. ASTER GDEM V2 is organized according to a regular grid of 30x30 m
Pulighe and Fava DEM extr action from archi ve aerial photos
368
UTM-WGS84 cartographic coordinates, while the orthometric heights were determined by
using the corresponding ellipsoidal Earth Gravitational Model (EGM 96) geoid [ASTER
Validation Team, 2011]. ASTER GDEM is made to order following the selection of tiles
through the Earth Observing System (EOS) Data Gateway.
The overall vertical accuracy of ASTER elevations is specied to vary between 8.68 m
and 18.31 m [ASTER Validation Team, 2011]. Generally error increases as the topography
become rougher. This ASTER product is available at no charge for any user pursuant to an
agreement between METI and NASA.
Check Points
In order to provide the benchmark to infer vertical accuracy, 3D coordinates of 1033 check
points have been collected from topographic map with a scale of 1:10000. These points
are homogeneously distributed in the area and the number is enough to guarantee error
control reliability. Check points have a particular importance as a reference or planimetric
and elevation, which are always located in areas which are presumed to be stable over the
last decades, and have been reported with particular care. The elevations H are ground truth
ellipsoidal heights. The vertical accuracies of the CPs are <±1.8 m, and is satisfactory for
the control objectives (matching the criteria of 1:10000 topographic mapping) since the
standard deviation expected of the DEM is twice the mean error of CPs.
Accuracy assessment
The magnitude of absolute and relative errors of DEMs data has been examined. The
absolute accuracy is a measurement of the error between a DEM and the coordinates of the
terrain. Accuracy is evaluated by indices such as the absolute mean error (AME), standard
deviation (SD) and the root mean square error (RMSE), whereas shape reliability is
evaluated through statistical analysis of a parameter set characterizing the spatial properties
of a surface such as slope and aspect. Absolute accuracy is expressed as the vertical RMSE,
who is an overall error indicator that takes into account both random and systematic errors
introduced during the data generation process. The relative vertical accuracy is especially
important for derivative products that make use of the local differences among adjacent
elevation values, such as slope and aspect. A DEM with good relative accuracy is one that
models the shape and dimensions of the terrain accurately, but may not necessarily be
accurately registered to real geographic coordinates. The relative accuracy is expressed
as the standard deviation of the vertical error [Jung Hum Yu, 2011]. The accuracy of the
DEM95 is determined as the difference between the CPs and DEM95 and is denoted as CP
- DEM95; similarly, the accuracy of the ASTER and TINITALY is denoted as CP - ASTER
and CP - TINITALY. In order to describe and compare the elevation distributions in each
DEM, elevations at the locations of CPs have been extracted from all DEMs and compared
with the elevations of CPs to determine several descriptive statistic measures.
The process is as follows: First of all, ASTER and TINITALY has been transformed in
Gauss-Boaga/Roma40 reference system using Traspunto software package, the most
widely software package used in Italy for the automatic transformation of coordinates
between reference systems [Travaglini, 2004]. After that, the DEMs have been clipped to
the common study area, in order to have equal areas for all DEMs and eliminate bias due to
the edge effect. Subsequently, all grids have been overlaid with CPs using Intersect Point
369
European J ournal of Remote Sensi ng - 2013, 46: 363-378
Tool from Hawth’s Analysis Tools for ArcGIS [Beyer, 2004]. Finally, the DEM attributes
(i.e. elevation, aspect and slope) have been extracted. All data points with their respective
attributes have been organized in a spreadsheet table and analyzed by using the software
IBM SPSS Statistics 19. Considering that the error possibly correlates with the elevation, the
values of the CPs have been taken as the control variable, and carry out partial correlations
between the errors of the DEMs. In fact, neither data scatter in the linear regression models
nor vertical accuracy constitute sufcient criteria to achieve a full denition of DEM error
because the error is expressed globally [Gonga-Saholiariliva et al., 2011]. Slope gradient
and aspect are two of the most basic parameters affecting morphological and hydrological
problems, and represents one set of the recommended variables to analyse distribution and
concentration of certain spatial objects [Blaschke and Strobl, 2003; Moreno et al., 2003].
These have been computed in order to assess how well topographic details are represented
for all DEMs. In order to ensure compatibility and comparability among of dataset across
space, the grids could not be used with their original characteristics, then slope and aspect
have been resampled with a grid size of 30 meters. While this may have biased slope and
aspect evaluations by potentially inating accuracies for areas of complex terrain, it also
limits the effects of horizontal registration error [Bolstad and Stowe, 1994].
Results
Elevation prole analysis
Qualitative assessment of DEM95 is executed by comparing the elevation proles of the
three DEMs. The DEMs are depicted as elevation curves according to the directions of
proles North-South and West-East for the study area. The proles have been chosen based
on the area covering both low-gradient and high-gradient terrains (Fig. 2).
Figure 2 - 3D perspective view of the study area. DEM was draped by the mosaic image. Red lines
shown transects used for plotting elevation proles.
As shown in Figure 3, the black, blue and red curves represent DEM95, TINITALY and
ASTER, respectively.
Pulighe and Fava DEM extr action from archi ve aerial photos
370
Figure 3 - Prole comparisons between elevations of DEM95, ASTER and DEM
TINITALY: (a) North - South elevation prole, (b) West - East elevation prole.
In the prole North-South (Fig. 3a), length 4000 m, the three curves can be divided into
three parts: in the left part, the ASTER is lower than DEM95 and TINITALY; in the middle
part, the curves of DEM95 and TINITALY uctuate along that of the ASTER; in the right
part, the ASTER is for small sections higher than both DEM95 and TINITALY. This is
probably due to the increased geomorphology complexity of the northern prole. ASTER
curve has some evident uctuations, is generally lower than DEM95 and TINITALY, that
are relatively close and evidently differ, due to the higher accuracy of these. In the prole
West-East (Fig. 3b), length 8800 m, the differences are less marked and the ASTER is
generally close to but lower than the others DEMs. From the curves of the different data
types for the two proles, it can be seen that the trend of the DEM95 and TINITALY have
similar characteristics.
371
European J ournal of Remote Sensi ng - 2013, 46: 363-378
Assessment of relative and absolute accuracy
Descriptive statistics of elevation values of CPs and of the three DEMs are reported in Table 2.
Table 2 - Descriptive statistics of CPs and the three DEMs.
Data Mean (m) Min (m) Max (m) SD (m)
CP 350.70 9.20 864.70 252.17
DEM95 347.25 8.00 857.00 251.05
TINITALY 349.68 9.20 860.08 251.65
ASTER 344.88 9.00 837.00 249.71
SD: Standard Deviation
The CPs are equally distributed among plains, hills and moderate relief at medium elevation
and the TINITALY is a little closer to the CP than the other DEMs. A synthesis of the results
of the accuracy assessment is given in Table 3 which shows the values of min and max
and mean error, root mean square error, standard deviation, standard error and condence
interval (CI = 95%). As shown in the Table 3, the RMSE of DEM95 elevation is about twice
to that of TINITALY, much lower than the one of ASTER GDEM.
Table 3 - Statistical differences between CPs and the respective points in the three DEMs. (AME:
mean absolute error; Min: Minimum; Max: Maximum; Std err: Standard error; SD: Standard
Deviation; CI: Condence Interval for SD (95%); RMSE: Root Mean Square Error).
Data AME
(m)
Min
(m)
Max
(m) Std err (m) SD
(m)
CI
(95%)
RMSE
(m)
CP - DEM95 3.44 -15.15 28.25 0.14 4.42 ±0.27 4.90
CP - TINITALY 1.02 -5.39 26.92 0.07 2.31 ±0.14 2.53
CP - ASTER 5.82 -98.80 50.60 0.36 11.58 ±0.71 12.95
The error distribution frequencies of DEM95, TINITALY and ASTER are illustrated in
Figure 4.
Figure 4 – Error distribution frequencies of (a) CP - DEM95, (b) CP - TINITALY, (c) CP - ASTER.
Superimposed curve represents the normal distribution.
Pulighe and Fava DEM extr action from archi ve aerial photos
372
The overall range of the error distribution for ASTER elevation is wider than those for DEM95
and TINITALY and the most frequent errors are positives. The characteristics of the error
distributions of DEM95 and TINITALY resemble each other, and the frequencies of positive
errors are greater than those of the negative errors for both DEM95 and TINITALY, which
indicates a small positive bias for both models.
Error correlation analysis
Considering that the error possibly correlates with the elevation, the values of the CPs
have been taken as a control variable, and carry out partial correlations between the errors
of the DEMs. In the comparison DEM95 - TINITALY the error correlation is r = 0.43
(p<0.001). As shown in Figure 5a, the errors of both DEM95 and TINITALY are mainly
distributed from -10 to 10 m. In the comparison DEM95 - ASTER, the error correlation
is r = 0.37 (p<0.001). The errors of both DEM95 and ASTER are mainly distributed from
- 30 to 30 m (Fig. 5b). Therefore, the errors of the TINITALY and ASTER only have weak
correlation.
Figure 5 – Scatter plots between the errors of the (a) DEM95 - TINITALY and (b) DEM95 -
ASTER.
Geomorphometric analysis
The geomorphometric analysis revealed a signicant decrease in accuracy of all DEMs data
when measurements are performed on terrain characterized by slope values greater than 8°
(Tab. 4).
The relationship between the AME and aspect is shown in Figure 6. The aspect is divided
into eight directions. The AME in each direction is the value in the 45° range centred on
the specied direction. As the Figure 6 shows, the relationships for the DEM95, TINITALY
and ASTER completely differ in terms of magnitude. For DEM95 and ASTER, the AME
in each direction is almost the same and forms a near circle. For TINITALY, the AME
for the southern, western and southeast direction is remarkably lower than it is for other
directions.
373
European J ournal of Remote Sensi ng - 2013, 46: 363-378
Table 4 - Analysis of discrepancies between DEMs and CPs data with different slope values. (Std
err: Standard error; SD: Standard Deviation; Min: Minimum; Max: Maximum; CI: Condence
Interval for SD (95%)).
Area
Data
DEM95 TINITALY ASTER
Slope Slope Slope
< 4° 4°< x <8° > 8° < 4° 4°< x <8° > 8° < 4° 4°< x <8° > 8°
Mean (m) 2.90 3.28 3.83 0.22 0.92 1.48 3.56 4.68 7.16
Std err (m) 0.18 0.24 0.23 0.05 0.11 0.12 0.43 0.59 0.57
SD (m) 3.24 3.35 5.23 0.87 1.51 2.89 6.57 8.82 13.74
Min (m) -5.82 -9.05 -15.15 -5.14 -2.67 -5.39 -25.50 -20.10 -98.80
Max (m) 11.50 11.61 28.25 6.11 8.45 26.92 40.00 50.60 42.70
CI (95%) 0.36 0.48 0.45 0.10 0.22 0.24 0.84 1.17 1.13
Figure 6 - Relationships between aspect and absolute
mean error.
Discussion
Based on the results obtained in this study, the generation of DEM from aerial photos in
areas of complex topography has showed that elevation RMSE range up to ± 4.90 m.
Specically, the RMSE values obtained for DEM95 are comparable to those obtained by
Walstra et al. [2007] and Zanutta et al. [2006], that used archive aerial photos for DEM
generation. Previous studies report that the vertical accuracy of the DEM is a function of
photo scale and is estimated as 1/9000th of the ying height of the aircraft carrying the
camera system [Maune et al., 2001; Hapke, 2005]. In this study, the ying height of the
photographs was 5000 m and this results in a vertical error of 0.55 m. However, this accuracy
may be good enough for ne scale applications over large areas or where limitations on the
accessibility of remote areas makes the collection of ground truth data impossible [Oksanen
Pulighe and Fava DEM extr action from archi ve aerial photos
374
and Sarjakoski, 2006; Hobi and Ginzler, 2012].
The RMSE values obtained for the other DEMs are consistent with the reported ofcial
accuracy (i.e. TINITALY: 0.8 m < RMSE < 6.0 m – [Tarquini et al., 2007]; ASTER: 6.1
m < RMSE < 15.1 m – [ASTER validation Team, 2011]), supporting the robustness of the
validation dataset.
DEM95 had a lower accuracy than TINITALY, probably due to the fact that the area contains
a great deal of vegetation, for which a lower accuracy is to be expected as compared with
the latter, obtained from heterogeneous vector datasets. The errors of the TINITALY and
ASTER have only weak correlation. One interpretation of this nding is that the poor
correlation relates to the different resolutions of the original datasets.
For all DEMs, greater error values were associated with rugged terrain, while smaller errors
were associated with plain, suggesting that terrain characteristics such as slope and aspect
can inuence DEMs accuracy [Gorokhovich and Voustianiouk, 2006].
Different results were obtained investigating the effects of land slope and aspect on the
DEMs. As expected, the vertical error increased with surface slope. The magnitude of SD
was about twice for terrains with slope values exceeding 8° compared to areas where slope
values are less than 4° in DEM95 (3.2 m vs 5.23 m) in TINITALY (0.87 m vs 2.89 m) and
ASTER (6.57 m vs 13.74 m). This result demonstrates that there is a correlation between
the elevation error and land slope.
Also terrain aspect was found to have a relative inuence on the magnitude of errors in the
DEMs. The highest magnitude of errors was observed for measurements made on ASTER,
the lowest magnitude of errors was observed in TINITALY. The error magnitude did not
vary considerably with aspect in DEM95 and ASTER, while in TINITALY an irregular
shape of the error was observed, with the lowest errors on slopes facing south (S), southeast
(SE) and west (W), due to the better accuracy, while DEM95 did not show trends. In this
case aspect did not particularly inuenced the absolute mean error.
The accuracy of DEM95 is affected by the quality and parameters of the aerial photos taken,
the height of the ight, the topographic characteristics of the area, the status and type of the
vegetation cover as well as the human factor during the photogrammetric evaluation. The
potential limitations of these data could be a lack of sensitivity to the variation in terrain
surface due to the presence of trees and areas of steeper slopes.
Overall, the results obtained in this study indicate that the accuracy DEM95 is comparable
to DEMs extracted with photogrammetric techniques from previous studies, and strongly
encourage the use of archival materials to improve the geomorphological studies.
Developments on data acquisition and processing software allows the reuse and enhancement
of historical data for DEM generation, leading to better and faster results.
Conclusions
In this article we describe the accuracy of DEM95 created with photogrammetric techniques
from archive aerial photos, and we compare it to ASTER GDEM version 2 and DEM
TINITALY. Check points collected from digital cartography were used for the vertical
accuracy control.
According to the results achieved, it is possible to state that photo-based DEM95 can be
used for computation of terrain attributes such as slope and aspect, and is suitable for a range
of environmental mapping tasks involving the use of DEM with a grid step of 5-20 m. This
375
European J ournal of Remote Sensi ng - 2013, 46: 363-378
implies that an automated DEM extraction of ne toposcale can be an efcient method for
analysis of hydrological modeling, soil properties and vegetation pattern. These evaluations
showed that photo-based DEMs are very competitive, very cheap and affordable, readily
available and relatively precise, especially when other expensive high-resolution data (e.g.
LiDAR) are not accessible.
Acknowledgements
The authors would like to thank Prof. Roberto Scotti of Nuoro Forestry School, Dipartimento
di Agraria, Università di Sassari, for provision of aerial photos of 1995.
References
ASTER Validation Team (2011) - ASTER Global Digital Elevation Model Version 2 –
Summary of Validation Results. Published by The ASTER GDEM Validation Team:
METI, NASA and USGS in cooperation with NGA and other colloborators. August
2011, 27 pages. Available online: http://asterweb.jpl.nasa.gov/gdem.asp (last accessed
28/01/2012).
Beyer H.L. (2004) - Hawth’s Analysis Tools for ArcGIS. Available online: http://www.
spatialecology.com/htools/tooldesc.php (last accessed 29/01/2012).
Blaschke T., Strobl J. (2003) - Dening landscape units through integrated morphometric
characteristics. In: Buhmann, E., Ervin, S. (Eds.), Landscape Modeling: Digital
Techniques for Landscape Architecture. Wichmann-Verlag, Heidelberg, pp. 104-113.
Bolstad P.V., Stowe T. (1994) - An evaluation of DEM accuracy: Elevation, slope and
aspect. Photogrammetric Engineering and Remote Sensing, 60 (1): 1327-1332.
Christoph S., Olaf B., Bernhard E. (2009) - Combining digital elevation data (SRTM/ASTER),
high resolution satellite imagery (Quickbird) and GIS for geomorphological mapping: a
multi-component case study on Mediterranean karst in Central Crete. Geomorphology,
112: 106-121. doi: http://dx.doi.org/10.1016/j.geomorph.2009.05.010.
Fabris M., Pesci A. (2005) - Automated DEM extraction in digital aerial photogrammetry:
precisions and validation for mass movement monitoring. Annals of Geophysics, 48 (6):
973-988. doi: http://dx.doi.org/10.4401/ag-3247.
Fabris M., Menin A., Achilli V. (2011) - Landslide displacement estimation by archival
digital photogrammetry. Italian Journal of Remote Sensing, 43 (2): 23-30. doi: http://
dx.doi.org/10.5721/ItJRS20114322.
Favalli M., Paresci M.T. (2004) - Digital elevation model construction from structured
topographic data: the DEST algorithm. Journal of Geophysical Research, 109: F04004.
doi: http://dx.doi.org/10.1029/2004JF000150.
Gonga-Saholiariliva N., Gunnell Y., Petit C., Mering C. (2011) - Techniques for quantifying
the accuracy of gridded elevation models and for mapping uncertainty in digital terrain
analysis. Progress in Physical Geography, 35 (6): 739-764. doi: http://dx.doi.org/10.11
77/0309133311409086.
Gorokhovich Y., Voustianiouk A. (2006) - Accuracy assessment of the processed SRTM-
based elevation data by CGIAR using eld data from USA and Thailand and its relation
to the terrain characteristics. Remote Sensing of Environment 104: 409-415. doi: http://
dx.doi.org/10.1016/j.rse.2006.05.012.
Grosse P., van Wyk de Vries B., Euillades P.A., Kervyn M., Petrinovic I.A. (2012)
Pulighe and Fava DEM extr action from archi ve aerial photos
376
- Systematic morphometric characterization of volcanic edices using digital
elevation models. Geomorphology, 136 (1): 114-131. doi: http://dx.doi.org/10.1016/
j.geomorph.2011.06.001.
Hapke C.J. (2005) - Estimation of regional material yield from coastal landslides based on
historical digital terrain modeling. Earth Surface Processes and Landforms, 30: 679-
697. doi: http://dx.doi.org/10.1002/esp.1168.
Hirt C., Filmer M.S., Featherstone W.E. (2010) - Comparison and validation of recent
freely-available ASTER-GDEM ver1, SRTM ver4.1 and GEODATA DEM-9S ver3
digital elevation models over Australia. Australian Journal of Earth Sciences, 57 (3):
337-347. doi: http://dx.doi.org/10.1080/08120091003677553.
Hobi M.L., Ginzler C. (2012) - Accuracy Assessment of Digital Surface Models Based on
WorldView-2 and ADS80 Stereo Remote Sensing Data. Sensors, 12 (5): 6347-6368. doi:
http://dx.doi.org/10.3390/s120506347.
Hohle J. (2009) - DEM Generation Using a Digital Large Format Frame Camera.
Photogrammetric Engineering & Remote Sensing, 75 (1): 87-93.
Jarvis A., Rubiano J., Nelson A., Farrow A., Mulligan M. (2004) - Practical use of
SRTM data in the tropics: Comparisons with digital elevation models generated from
cartographic data. Working Document: Centro Internacional de Agricultura Tropical
(CIAT), 198: 32 p.
Jung Hum Yu (2011) - Interferometric Synthetic Aperture Radar and Radargrammetry
for Accurate Digital Elevation Model Generation in New South Wales, Australia. PhD
thesis, The University of New South Wales, Sydney NSW 2052, Australia.
Kuhni A., Pffner O.A. (2001) - The relief of the Swiss Alps and adjacent areas and its relation
to lithology and structure: topographic analysis from a 250-m DEM. Geomorphology,
41 (4): 285-307. doi: http://dx.doi.org/10.1016/j.bbr.2011.03.031.
Lisle R. J. (2006) - Google Earth: a new geological resource. Geology Today, 22: 29-32.
doi: http://dx.doi.org/10.1111/j.1365-2451.2006.00546.x.
Marzolff I., Poesen J. (2009) - The potential of 3D gully monitoring with GIS using high-
resolution aerial photography and a digital photogrammetry system. Geomorphology,
111 (1-2): 48-60. doi: http://dx.doi.org/10.1016/j.geomorph.2006.11.003.
Maune D.F., Kopp S.M., Crawford C.A., Zervas C.E. (2001) - Introduction: digital elevation
models. In Digital Elevation Model Technologies and Applications: The DEM Users
Manual, Maune DF (ed.). American Society of Photogrammetry and Remote Sensing:
Bethesda, Maryland; pp. 1-34.
Mitasova H., Overton M., Harmon R.S. (2005) - Geospatial analysis of a coastal sand
dune eld evolution: Jockey’s Ridge, North Carolina. Geomorphology, 72: 204-221.
doi: http://dx.doi.org/10.1016/j.geomorph.2005.06.001.
Moreno M., Levachkine S., Torres M., Quintero R. (2003) - Geomorphometric analysis of
raster image data to detect terrain ruggedness and drainage density. Lecture Notes in
Computer Science, 2905: 643-650. doi: http://dx.doi.org/10.1007/978-3-540-24586-5_79.
Nelson A., Reuter H.I., Gessler P. (2009) - DEM production methods and sources. In: Hengl
T., Reuter H. (Eds.), Geomorphometry, Concepts, Software, Applications. Developments
in Soil Science, Vol. 33. Elsevier, Oxford.
Oksanen J., Sarjakoski T. (2006) - Uncovering the statistical and spatial characteristics of
ne toposcale DEM error. International Journal of Geographical Information Science,
377
European J ournal of Remote Sensi ng - 2013, 46: 363-378
20 (4): 345-369. doi: http://dx.doi.org/10.1080/13658810500433891.
Pieczonka T., Bolch T., Buchroithner M.F. (2011) - Generation and evaluation of
multitemporal digital terrain models of the Mt. Everest area from different optical
sensors. ISPRS Journal of Photogrammetry and Remote Sensing, 66 (6): 927-940. doi:
http://dx.doi.org/10.1016/j.isprsjprs.2011.07.003.
Pilesjö P., Persson A., Harrie L. (2006) - Digital elevation data for estimation of potential
wetness in ridged elds - Comparison of two different methods. Agricultural Water
Management, 79 (3): 225-247. doi: http://dx.doi.org/10.1016/j.agwat.2005.02.017.
Prokešová R., Kardoš M., Medvedová A. (2010) - Landslide dynamics from high-
resolution aerial photographs: a case study from the Western Carpathians, Slovakia.
Geomorphology, 115: 90-101. doi: http://dx.doi.org/10.1016/j.geomorph.2009.09.033.
Rivera A., Casassa G., Bamber J., Kääb A. (2005) - Ice-elevation changes of Glaciar Chico,
southern Patagonia, using ASTER DEMs, aerial photographs and GPS data. Journal of
Glaciology, 51 (172): 105-112. doi: http://dx.doi.org/10.3189/172756505781829557.
Rocchini D., Metz M., Frigeri A., Delucchi L., Marcantonio M., Neteler M. (2012) - Robust
rectication of aerial photographs in an Open Source environment. Computers &
Geosciences, 39 (2): 145-151. doi: http://dx.doi.org/10.1016/j.bbr.2011.03.031.
Sawada M., Cossette D., Wellar B., Kurt T. (2006) - Analysis of the urban/rural
broadband divide in Canada: Using GIS in planning terrestrial wireless deployment.
Government Information Quarterly, 23 (3-4): 454-479. doi: http://dx.doi.org/10.1016/
j.giq.2006.08.003.
Schumann G., Matgen P., Cutler M.E.J., Black A., Hoffmann L., Pster L. (2008) -
Comparison of Remotely Sensed Water Stages from Li-DAR, Topographic Contours and
SRTM. Journal of Photogrammetry and Remote Sensing, 63: 283-296. doi: 10. 1016/
j.isprsjprs.2007.09.004.
Tarquini S., Isola I., Favalli M., Mazzarini F., Bisson M., Pareschi M., Boschi E. (2007)
- TINITALY/01: a new Triangular Irregular Network of Italy. Annals Of Geophysics, 50
(3): 407-425.
Tarquini S., Vinci S., Favalli M., Doumaz F., Fornaciai A, Nannipieri L. (2012) - Release
of a 10-m-resolution DEM for the Italian territory: Comparison with global-coverage
DEMs and anaglyph-mode exploration via the web. Computers & Geosciences, 38 (1):
168-170. doi: http://dx.doi.org/10.1016/j.cageo.2011.04.018.
Toutin T. (2004) - Review article: Geometric processing of remote sensing images: models,
algorithms and methods. International Journal of Remote Sensing, 25 (10): 1893-1924.
doi: http://dx.doi.org/10.1080/0143116031000101611.
Tsombos P.I., Nikolakopoulos K.G., Lathourakis G. (2008) - DEM from Cartosat data
and comparison to DEM from other sources. Proc. SPIE 7106, Sensors, Systems,
and Next-Generation Satellites XII, 71061C (October 2, 2008). doi: http://dx.doi.
org/10.1117/12.800019.
Travaglini D. (2004) - Trasformazioni tra sistemi di coordinate: software disponibili,
limiti e potenzialita. Forest@ 1 (2): 128-134. doi: http://dx.doi.org/10.3832/efor0228-
0010128.
Tucker G.E., Catani F., Rinaldo A., Bras R.L. (2001) - Statistical analysis of drainage
density from digital terrain data. Geomorphology, 36 (3-4): 187-202. doi: http://dx.doi.
org/10.1016/S0169-555X(00)00056-8.
Pulighe and Fava DEM extr action from archi ve aerial photos
378
Walstra J., Dixon N., Chandler J.H. (2007) - Historical aerial photographs for landslide
assessment: two case histories. Quarterly Journal of Engineering Geology &
Hydrogeology, 40 (4): 315-332. doi: http://dx.doi.org/10.1144/1470-9236/07-011.
Zanutta A., Baldi P., Bitelli G., Cardinali M., Carrara A. (2006) - Qualitative and
quantitative photogrammetric techniques for multi-temporal landslide analysis. Annals
of Geophysics, 49 (4-5): 1067-1080. doi: http://dx.doi.org/10.4401/ag-3117.
Received 12/09/2012, accepted 20/02/2013