Ground-based and UAV-Based photogrammetry: A multi-scale, high-resolution mapping tool for Structural Geology and Paleoseismology

Full text

Review article
Ground-based and UAV-Based photogrammetry: A multi-scale, high-
resolution mapping tool for structural geology and paleoseismology
Sean P. Bemis
a
,
*
, Steven Micklethwaite
b
, Darren Turner
c
, Mike R. James
d
, Sinan Akciz
e
,
Sam T. Thiele
b
, Hasnain Ali Bangash
b
a
Department of Earth and Environmental Sciences, University of Kentucky, 101 Slone Research Building, Lexington, KY 40506, USA
b
CET (M006), School of Earth and Environment, The University of Western Australia, 35 Stirling Highway, Crawley, WA 6009, Australia
c
School of Land and Food, University of Tasmania, Hobart, Tasmania 7001, Australia
d
Lancaster Environment Centre, Lancaster University, Lancaster LA1 4YQ, UK
e
Department of Earth, Planetary, and Space Sciences, University of California, Los Angeles, Los Angeles, CA 90095, USA
article info
Article history:
Received 21 May 2014
Received in revised form
13 October 2014
Accepted 14 October 2014
Available online 27 October 2014
Keywords:
Photogrammetry
Structural geology
Neotectonics
3D surface modelling
UAVs
Structure-from-Motion
abstract
This contribution reviews the use of modern 3D photo-based surface reconstruction techniques for high
fidelity surveys of trenches, rock exposures and hand specimens to highlight their potential for paleo-
seismology and structural geology. We outline the general approach to data acquisition and processing
using ground-based photographs acquired from standard DSLR cameras, and illustrate the use of similar
processing approaches on imagery from Unmanned Aerial Vehicles (UAVs). It is shown that digital map
and trench data can be acquired at ultra-high resolution and in much shorter time intervals than would
be normally achievable through conventional grid mapping. The resulting point clouds and textured
models are inherently multidimensional (x,y,z, point orientation, colour, texture), archival and easily
transformed into orthorectified photomosaics or digital elevation models (DEMs). We provide some
examples for the use of such techniques in structural geology and paleoseismology while pointing the
interested reader to free and commercial software packages for data processing, visualization and 3D
interpretation. Photogrammetric models serve to act as an ideal electronic repository for critical outcrops
and observations, similar to the electronic lab book approach employed in the biosciences. This paper
also highlights future possibilities for rapid semi-automatic to automatic interpretation of the data and
advances in technology.
©2014 Elsevier Ltd. All rights reserved.
1. Introduction
High-resolution three dimensional (3D) data capture is required
at all scales in the geosciences, from hand specimen to landscapes,
and a range of tools are available for addressing different portions
of the scale spectrum (e.g., McCaffrey et al., 2005). In particular,
recent advances in high-resolution digital 3D data collection are
dominated by active source sensors, predominantly based upon
laser scanning technologies (e.g., LiDAR), which measure distance
to a target based upon the travel time of reflected light (e.g.,
Hodgetts, 2013). However, a new development in high-resolution
3D data collection exploits a very common and widely accessible
passive imaging source edigital photography. Geoscientists have
long utilized the 3D information available through
photogrammetric techniques; most notably the ability to visualize
the Earth's surface and extract topographic data from stereo aerial
photographs (e.g., Birdseye, 1940; Eardley, 1942). With the funda-
mental principles of photogrammetry now combined with robust
algorithms from the computer vision community, collections of
overlapping photographs can be automatically processed to rapidly
extract the relative 3D coordinates of millions of surface points
(Lowe, 2004; Snavely et al., 2008a, 2008b, 2006). Therefore, the
only specialized resource required for acquisition of 3D data
through photogrammetric techniques is access to suitable software
which depending on computer skills and requirements, is available
through both commercial and open-source options (Table 1).
The limited infrastructure requirements of modern photo-
grammetric techniques present a wide range of opportunities for
geoscience research and education. In the most basic form, the raw
data consist of only digital photographs, which can be collected
with any commonly available digital camera, including those on
smartphones and tablets. As such, the technique facilitates rapid
*Corresponding author.
E-mail address: sean.bemis@uky.edu (S.P. Bemis).
Contents lists available at ScienceDirect
Journal of Structural Geology
journal homepage: www.elsevier.com/locate/jsg
http://dx.doi.org/10.1016/j.jsg.2014.10.007
0191-8141/©2014 Elsevier Ltd. All rights reserved.
Journal of Structural Geology 69 (2014) 163e178

collection of large amounts of data in remote settings where
portability and efficiency may be critical. Because the source data
are photographs, the derivative 3D data can readily be coloured or
draped with the source photography to produce realistic 3D models
of the feature of interest, which can then be exported to 3D visu-
alization environments for analysis. In addition, with free viewers
for many 3D formats (including the ubiquitous Adobe Reader for 3D
PDFs), models can be widely shared for exploration and enhanced
understanding of the 3D nature of many geoscience examples. 3D
photogrammetric models and their accompanying digital photo-
graphs are inherently archival, easily shared and provide a record
that is faithful to the primary observations, thereby allowing future
generations of geoscientists to extract additional 3D and oriented
data, or reinterpret the outcrop/samples.
The purpose of this review is to introduce these easy to use
photogrammetric techniques to the general structural geology and
paleoseismology communities while illustrating a range of appli-
cations. A variety of 3D data collection tools are now available to the
geoscientist, including a range of laser scanning options and
traditional surveying methods, and each method has different ad-
vantages/disadvantages in terms of resolution, scalability, porta-
bility, and computer processing requirements. Photo-based 3D
reconstruction techniques can provide a comparable resolution to
laser scanning tools with a significant reduction in cost, infra-
structure, and processing requirements. To assist with adoption of
such techniques, we provide an overview of the technical basis
along with workflows and discussion of best practices for photo-
graph collection that will provide optimal results.
2. Photogrammetry in the geosciences
For most of the 20th century, photogrammetry principles
implemented through stereoscopic instruments were the primary
means of the construction of topographic maps (e.g., Birdseye,
1940)ea critical element of traditional field-based geologic
studies. Furthermore, photogrammetry has been directly employed
for many years in the geosciences through stereoscopic viewing
and analysis of overlapping pairs of aerial photographs (e.g.,
Eardley, 1942; Pillmore, 1964). This stereoscopic viewing provides
the researcher with the ability to visualize and map a study area
remotely and from a perspective that is impossible to attain in the
field. These photogrammetric techniques were adapted for working
at the outcrop scale for simple visualization purposes by taking
photo pairs of the outcrop of interest for later viewing through a
mirror stereoscope (Kuenen, 1950). This approach is valuable for a
more representative visualization of a site once the researcher
returns to the office, but requires tedious procedures to enable
extraction of data that are fully 3D and oriented (Hagan, 1980).
With the advent of modern digital camera technology, re-
strictions around the number of photos that can be collected have
been relaxed and picture quality can be quickly and easily assessed
in the field. Now, the greater limitation lies in achieving optimal
camera positioning relative to the object of interest, whether due to
vegetation, topography, objective hazards, etc. However, even this
limitation is being reduced through the use of digital photography
from balloons, kites, and UAVs, enabling drastically improved
synoptic views from overhead (e.g., Smith et al., 2009; Niethammer
et al., 2012; Stumpf et al., 2013).
3. Principles and capability of photo-based 3D reconstruction
3.1. Basic principles
New methods of photo-based 3D reconstruction reflect the
evolution of photogrammetry from a highly specialized technique,
requiring expensive software and restrictive image collection re-
quirements, to a user-friendly and scalable methodology. Funda-
mentally, photogrammetry works on the basis that, from two
overlapping photographs, it is possible to calculate the unique
three-dimensional (3D) location of a set of given points shared in
Table 1
Examples of open source and commercial software for photo-based 3d reconstruction.
Software Url (valid on 17 May, 2014) Notes
Freely available
Bundler Photogrammetry
Package
a,b
http://blog.neonascent.net/archives/bundler-
photogrammetry-package/
Used in James and Robson (2012). Script-based, no graphical user interface
(GUI). Windows OS only.
SFMToolkit
a,b
http://www.visual-experiments.com/demos/sfmtoolkit/ Similar software to above.
Python Photogrammetry
Toolbox (PPT)
a,b
http://code.google.com/p/osm-bundler/ Formerly OSM-bundler. Python-driven GUI and scripts, with a Linux
distribution.
VisualSFM
b
http://www.cs.washington.edu/homes/ccwu/vsfm/ Advanced GUI with Windows, Linux and Mac. OSX versions. Georeferencing
options, but camera model is more restricted than that used in Bundler.
3DF Samantha http://www.3dflow.net/technology/samantha-structure-
from-motion/
SfM only, but with more advanced camera models than all above (Farenzena
et al., 2009). Provides output compatible with several dense matching
algorithms.
Web sites and services
Photosynth http://photosynth.net/ Evolved from Bundler. SfM only, no dense reconstruction. Can incorporate a
very wide variety of images, but does so at the cost of reconstruction
accuracy.
Arc3D http://www.arc3d.be/ Vergauwen and Van Gool [2006]
CMP SfM Web service
a
http://ptak.felk.cvut.cz/sfmservice/
Autodesk 123D Catch http://www.123dapp.com/catch/
Pix4D http://pix4d.com/ Also available as standalone software.
My3DScanner http://www.my3dscanner.com/
Commercial
PhotoScan http://www.agisoft.ru/products/photoscan/ Full SfM-MVS-based commercial package.
Acute3D http://www.acute3d.com/
PhotoModeler http://www.photomodeler.com/ Software, originally based on close-range photogrammetry, now also
implements SfM.
3DF Zephyr Pro http://www.3dflow.net/ Underlying SfM engine is 3DF Samantha
Note: Table modified from http://www.lancaster.ac.uk/staff/jamesm/research/sfm.htm.
SfM ¼Structure from Motion; MVS ¼Multi-View Stereo.
a
Uses Bundler (http://phototour.cs.washington.edu/bundler/) to compute structure from motion.
b
Uses PMVS2 (http://grail.cs.washington.edu/software/pmvs/) as a dense multi-view matcher.
S.P. Bemis et al. / Journal of Structural Geology 69 (2014) 163e178164

both photographs, relative to the cameras (Fig. 1). The unknowns
comprise a ‘camera model’that describes how the camera repre-
sents the 3D world as a 2D image, the relative camera positions and
pointing directions, and the 3D point coordinates. In ‘conventional’
photogrammetry, which has evolved through the surveying, engi-
neering and remote sensing communities, initial estimates are
generally derived by providing additional control data, such as the
positions of known control points within images, prior to pro-
cessing. This approach allows error estimates (e.g., the accuracy of
the control measurements) and a real-world coordinate system to
be embedded from the outset. Processing then consists of refining
parameters through a ‘bundle adjustment’(Granshaw,1980), which
simultaneously optimizes all variables to produce a self-consistent
3D model, with minimized overall residual error. Such software
enables a highly rigorous approach and provides accurate results in
which error estimates are widely visible. However, the software
often has complexities and error intolerance that can present dif-
ficulties for inexperienced users, and the requirements for collec-
tion of suitable imagery and control data can be arduous.
Over the last decade, parallel advances in an area of computer
vision research called ‘structure from motion’(SfM) has been
driven by a different rationale eto enable automated model pro-
duction from unconstrained imagery, for which metric accuracy is
not the primary goal. Some of the most significant advances in SfM
have arisen from the development of image feature descriptors that
are tolerant to changes in view point (e.g., Lowe, 2004), and robust
matching algorithms that can identify and reject errors when they
occur (e.g., Fischler and Bolles, 1981). With these, bundle adjust-
ment can be initialized from automated image measurements
alone. For example, to work efficiently, coarse estimates for camera
model parameters are also required, e.g., focal length, but most
software will just automatically extract appropriate values from
image file metadata. Thus, 3D models can be effectively constructed
from a wide variety of imagery, with no user intervention (e.g.,
Snavely et al., 2008a, 2008b, 2006). However, the results of such a
3D reconstruction will be in an arbitrary coordinate system so, to
reference to a real-world system, the model needs to be trans-
formed through the use of some control data. The control re-
quirements are usually significantly less arduous than for
‘conventional’photogrammetry, but control data are not neces-
sarily incorporated throughout model construction and their error
estimates are more weakly integrated.
The ongoing convergence of photogrammetry- and computer
vision-based workflows is now providing powerful tools for geo-
science use (Favalli et al., 2012), enabling automated model pro-
duction from flexible image input and with increasing access to
integrated georeferencing and error analysis. The rapidly widening
availability of such software is the motivation for this paper. On that
basis, a typical photo-based reconstruction workflow is described
below and summarized in Fig. 2.
3.2. Workflow for high precision data collection
3.2.1. Image acquisition
Images can be collected with almost any digital camera, and
there are few limitations in number or types of digital cameras used
for a single set. As expected, a higher quality model output is
facilitated by higher quality input images. Additional data collec-
tion flexibility derives from the ability of SfM algorithms to match
images taken at varying scales and perspectives, provided the
photos still produce sufficient overlap. Stereoscopic aerial
target surface
camera
positions
camera
field of view
Fig. 1. Basic principles of photogrammetry for 3D reconstruction. Two overlapping
photographs taken from different positions allow each feature in the overlapping area
to be defined by a unique 3D position. Dashed lines illustrate convergent views of
discrete features from overlapping photographs.
Establish control
points or scale
Collect photographs
Mask non-stationary
portions of images
Pixel grid based
matching
Build mesh/
interpolate surface
Georeferencing and
scale
Texture mapping Reprojection
Feature detection,
bundle adjustment,
and 3D scene
reconstruction
Dense
point cloud
DEM
generation
orthomosaic
photo-
realistic model
Sparse
point cloud
FieldworkSfMMVS
Fig. 2. General workflow illustrating the photo-based 3D reconstruction process.
Diamond fields indicate potential output products at different stages during recon-
struction. MVS (Multi-view Stereo) and SfM (Structure-from-Motion) illustrate the
portions of the workflow that specifically related to these processes described in text.
S.P. Bemis et al. / Journal of Structural Geology 69 (2014) 163e178 165

photography is typically collected with 50e60% overlap between
adjacent images, under near-parallel viewing conditions (e.g.,
Krauss, 1993; Abdullah et al., 2013). SfM-based 3D reconstruction
methods also require overlapping images but, because they can
operate on unordered collections of photographs, the overlap
requirement is best considered in terms of coverage and angular
change between overlapping images. In terms of coverage, every
surface that will be reconstructed needs to be covered by at least 2
images taken from different positions, and preferably more (Fig. 3).
Increasing angles of convergence between overlapping images will
tend to increase reconstruction accuracy up to a point, but will
eventually prevent matching due to the surface texture appearing
too dissimilar in images from different directions. Moreels and
Perona (2007) found that popular feature detectors used for auto-
mated image matching did not perform well with angular changes
greater than 25e30

between images. Thus, while angular changes
between photos can increase the accuracy of reconstructed 3D
surfaces, differences should be limited to 10e20

for overlapping
photos.
The simplest approach for image acquisition over relatively
planar surfaces mimics the approach of collecting traditional ste-
reoscopic aerial photographs where images are collected in a
continuous line with a camera position orthogonal to the surface of
interest and with a frequency to produce image overlap >60%
(Fig. 3). However, the scale and layout of the target surface will
frequently necessitate adjustments to this simple approach.
Furthermore, James and Robson (2014) document systematic errors
that can be introduced across models derived from image collec-
tions where all photos are collected with parallel viewing di-
rections. This error is manifest in the axis parallel to the image view
direction, for example producing broad-scale elevation error from
images collected with a vertical orientation. James and Robson
(2014) demonstrate that one approach to mitigate this systematic
error by combining additional images with a view direction that is
inclined relative to the view direction of the rest of the image
collection (Fig. 3).
In addition to image coverage, the other key consideration in the
planning of any survey is how the texture of the target will resolve
in individual photographs. Automated feature matching relies upon
the ability of computer algorithms to identify unique
corresponding features in overlapping photos. For good results, any
effects that reduce textural variability within images or increase
feature variability between images should be minimized during
acquisition. Common issues preventing algorithms from resolving
coincident points include homogeneous surface texture, changes in
the target, and changes in illumination. The latter two have the
same effect of making a unique feature appear differently between
images, although both require different strategies for reducing the
deleterious effect on model construction. Poor or variable image
texture is often due to surface reflections, flat surfaces with little
textural variation, and the occurrence of deep shadows. The target
itself may appear to change between images due to wind shifting
vegetation, or the movement of people and vehicles. Changes in
illumination can result from accidental shading by the photogra-
pher, changes in the sun position, or filtering by clouds. Strategies
for circumnavigating these issues in structural studies are outlined
further in Section 5.2.
3.2.2. Scale and coordinates
The scaling and georeferencing requirements for the target
surface will vary with the intended use of the 3D data/model and
should be considered prior to image acquisition. An object or sur-
face can be fully reconstructed in 3D without any scale or position
information but, to extract oriented and scaled data, additional
control data must be provided. Scale can be added to a model
simply by knowing the distance between two points on an input
image or on the model. These distance measurements for scale are
best taken over the width of the target area rather than smaller,
isolated lengths. Greater accuracy in scale and full georeferencing
to local or geographic coordinates requires three or more ground
control points, usually collected with survey-grade, carrier phase
differential GPS or total station surveying. These ground control
points should be distributed widely across the target area, not
neglecting the margins. An alternate option for georeferencing is
the use of high-precision camera locations for the input images, but
current GPS units in hand-held cameras are insufficient for the
typical accuracy requirements of high resolution structural map-
ping and paleoseismology projects conducted over a few hundred
meters or less.
portion of surface resolvable by photo-based reconstructions
Fig. 3. Representation of simple image acquisition for photo-based 3D reconstruction. Grey triangles represent the field of view for each camera position and the increasing
darkness of the triangles corresponds with the number of camera positions that the surface is visible from. A greater number of overlapping images is likely to produce a denser
point cloud because more features should be resolvable, leading to a higher resolution model for that portion of the reconstruction. The sensitivity of image overlap to the image
separation, inclination, and the distance to the surface is illustrated by the size of the areas visible from 3 or more camera positions (darkest grey). Although the inclined camera
positions are not required, they are recommended for reducing possible systematic errors due to camera calibration error (James and Robson, 2014).
S.P. Bemis et al. / Journal of Structural Geology 69 (2014) 163e178166

3.2.3. Determining the imaging geometry: structure from motion
The first stage of a SfM-based 3D reconstruction involves the
analysis of the individual images for distinct image features that
can be matched to their corresponding features in other images
within the collection (e.g., Fig. 1). The SfM process then uses the
resulting network of matched points to establish the relative lo-
cations of each camera and simultaneously determines the camera
parameters for each image, the collection position and orientation
for each input image, and the 3D coordinates for each matched
feature ecommonly referred to as the bundle adjustment. Algo-
rithms typically adopt an incremental approach where bundle
adjustment of an initial image pair is sequentially repeated, with
more images incorporated at each iteration. The matched features
thus constitute a sparse 3D point cloud that represents the struc-
ture of the target surface, defined within a local coordinate system.
This point cloud is limited in precision and point density because it
is derived from robust feature matching, which has lesser accuracy
(e.g., ~0.5 pixel) than some other matching approaches
(Remondino, 2006;Barazzetti et al., 2010).
3.2.4. Densifying the measurements: multi-view stereo
With known camera models and orientations, a multi-view
stereo algorithm will produce a dense point cloud representation
of the surface. Typically, this technique will be implemented as a
systematic search over a pixel grid to identify best matches be-
tween images, with the results providing significantly more 3D
points, having greater precision than the feature matching of the
initial SfM step. Multi-view stereo is particularly intensive
computationally if the full image collection is processed simulta-
neously. However, most photo-based 3D reconstruction programs
and algorithms have the option to subset image collections (e.g.,
Furukawa et al., 2010) or to adjust the grid-cell size at which multi-
view stereo is performed so as to manage the resolution and time
required to produce the resultant dense point cloud.
3.2.5. 3D model and orthophoto generation
For many applications within structural geology and paleo-
seismology, 3D models and orthophotos are useful for high reso-
lution mapping of outcrop, rock faces or trenches. Using
triangulation or grid interpolation, it is relatively straightforward to
generate a digital elevation model (DEM) and ortho-rectified pho-
tomosaics for any selected orientation from the dense, georefer-
enced point cloud. True 2D orthophotos of any portion of the 3D
model can then be derived, with the known 3D model and imaging
geometry, allowing correction for the viewing characteristics of the
input images and the images providing the texture for the ortho-
photo mosaic. Furthermore, depending on the software used, the
3D model itself can be textured and exported in common 3D
visualization formats (e.g.,.obj,.ply,.pdf, etc.) for visualization and
interactivity.
3.3. Precision and applicability
In general terms, the accuracy of a photo-based model depends
upon the scale and resolution of the input images, the distribution
and accuracy of control data (whether ground control points, scale
measurements or camera positions), the precision and distribution
of matched image points, and the network geometry, which in-
cludes the number of photos, how much they overlap and how
convergent the views are. In close-range photogrammetry (e.g., as
often used for high accuracy engineering applications), where im-
age networks are usually multi-image and highly convergent, the
strength of a network can be described by its relative network
precision, which is a ratio of the mean 3D point uncertainty esti-
mate to the longest dimension of the network. For a given image
measurement precision, a stereo image pair with only two near-
parallel images would represent a weaker network than a conver-
gent multi-image arrangement. For projects using digital SLR im-
agery covering sub-meter to kilometer scales processed with an
SfM-based approach, James and Robson (2012) estimated relative
precision ratios of ~1:1000 or greater. These ratios were shown to
be similar to those of theoretical estimates for stereo photogram-
metry, but approximately an order of magnitude poorer than
equivalent theoretical estimates for close-range (convergent)
networks.
In most structural geology and neotectonic applications, we are
interested in a bare surface model. In areas of light to dense
vegetation, this preference for bare surfaces becomes the primary
disadvantage to photo-based 3D reconstruction relative to active
source 3D data collection tools. LiDAR collects a 3D point location
with a single pulse of light, thus is capable of collecting ground
surface points wherever the pulse of light is able to penetrate the
vegetation, reflect off the ground surface, and return to the in-
strument. The requirement in photo-based 3D reconstructions for
multiple images collected from different perspectives for 3D point
geometry reconstruction dramatically reduces the number of
resolvable ground surface points due tothe occlusion of the ground
surface by vegetation when moving from one camera position to
the next. However, because the SfM process creates a point cloud,
as long as the vegetation is sparse enough to allow a sufficient
number of ground features to be identified and matched, some of
the approaches developed for classification of LiDAR data can be
applied to SfM-derived point clouds. Furthermore, one of the
popular commercial photo-based 3D reconstruction packages
(Table 1), Agisoft PhotoScan, has implemented a point cloud clas-
sification scheme for classification of ground surface points and
vegetation (http://www.agisoft.ru/tutorials/photoscan/08).
4. Applications to paleoseismology and neotectonics
Major advances in the ability to collect, process, and visualize
high resolution 3D topographic data in the form of LiDAR has
revolutionized paleoseismology and neotectonics research over the
past ~15 years (e.g., Haugerud et al., 2003; Hudnut et al., 2002). The
ability to visualize bare-earth topography on a regional scale with
sub-meter resolution in a wide variety of landscapes using airborne
laser scanning (ALS) is currently unparallelled. ALS has been
implemented in neotectonic studies for fault mapping (e.g.,
Arrowsmith and Zielke, 2009; Bevis et al., 2005; Oskin et al., 2007),
measurement of geomorphic offsets (e.g., Zielke et al., 2010), and
measurement of coseismic surface displacements (e.g., Borsa and
Minster, 2012; Duffy et al., 2013; Nissen et al., 2012; Oskin et al.,
2012). Terrestrial laser scanning (TLS) provides a higher resolu-
tion data collection (sub-decimeter), but usually over more
restricted distances due to the limited range of most instruments
and the reduced visibility from working near ground-level. This
technique has been similarly used to measure geomorphic offsets,
determine coseismic surface displacements (Gold et al., 2012), and
scan paleoseismic excavations (Haddad et al., 2012). The high-
resolution capability of photo-based 3D reconstruction accommo-
dates many of the same applications as ALS and TLS surveying, but
also presents several site-specific and technical advantages in its
implementation due to portability, low power consumption, rapid
data collection, and low cost (Morelan et al., 2010; Johnson et al.,
2014). We provide some examples illustrating these advantages.
4.1. Neotectonics applications
One of the critical roles for geoscientists following a major
surface-rupturing earthquake is the systematic measurement of
S.P. Bemis et al. / Journal of Structural Geology 69 (2014) 163e178 167

offset features along the length of the surface rupture. Even though
these measurements may underestimate total coseismic displace-
ment due to distributed deformation off the main fault trace (Dolan
and Haravitch, 2014), these offset measurements provide critical
information about the earthquake rupture and provide markers for
assessment of post-seismic processes such as afterslip, which is the
increasing displacement that occurs after the primary earthquake.
Although many landforms that are used to record slip over multiple
earthquake cycles are persistent in the landscape, most targets that
would uniquely record the offset during a single earthquake are
transient features, including the fault scarp itself, broken and offset
trees, channel margins in active floodplains and, in the case of the
2002 M7.9 Denali fault earthquake sequence, offset glacial cre-
vasses (Fig. 4;Haeussler et al., 2004). Some of these features could
be adequately documented with a post-earthquake ALS survey, but
others require site-by-site documentation because of the scale of
the offset. Traditionally, field measurements are collected with
standard geologic field equipment etape measures and compasses,
but we propose that each coseismic offset be photographed for 3D
reconstruction so as to archive the offset feature and to facilitate
further analysis, such as automated slip-vector calculation (Gold
et al., 2012). Because coseismic offsets are a relative displacement
measurement, georeferencing is not required but scale is critical.
Scale is provided through introducing scale bars into the scene or
physically measuring and recording the distance between features
within the scene. The number of photographs required depends
upon the nature of the offset feature and the magnitude of the
offset, although a minimum of 18 photos should be considered to
capture the offset feature from all directions (360

with 20

rota-
tion between individual photographs) and additional photographs
from higher and lower perspectives will reduce shadowing and
improve the model geometry. Fig. 5 shows an example of a photo-
based 3D reconstruction for an ephemeral stream channel offset
6.6 m during the 1857 Fort Tejon earthquake on the San Andreas
fault (a fully interactive 3D model of this offset is provided in the
Supplemental Materials). The figure illustrates both the digital
elevation model from which morphological and displacement data
can be extracted, and the fully textured model that preserves a
natural visual depiction of the site. With simple scaling re-
quirements and rapid photograph collection, the additional work
required for photo-based 3D reconstructions of individual coseis-
mic offset measurements is nominal and will add to the robustness
of the overall coseismic slip distribution dataset. Furthermore, as
observed following the 2014 M6.0 Napa, California, earthquake,
afterslip processes contributed to increasing surface displacements
during the days following the earthquake (Brooks, 2014). The ease
of collecting data for photo-based 3D reconstruction of individual
offsets could accommodate high spatial and temporal resolution
monitoring of afterslip following future earthquakes.
In regions of sparse vegetation, photo-based 3D reconstructions
from aerial platforms can produce topographic surface models with
comparable accuracy to ALS surveys (e.g., Fonstad et al., 2013;
James and Robson, 2012; Johnson et al., 2014; Westoby et al.,
2012). Although spatial coverage of ALS-derived topographic data
is expanding, the availability is concentrated in relatively few
countries and regions. Therefore, in regions where ALS-derived
Fig. 4. Examples of extremely short-lived geomorphic markers of coseismic fault displacement from the November 3rd, 2002, M7.9 Denali fault earthquake sequence in on the
Denali fault in south-central Alaska. Tremendous effort was invested in capturing these offsets, many of which disappeared by the following summer. Image collection for photo-
based 3D reconstruction techniques would not require more than a few extra minutes per site and preserve a richer record of the fault offset. (a) Patty Burns measures the vertical
separation of a fault scarp that is expressed as an offset glacier surface across the Susitna Glacier fault. (b) A 5.5 m offset of an active stream bank. (c) Measuring the displacement of
a tree that was broken and offset across the Denali fault. (d) Peter Haeussler attempts to measure the displacement of a glacial crevasse. The 5.5 m offset documented in (b)
immediately after the earthquake in November 2002 increased to 6.6 m by July 2003 (Haeussler et al., 20 04). Photos (a), (b), and (d) are courtesy of the U.S. Geological Survey
(http://earthquake.usgs.gov/earthquakes/eqinthenews/2002/uslbbl/photos/pr071102/).
S.P. Bemis et al. / Journal of Structural Geology 69 (2014) 163e178168

topographic data is not available, photo-based 3D reconstruction
techniques can accommodate many of the innovative neotectonic
analyses being performed with ALS-derived topography (e.g., Akçiz
et al., 2010; Hilley et al., 2010; Nissen et al., 2012). Even where ALS-
or TLS-derived topographic data already exist, photo-based 3D
reconstruction techniques can add to the temporal dimension of
the topographic data by enabling more frequent surveys when
repeat ALS or TLS surveys would be cost-prohibitive or logistically
difficult. Analyses of pre- and post-earthquake ALS topographic
data for synthetic earthquakes (Borsa and Minster, 2012) and actual
earthquake ruptures (Duffy et al., 2013; Nissen et al., 2012; Oskin
et al., 2012) demonstrate the ability to extract 3D displacements
from high-resolution topography, and Krishnan et al. (2012) pre-
sent methods for registration of photo-based reconstruction sur-
face models with ALS topography to enable similar analyses with
data derived from different methods.
4.2. Photo-based 3D reconstruction in paleoseismic investigations
A fundamental product of any paleoseismic investigation is a
representation of the exposed stratigraphy and faulting eessen-
tially producing a geologic map (the so-called “trench log”) of the
exposure. Early methods for producing the trench log relied upon
surveying or measurement from a reference grid to points along
contacts and faults and transferring this information onto a sheet of
graph paper (McCalpin, 2009). To complete the logs in the field,
lines are visually interpolated on the log between measured points,
and characteristics of mapped units sketched in and described. To
increase the information contained and communicated by a trench
log, many paleoseismologists produce photomosaics for the expo-
sure, map directly onto these photos in the field, and create final
publication-quality logs. Using photographs as a base map captures
the rich tonal and textural information, but the traditional
approach to photomosaic production required time-intensive ad-
hoc rectification of individual photos. This rectification process
requires having a complete system of measured grid lines estab-
lished on the trench walls at a scale that allows individual photos to
capture ideally all four margins of a grid rectangle. The rectification
is performed by manually warping and distorting the image to
restore the reference grid lines within the photos to vertical and
horizontal, and then these individual photos are cropped and
assembled into the photomosaic. This manual rectification and
mosaicking process is hindered by holes in the trench wall, clasts
and other features that protrude from the wall, and places where
the reference grid is not flush with the trench wall surface. Because
these photos are often taken at close range with wide angle camera
lenses, surface irregularities on the margins of the photo are
particularly problematic, potentially introducing positional errors
of several cm or more during manual rectification and obscuring
stratigraphic relationships. The resulting photomosaic then re-
quires adjustment of tonal and lightness characteristics for
adjoining photos to provide even colour and contrast across the full
photomosaic.
This complex and time-consuming approach to constructing 2D
photomosaics for paleoseismology can be largely circumvented
through photo-based 3D reconstruction techniques. A properly
planned collection of photographs from a trench wall will facilitate
the full 3D reconstruction of the topography of that surface, which
in turn accommodates the precise and automated orthor-
ectification of the input photographs. For a planar trench wall, we
have achieved consistently high-quality results by taking photo-
graphs in a similar fashion as traditional photomosaics, but
increasing photograph overlap so that ~3 times more photos are
collected (Fig. 6). In this case, rather than just photographing each
measured grid cell, the image coverage is increased by collecting an
intermediate photograph centered on the gridlines between each
grid cell ehorizontally and vertically. This approach includes tak-
ing additional photos of the trench margins recognizing that these
marginal photos will only partially cover portions of the trench
wall. Most photographs should be taken orthogonal to the wall to
Fig. 5. Offset channel on the Carrizo Plain, California, from the 1857 Fort Tejon earthquake on the San Andreas fault. Photos were collected with a GoPro camera mounted on a long
pole and held overhead. This model is derived from 56 photos with (a) showing an oblique view of the resulting shaded relief surface model and (b) showing the identical oblique
view with the mosaicked image texture. This is channel offset ZA10543 from Zielke et al. (2010) who documented an offset of 6.6 m (þ0.5/1.0). Black arrows illustrate the offset of
the channel thalweg and point in the downstream direction. View is looking north. The 3D model these images are derived from is provided as a 3D PDF in the Supplementary
Material.
Fig. 6. Example of photomosaic production within Agisoft PhotoScan. Blue rectangles
with black normal vectors (labelled with the image file name) represent the camera
positions and orientations determined from the SfM process. The background image is
the orthorectified photomosaic that was exported to create Fig. 7.
S.P. Bemis et al. / Journal of Structural Geology 69 (2014) 163e178 169

ensure that complete wall coverage by photographs is attained and
to provide a systematic angular change between overlapping
photographs. Additionally, a few photographs should be collected
oblique to the trench wall to reduce possible error from broad-scale
warping of the reconstructed surface (e.g., James and Robson,
2014). An example of a photomosaic produced by this process is
shown in Fig. 7 (with the source 3D model provided in
Supplementary Materials) for a paleoseismic trench at a new
paleoseismic site on the Mojave section of the southern San
Andreas fault near Elizabeth Lake, California. For this photo-based
reconstruction, we collected 191 photographs with a Tokina
11e16 mm f/2.8 AT-X 116 Pro DX lens on a Nikon D7000 DSLR over
the span of 29 min, covering one wall of the 17 m long, up to 4 m
deep, and ~1 m wide trench following the photo spacing illustrated
in Fig. 6. We imported this image collection into Agisoft PhotoScan
and used the masking tools to hide non-stationary objects and
features that lie in the distant background outside of the trench
wall. The 3D reconstruction of these photographs took ~1 h total
using low resolution processing settings on a desktop computer
with a 2.8 GHz processor and 12 GB RAM. Lower resolution pro-
cessing may reduce the resolution of the 3D reconstruction of the
wall surface but does not impact the resolution of an output
photomosaic. The output photomosaic resolution is controlled by
the quality of the input photographs and resolution parameters set
at the time of export. However, lower resolution processing may
impact the accuracy of the photomosaic on the scale of several mm
if the resolution of the trench wall surface model is insufficient for
proper orthorectification of input images. Some photo-based 3D
reconstruction software packages perform automatic colour and
contrast matching across the mosaicked images to produce an
orthomosaic with consistent tone and colour spectrum. In our
experience, we find this capability has the relatively minor disad-
vantage of reducing overall contrast, and possibly clipping or flat-
tening portions of the colour spectrum. This disadvantage is far
outweighed by the ability to manually enhance contrast, colour,
saturation, etc. evenly across the entire photomosaic without
having to perform these adjustments individually for photos within
the photomosaic.
Producing a 2D photomosaic from more complex trench-wall
configurations simply requires strategic photograph collection to
ensure capture of the full range of angles required to prevent oc-
clusion of portions of the trench wall. A particular example may be
paleoseismic trenches that are constructed of benched walls due to
their depth or unstable substrate, such that vertical faces are up to
~1.5 m tall and separated by a bench to the set-back higher wall
face. Nonetheless, in one instance, we were able to revisit photo-
graphs collected in 2006 from a paleoseismic investigation on the
southern San Andreas fault near Coachella, California, that utilized
a 7 m deep trench with benched walls to expose a section of
alternating lacustrine and aeolian deposits spanning the past ~1000
years (Philibosian et al., 2009). Using just 15 photographs taken
from the opposite side of the trench, we rapidly reconstructed the
geometry of a portion of the benched trench wall in 3D (Fig. 8a;
Supplementary Material). Although mapping the stratigraphy has
historically been done on a 2D representation of the trench wall,
compiling this mapping onto the 3D reconstruction provides a
more complete model for fault geometry and along-strike (cross-
trench) variation.
In addition to the trench wall orthophotos, establishing the full
3D geometry of a trench facilitates robust structural and strati-
graphic analysis while avoiding the time-consuming process of
surveying each fault and stratigraphic contact. Fig. 8 shows two
examples where we used a relatively small collection of photo-
graphs to reconstruct the geometry of an entire trench or complex
trench wall (Fully interactive 3D models provided in
Supplementary Material). A small reconnaissance trench (3 m long,
1.5 m deep, and ~1 m wide) on the Denali fault, near Cantwell,
Alaska, was fully captured by 13 photographs taken while standing
on the ground surface outside the trench (Fig. 8b). The trench ge-
ometry was properly reconstructed in 3D from this small, highly
oblique, photograph collection. Additional photographs taken
orthogonal to the walls would improve resolution of the stratig-
raphy but are not critical for reconstructing trench geometry.
Although detailed surveying of contacts and gridlines are not
necessary when photo-based 3D reconstructions are utilized, we
suggest that all paleoseismic studies should utilize high-precision
surveying instruments (e.g., total station or differential GPS) to
locate the 3D model in real world coordinates and provide precise
relative positioning between successive excavations and for future
examinations of the site.
An advantage for structural studies in paleoseismology vs. hard-
rock structural geology is the ability in many paleoseismic studies
to easily modify the target exposure. If faulting and/or stratigraphic
relationships are not clear in an exposure, the trench wall can be cut
back to expose a new view withthe hope that the relationships will
become clearer. Furthermore, this strategy can be implemented in a
progressive fashion where the trench wall is cut back incremen-
tally, with structure and stratigraphy documented on each fresh
face for tracking changes in fault geometry and stratigraphic re-
lationships in the 3rd dimension. Unfortunately this destructive
Fig. 7. Orthorectified photomosaic created with photo-based 3D reconstruction techniques utilizing Agisoft Photoscan. No additional image adjustments were performed following
orthomosaic production except for rotation and downsampling. The even tone and contrast across the photomosaic along with the accuratespatial geometry improves the quality of
detailed mapping in the field and more completely preserves these characteristics for archiving and publication. This image is the east wall of one of the trenches at the Elizabeth
Lake paleoseismic site on the Mojave section of the southern San Andreas fault, California (view is looking to the southeast). The black box shows the area of photomosaic in Fig. 6.
The 3D model this image is derived from is provided as a 3D PDF in the Supplementary Material.
S.P. Bemis et al. / Journal of Structural Geology 69 (2014) 163e178170

technique obliterates the original stratigraphy and prevents the re-
examination of the stratigraphic record in the field. An important
positive outcome is that the photo-based 3D reconstruction pro-
vides a solution for rapid and complete archival documentation of
each face created during a progressive excavation. After the face is
cut, cleaned, and prepared, the face is photographed with either
fixed reference points outside the zone of excavation or reference
points that are surveyed on each face. The faces are then recon-
structed individually and compiled into a 3D model that records the
removed stratigraphy. This 3D model enables the detailed inter-
pretation of the 3D geometry of faults and stratigraphic horizons, as
well as preserving the original tonal and textural information for
future investigations. A 3D archive of recent stratigraphy and
structure may prove to be a powerful tool for researchers of other
geologic subdisciplines or for paleoseismologists to re-interpret a
site using new insights that develop during later research.
5. Applications to structural geology
5.1. High-resolution mapping using UAV surveys
High-resolution fault, vein and fracture maps are routinely
created in structural geology to constrain, amongst other things,
the nucleation, growth, mechanics and scaling properties of frac-
tures (e.g. Shipton and Cowie, 2001; Wilson et al., 2009; Nixon
et al., 2011) and the permeability characteristics and sealing capa-
bilities of fault systems (e.g., Willemse et al., 1997; Peacock et al.,
1998; Antonellini et al., 2008). Combinations of grid mapping, in-
terpretations from overlapping outcrop photographs, aerial pho-
tographs and detailed sketches are employed. More recently, digital
techniques such as terrestrial LiDAR scans are increasingly used,
linked to satellite imagery and field observations (e.g., Pringle et al.,
2006; Wilson et al., 2009).
Conventional, high-resolution grid-mapping techniques suffer
from time constraints and are either impractical to collect data at
cm-scale resolutions over significant areas or take weeks to ach-
ieve. The results are also limited because the final product remains
a map interpretation that can only be verified by further visits to
the study site. Photogrammetric datasets derived from UAV plat-
forms (Fig. 9) offer a cost-effective, ultra-high resolution alternative
with a rapid acquisition time. They have the added advantage of
providing access to vertical or unstable exposures, while delivering
visual data in digital form that can be shared and reanalysed
without need to revisit the outcrop.
Fig. 10 is a rendering and interpretation of coastal outcrop from
Piccaninny Point on the northeast coast of Tasmania, Australia,
which exposes a spectacular series of strike-slip faults. These faults
comprise a damage zone of intersecting structures, crosscutting a
subvertical succession of metasedimentary sandstones and silt-
stones, belonging to the Mathinna Group (Banks, 1962; Gee and
Groves, 1971; Groves et al., 1977). We deployed an eight-rotor
Oktokopter (Fig. 9) with a small format digital camera (Canon
550D 15 Megapixel, DSLR with Canon EF-S 18e55 mm F/3.5-5.6 IS
lens). An onboard, navigation grade GPS receiver is integrated with
a Mikrokopter Flight Controller ME V2.0 and these permit an
autonomous flight dictated by programmed waypoints. During this
particular deployment, wind gusts reached speeds of 35 km/h,
requiring manual control of the UAV. Nevertheless, the strong wind
conditions encountered demonstrated the robustness of the survey
method. Approximately 140 outcrop photographs were collected
by the UAV. An area 100 100 m was covered in <5 min, at alti-
tudes of 30e40 m, which produced high resolution imagery (1
pixel z10 mm). Even lower altitudes of 15e20 m allow sub-cm
resolutions, although the advantage of this increase is countered
by larger photographic datasets and associated increases in data-
processing time. In contrast, the best resolution available from
Fig. 8. Examples of geometric reconstructions from small collections of overview photographs. (a) oblique views of a section of an ~7 m tall, benched trench wall across the
southern San Andreas fault near Coachella, California. Blue rectangles and black vectors indicate the position and orientation of the photographs used to build this model. Although
these photos were not collected with photogrammetry in mind, there was sufficient overlap to reconstruct a detailed 3D model. Reed Burgette kneeling on the bottom bench for
scale. (b) A small (3 m wide, 1.5 m deep, 1 m wide) reconnaissance trench across the Denali fault near Cantwell, Alaska. As illustrated, this model was reconstructed from just 13
pictures taken from above the trench, but even with this highly oblique photography, the stratigraphy near the base of the trench resolves well. In particular, note how the
stratigraphy drapes over the boulder protruding from the trench wall. The 3D model these images are derived from is provided as a 3D PDF in the Supplementary Material.
S.P. Bemis et al. / Journal of Structural Geology 69 (2014) 163e178 171

manned aircraft or satellites is typically 100e500 mm/pixel (e.g.
Nixon et al., 2011).
As with the paleoseismic studies, the workflow that followed
acquisition of the photographs began with manual selection of the
most appropriate photos from the dataset. In this example, the
images were processed using the Bundler software (Snavely, 2010).
Processing of the imagery was a semi-automated task, taking
around 6 h to complete on a high-end desktop computer. Initially
the SIFT algorithm ( Lowe, 2004) is used to detect features across the
images, and these features are then matched between overlapping
images. Bundler then used these matching features to complete a
bundle adjustment and to align the images using an arbitrary co-
ordinate system. An output file was generated that listed the
calculated position of each camera and each matched feature in the
Bundler arbitrary coordinate system. The result was a sparse point
cloud in which the x,y,zposition of each matched feature is listed
along with its RGB colour from the original imagery. The point
cloud was densified by use of the Patch-based Multview Stereo
software (PMVS2; Furukawa and Ponce, 2009) to produce a point
cloud containing many millions of points (Fig. 9bec). A real-world
coordinate system was derived for the output using the direct
georeferencing technique described in Turner et al. (2012). The
direct georeferencing technique links the timestamp of each
photograph to both the GPS position logged onboard at the time of
exposure and elevation as provided by the OktoKopter's barometric
altimeter (accuracies of ±1 m during UAV flight), which derives
camera coordinates for each photograph. These coordinates are
then matched to the computed (Bundler) coordinates to solve for
the Helmert transformation parameters (3 translations, 3 rotations,
1 scale parameter). The derived transform is subsequently applied
to the sparse point cloud, resulting in thousands of real-world co-
ordinates linked to points in each image and allowing the imagesto
be orthorectified using a Delaunay triangulation. Turner et al.
(2012) report the absolute accuracy of the orthorectified images
produced from this method is typically <400 mm. Using this direct
georeferencing technique, absolute accuracies are dominated by
the navigation grade GPS errors, estimation of the camera focal
length and interior/exterior orientation parameters and imprecise
synchronization between the GPS receiver and camera (Turner
et al., 2013). This series of steps is now largely implemented in
off-the-shelf software products such as Photoscan (Table 1).
Once georeferenced point cloud data are derived, obtaining
renderings is straightforward for direct use in structural geology,
such as textured wireframe models, orthorectified photomosaics
and DEMs. For example, the Piccaninny Point data were converted
to these datasets and rapidly digitized in a GIS environment
(Fig. 10), revealing a complex damage zone around an eroded fault
core (Buckley, 2013), which is the subject of ongoing research at The
University of Western Australia. The high-resolution nature of the
imagery, in combination with the intense layering of the outcrop,
allowed fault offset directions to be detected (Fig. 10d) and dis-
placementelength profiles to be calculated for each damage zone
structure. The pixel resolutions of 10 mm in the photomosaic and
20 mm in the DEM allowed for offsets greater than ~20e50 mm to
be identified with certainty over the 10,000 m
2
outcrop area.
In cases where absolute location or improved accuracies are
required, real-world coordinates of Ground Control Points (GCPs)
can be established within the scene (James and Robson, 2012;
Turner et al., 2012). Typically, painted target markers are placed
in the area and surveyed using dual frequency differential GPS
(horizontal accuracies of 20 mm and vertical accuracies of 40 mm)
or total station techniques (<10 mm accuracies). For such an
outcome, 10e15 markers placed throughout the area would be
sufficient. When scale alone is required, only control distances
across the scene are required (James and Robson, 2012). Other
important considerations to minimise error include using a high
quality pre-calibrated camera (i.e. DSLR) with fixed optics, collect-
ing slightly convergent imagery (Wackrow and Chandler, 2011;
James and Robson, 2014), and surveying the outcrop using flight
lines from 2 orthogonal directions (P. Kovesi pers comm.). Recently,
Turner et al. (2013) obtained large improvements to the absolute
accuracy of the direct georeferencing technique with modifications
to the onboard GPS receiver, synchronization between the GPS
receiver and camera and corrections between the GPS antenna and
camera position. Absolute accuracies of 100e200 mm were ach-
ieved (Turner et al., 2013). Similar improvements are only likely to
accelerate in the next few years, and the need for GCPs may well be
eliminated in the near future at the point where direct georefer-
encing techniques achieve similar accuracies to GCP studies con-
strained by differential GPS. Nonetheless, GCPs will continue to be
useful where outcrop studies require ground-based verification or
sub-cm accuracy (e.g. using total station surveys).
As a result of these developments, UAVs are likely to be a
commonly utilized tool by structural geologists in the near-future.
Fig. 9. (a) Oktokopter UAV undergoing a test flight. A DSLR is suspended on a gimble
and can be rotated in flight. (bec) SfM-derived point cloud of Piccaninny Point, viewed
from the southeast and above, respectively. The outcrop is 100 m long in this scene. A
wavecut rock platform is captured. Vegetation and a vehicle are present as noise.
S.P. Bemis et al. / Journal of Structural Geology 69 (2014) 163e178172

A very large range of potential applications can be envisaged
beyond the scope of this review. Nonetheless, it is worth providing
a brief note about the technology. UAVs are broadly subdivided into
fixed-wing types (e.g. model airplanes) and multi-rotor types such
as used in the example above. For high-resolution studies, multi-
rotor UAVs have significant advantages over fixed wing UAVs
because they can fly at exceptionally low altitudes with generally
higher quality cameras (relative to the size of the craft) mounted on
a stabilised platform. The Oktokopter employed in this case study is
capable of covering approximately 20,000 m
2
(2 ha) in a single
flight with a 1.5 kg payload, and is particularly useful for vertical
faces where imagery is required near to corners and surfaces at a
Fig. 10. (a) A selection of the overlapping photos captured by UAV, and their coverage of Piccaninny Point. (b) DEM derived from the point cloud shown in Fig. 9b, with an un-
derlying hillshade. Black dash etrace of large fault. (c) Orthorectified photomosaic for a portion of the outcrop. Pixel resolution 1 pixel ¼10 mm. (d) Structural interpretation of fault
and associated damage zone, showing dextral (red), sinistral (blue) and unidentified offset faults (grey) developed around a large fault (black dash), superimposed over the DEM.
S.P. Bemis et al. / Journal of Structural Geology 69 (2014) 163e178 173

range of angles. On the other hand, fixed wing UAVs cover far larger
areas of ground in a smaller timeframe, and are less power hungry.
The main restrictions on the use of UAVs are from national regu-
latory frameworks, which vary widely from country to country. In
Australia for example, UAV use is governed by the Civil Aviation
Safety Authority (CASA; http://www.casa.gov.au/), which stipulates
that research use of UAVs requires controller's and operators cer-
tification to fly. This certification requires general aviation knowl-
edge in line with a pilot's licence. The UAV can fly no more than
120 m above ground unless special approval is provided, must
remain within line-of-sight, and must be flown over unpopulated
areas and outside controlled airspace.
5.2. Ground-based outcrop and open pit surveys
With the advent of SfM and its implementation in off-the-shelf
products (Table 1), photogrammetry is now an ideal tool for
ground-based structural studies because it generates digital 3D
models of natural outcrop, quarries/mine sites and hand specimens
(Fig. 11;Supplementary Material). Rock textures and fabrics can be
produced with high fidelity, using processing and visualization
tools such as Agisoft Photoscan™, which perform well in both the
ground-based studies discussed here and UAV-based studies
(Turner et al., 2013). Photo-based models also compare favourably
with laser scanning techniques when used to identify joints, dis-
continuities and orientations (e.g. Coggan et al., 2007), possibly
because photogrammetric data points inherently contain colour as
well as location and it is easier to derive information on surfaces
from multiple orientations using a hand-held DSLR rather than a
tripod-mounted scanner. It should be noted that laser-scanning
techniques are advancing as rapidly as photogrammetry so that a
number of these issues are now avoided (Hodgetts, 2013).
The vein array shown in Fig. 11 is a photo-based model of an en-
echelon sigmoidal vein array, which changes into planar en-
echelon veins on the alternate side of the hand specimen
(Fig. 11a). In this example, an ultra-high resolution model was
required and 100 photographs were collected under diffuse light,
using a 100 mm fixed focal length lens and Canon EOS-5 mark III
DSLR. Several different models were generated from the dataset
and it was found models with the best texture actually used less
than the full complement of 100 photographs. Fig. 11 was con-
structed from 51 photographs, which generated a densified point
cloud of 2,442,780 points and a wireframe with 493,195 elements.
The model was processed in ~4 h on a laptop with Intel i7 CPU and
8 Gb RAM.
The model faithfully reproduces the trajectories of individual
calcite vein fibres, sets of parallel, closely spaced microveins and
pressure solution seams, with only minor distortion present in the
orthorectified photomosaic of the front face (Fig. 11b). Photo-
grammetric datasets such as these are being used to refine our
existing kinematic and mechanical models for vein formation (e.g.
Beach, 1975; Olson and Pollard, 1991; Bons et al., 2012). As with the
UAV-based models of Piccaninny Point, hand specimens and out-
crops derived from ground-based studies can be georeferenced and
converted to DEM and orthorectified photomosaics for mapping
and analysis in a GIS environment. In addition, the open source 3D-
graphics package Blender (www.blender.org), allows true 3D
mapping to be carried out and projected onto the surface of the
photo-based reconstruction (Fig. 11c). Spatial information
describing vein geometry and the orientation of features such as
vein surfaces or internal crystal fibres can be extracted. Given the
imagery, we have since been able to generate super high-resolution
models of individual veins and their fibres using overlapping im-
ages captured through bifocal microscopes.
Fig. 11. (a) High fidelity textured photogrammetric model of a hand specimen from the
Cape Liptrap Formation, Wilsons Promontory, Victoria, Australia. The front and rear
faces of the specimen show that 6 en-echelon sigmoidal veins link and develop into 3
planar en-echelon veins. The top and lower thirds of the hand specimen have been
digitally removed. (b) Orthorectified photomosaic of the front face. Geological fabrics
and geometries are reproduced in fine detail, with limited distortion in 2 spatially
restricted domains. (c) Photogrammetric model. The left-hand third of the model is
textured, the other two-thirds reveal the underlying wireframe with such high reso-
lution that individual triangular elements are not discernible. Pressure solution seams
(red), calcite veins (green) and calcite vein fibre trajectories (white) were mapped in
3D using the open-source Blender graphics environment. The 3D model these images
are derived from is provided as a 3D PDF in the Supplementary Material.
S.P. Bemis et al. / Journal of Structural Geology 69 (2014) 163e178174

For lower resolution models, the same sample was digitally
reproduced using just 30 photographs, captured by a compact
Canon S95 digital camera (i.e., not a DSLR). The photos were pro-
cessed in <1 h on a laptop, and even in this case, individual calcite
fibres are visible. As such, the application of photogrammetry for
pedagogical purposes is self-evident, because it is now being rela-
tively easy to build a virtual library of useful hand specimens.
Furthermore, these models may be printed in 3D using inexpensive
web-based providers (e.g. http://www.shapeways.com/), which
allows precious or fragile samples to be reproduced for lab-based
exercises.
At larger scales, the same workflow is followed for ground-
based digital mapping of open-pit minesites and large outcrops.
Trials involving ground-based digital mapping of a >500 m long
open pit required 150e250 photos with a 28 mm focal length lens,
and 500e700 photos with a 50 mm lens (Fig. 12). Larger focal
lengths lead to higher resolution models but, as a general rule,
twice the number of photographs requires a four-fold increase in
processing time. Our preliminary trials have shown a number of
additional parameters must be considered to achieve satisfactory
results, and that model quality is influenced by three main
parameters:
(1) Lighting conditions:Reflective surfaces and strong contrasts
in light across a scene negatively affect point matching.
Diffuse lighting conditions are preferable.
(2) Duration of survey: Because the sun's azimuth continues to
change as a survey progresses, point matching between
photographs becomes complicated by changes in shadow
length and surface albedo. It was found that model quality
degrades significantly for durations >30 min. For long-
duration surveys, this effect can be circumnavigated by
returning to the outcrop at approximately the same time the
next day if similar weather conditions prevail.
(3) Image network geometry: The capture of photographs from a
limited number of poorly distributed locations (stations) can
lead to model distortions (e.g. Wackrow and Chandler, 2011;
James and Robson, 2014) and missing regions. The use of
GCPs and convergent imagery is important to minimize any
such distortions. Convergent imagery will also allow recon-
struction of complex surfaces with a wide variety of face
directions. Outcrops, mine sites and quarries are best
reproduced if access is available all around (including inside)
and images are captured in an organised semi-continuous
manner.
Practical difficulties associated with these considerations are
typically overcome by UAV-based surveys, which can provide su-
perior coverage in a short period of time (<10 min). Nonetheless,
from ground-based surveys, we have been able to generate 3D
geological maps, identify and track lithologies across inaccessible
sub-vertical faces and extract planimetric orthorectified photo-
mosaics with equivalent or higher resolution and less expense than
achieved by industry-standard aerial photographs (e.g., Fig. 11a).
Following discretization of the wireframe meshes, mapping was
completed in the interpolation modelling package Leapfrog Mining,
which is based on Fast Radial Basis Functions and has the potential
for true 3D mapping (sensu stricto McCaffrey et al., 2005), where
photogrammetric datasets can be integrated with drill core logging,
multi-element chemistry and geophysical data etc.
5.3. The future of photogrammetry and structural interpretation
Photogrammetry and other digital techniques, such as photo-
realistic laser scanning (Hodgetts, 2013), offer a step-change in
the amount of data available from outcrop. These techniques are
part of an arsenal of digital approaches to field mapping that are
now available (McCaffrey et al., 2005) and may become particularly
powerful as we develop techniques to merge and automatically
analyse photogrammetric data with other potential field and
remote sensing data (e.g. aeromagnetics, hyperspectral and radio-
metric data). In addition, with the advent of Virtual Globes, such as
Google Earth, it becomes possible to visualize photogrammetric
alongside other structural data (Blenkinsop, 2012) on a carto-
graphic representation of the Earth (De Paor and Whitmeyer, 2011).
As demonstrated, photo-based 3D approaches permit the ca-
pacity for mm-cm scale resolution, over many hundreds of metres,
if not more in the case of fixed wing UAVs. Nonetheless, digital
mapping of the data using manual approaches remains time
consuming. The fault-fracture interpretation of orthorectified im-
ages extracted from Piccaninny Point (Fig. 10d) required ~2 weeks
of manual digitizing of polylines and the building of a database of
attributes (e.g. slip sense, offset etc). This raises the twin questions
of how can we use such data more efficiently, and how can we
extract information not readily available to manual interpretation?
Two promising approaches exist to aid rapid mapping and in-
formation extraction. The first involves image analysis of the DEM
and orthorectified photomosaic data (e.g. Stumpf et al., 2013;
Vasuki et al., 2014). The second involves identification, mapping
and classification of point cloud attributes using Artificial Intelli-
gence approaches (Hodgetts, 2013). Recently, a semi-automatic
Fig. 12. (a) A high fidelity textured photogrammetric model of a legacy open pit mine from the Coolgardie greenstone domain, Western Australia. The pit is >500 m long and this
model was constructed using 600 ground-based photos from a 50 mm focal length lens. Features on the scale of ~10 cm can readily be detected in the model and the spatial
resolution of orthorectified images extracted from the data compete with those obtained by the highest quality aerial photographs and can be obtained at any angle of projection.
(b) Oblique view of the same model.
S.P. Bemis et al. / Journal of Structural Geology 69 (2014) 163e178 175

method for the rapid mapping of discontinuities like faults was
developed using phase congruency and phase symmetry as edge
detection methods (Micklethwaite et al., 2012; Vasuki et al., 2013,
2014) and user interaction to guide the process. Fig. 13 shows
stages in the process and a comparison with a manually digitized
fault map. A user rapidly defines the broad location and length of
faults in the outcrop (Fig. 13a), then edges detected in the data are
used to construct the true geometry of the faults (Fig. 13bec). In the
example shown, the user-guided interpretation was completed in
10 min, while manual digitizing took approximately 7 h, repre-
senting a significant increase in efficiency when interpreting the
data. In a later stage (not shown), the detected fault traces were
combined with the point cloud data to extract orientation data
systematically along the faults (Vasuki et al., 2014) using the
RANSAC algorithm (Fischler and Bolles, 1981) to best-fit planes
through points lying along the fault. Simple triangulation or tensor
analysis approaches can also be used to identify surface orienta-
tions of faults, fractures or bedding surfaces (Feng et al., 2001;
Fern
andez, 2005).
Secondly, as highlighted by Hodgetts (2013 and references
therein) analysis of the attributes of point cloud data can emphasise
patterns and textures not obvious at first inspection. Artificial In-
telligence approaches such as Neural Networks, Fuzzy Logic and
Evolutionary Algorithms allow the automatic classification of point
cloud data, aiding the identification of varying stratigraphy (van
Lanen et al., 2009; Rarity et al., 2013) or the extraction and
upscaling of fault and fracture populations (e.g. Gillespie et al.,
2010;Seers and Hodgetts, 2013), especially when combined with
field observations. The approach is implemented in Virtual Reality
Geological Studio (VRGS; developed in-house at the University of
Manchester), and has been applied mostly to laser scan data
(Hodgetts, 2013) but is directly applicable to photogrammetric
point clouds.
6. Discussion and conclusions
With a variety of software implementations of photo-based 3D
reconstruction, including open-source options (Table 1), these
techniques represent the ‘democratization’of high-resolution 3D
geospatial data collection by making the collection of these data
available to anyone with a computer and a digital camera.
Photo-based 3D reconstruction techniques have a broad spec-
trum of applications in structural geology and neotectonics due
to the ability to collect vast quantities of high-resolution 3D
geospatial data across multiple spatial scales. The limited
infrastructure requirements that increase portability and
decrease cost may allow photo-based 3D reconstruction tech-
niques to supersede other common high-resolution surface
modeling techniques in many research settings. Photo-based 3D
reconstruction provides an opportunity for archiving and
sharing of high fidelity imagery and 3D geometry from critical
rock exposures and sediments. It is made particularly powerful
because of its ability to easily collect data from inaccessible or
unsafe exposures, or to record a time series of data such as when
paleoseismic trenches are successively cut back.
Resolution and accuracy during 3D reconstruction is dependent
upon a number of parameters. Nonetheless, it is relatively
straightforward to design a survey and processing routine that
suits the resolution needs of the project. The relative precision
ratio described by James and Robson (2012) is a useful guide
when designing a data collection routine by illustrating the
expected order of measurement precision relative to the
observation distance based upon observed capabilities of photo-
based 3D reconstruction software. For example, with this ratio
Fig. 13. (a) A sketch map of the approximate locations of faults in an orthorectified
image, using a limited number of clicks (approximately 10 min interpretation). (b) A
semi-automatic fault map constructed by matching the user defined approximate fault
locations with automatically detected edges, to construct realistic fault geometry,
segmentation and lengths. (c) Comparison between the semi-automatic fault map and
a manually digitized map derived in a standard GIS-environment. Dark blue esemi-
automatically identified discontinuities; red emanually digitized faults; light blue e
manually digitized joints; green emanually digitized extension fractures. The semi-
automatically mapped faults have the same geometry and approximate lengths as
the manually digitized counterparts, without false positives. Finer detail would have
been possible with slightly longer user interaction.
S.P. Bemis et al. / Journal of Structural Geology 69 (2014) 163e178176

generally exceeding 1:1000 (James and Robson, 2012), a study
that requires mm-scale precision will require photographs to be
collected within several meters of the target in addition to the
requirements for well-distributed camera positions and ground
control.
Future directions in structural geology and paleoseismology
using photo-based 3D reconstruction derive from the rich
dataset that includes complete tonal and textural information
combined with a high-resolution 3D model. Novel image anal-
ysis methods are becoming available to rapidly generate maps
from the large geospatial datasets, without spending days to
weeks manually digitizing (e.g. Vasuki et al., 2014). Further-
more, Artificial Intelligence-type algorithms provide new ways
to extract meaningful sedimentological and stratigraphic pa-
rameters from exposures using the combined 3D geometry and
image information (Hodgetts, 2013).
Finally, photo-based 3D models offer the ability to build virtual
archives of geoscientific data with relatively low cost or exper-
tise required. In the realm of the communication of scientific
results in Structural Geology and Paleoseismology, this ability
ought to lead to the development of electronic archives of crit-
ical outcrop or hand-specimen observations to accompany
publications. Such a process would be analogous to electronic
lab-books common to the biosciences. Secondly, there are
obvious pedagogical applications. Many students find difficulty
in visualizing the 3D nature of geological features when these
are shown as standard photographs. 3D models of outcrops and
hand specimens, which can be manipulated individually by
students, would help to overcome this hurdle in visualization
and learning. Rather than replacing field trips, these interactive
models could allow students to revisit key exposures as new
concepts are discussed in order to layer the concepts and rein-
force the connections between the concepts and real world
examples.
Acknowledgements
Cees Passchier is thanked for his ideas and encouragement to
submit this paper. David Hodgetts hosted and introduced SM to
VRGS and its application for photogrammetric data. Tom Blenkin-
sop and an anonymous reviewer are thanked for reviews that
contributed to the clarity of the paper. Southern California Earth-
quake Center award #13136 to SB supported the work at Elizabeth
Lake, CA, presented here. SM was supported by the Hammond-
Nisbet Endowment during this work.
Appendix A. Supplementary data
Supplementary data related to this article can be found at http://
dx.doi.org/10.1016/j.jsg.2014.10.007.
References
Abdullah, Q., Bethel, J., Hussain, M., Munjy, R., 2013. Photogrammetric project and
mission planning. In: McGlone, J.C. (Ed.), Manual of Photogrammetry. American
Society for Photogrammetry and Remote Sensing, pp. 1187e1220.
Akçiz, S.O., Ludwig, L.G., Arrowsmith, J.R., Zielke, O., 2010. Century-long average
time intervals between earthquake ruptures of the San Andreas fault in the
Carrizo Plain, California. Geology 38, 787e790. http://dx.doi.org/10.1130/
G30995.1.
Antonellini, M., Tondi, E., Agosta, F., Aydin, A., Cello, G., 2008. Failure modes in deep-
water carbonates and the impact for fault development: Majella Mountain,
Central Apennines, Italy. Mar. Pet. Geol. 25, 1074e1096.
Arrowsmith, J.R., Zielke, O., 2009. Tectonic geomorphology of the San Andreas fault
zone from high resolution topography: an example from the cholame segment.
Geomorphology 113, 70e81. http://dx.doi.org/10.1016/j.geomorph.2009.01.002.
Banks, M.R., 1962. The Silurian and Devonian systems. J. Geol. Soc. Aust. 9, 177e188 .
Barazzetti, L., Remondino, F., Scaioni, M., 2010. Automation in 3D Reconstruction:
Results on Different Kinds of Close-Range Blocks. IAPRS&SIS.
Beach, A., 1975. The geometry of en-echelon vein arrays. Tectonophysics 28,
245e263.
Bevis, M., Hudnut, K., Sanchez, R., Toth, C., Grejner-Brzezinska, D., Kendrick, E.,
Caccamise, D., Raleigh, D., Zhou, H., Shan, S., Shindle, W., Yong, A., Harvey, J.,
Borsa, A., Ayoub, F., Shrestha, R., Carter, B., Sartori, M., Phillips, D., Coloma, F.,
2005. The B4 Project: Scanning the San Andreas and San Jacinto Fault Zones.
AGU Fall Meet. Abstr. 34, 01.
Birdseye, C.H., 1940. Stereoscopic Phototopographic mapping. Ann. Assoc. Am.
Geogr. 30, 1e24. http://dx.doi.org/10.1080/00045604009357193.
Blenkinsop, T.G., 2012. Visualizing structural geology: from Excel to google earth.
Comput. Geosci. 45, 52e56.
Bons, P.D., Elburg, M.A., Gomez-Rivas, E., 2012. A review of the formation of tectonic
veins and their microstructures. J. Struct. Geol. 43, 33e62.
Borsa, A., Minster, J.-B., 2012. Rapid determination of Near-Fault earthquake
deformation using differential LiDAR. Bull. Seismol. Soc. Am. 102, 1335e1347.
http://dx.doi.org/10.1785/0120110159.
Brooks, B.A., 2014. Briefing on the M6.0 south Napa earthquake. In: Southern Cal-
ifornia Earthquake Center Annual Meeting, September 6-10, 2014. Palm Springs,
California.
Buckley, D., 2013. An Analysis of Fault Damage at Piccaninny Point, North-East
Tasmania. Unpublished Hons thesis. University of Western Australia.
Coggan, J.S., Wetherelt, A., Gwynn, X.P., Flynn, Z., 2007. Comparison of hand-
mapping with remote data capture systems for effective rock mass character-
isation. In: Proceedings of 11th Congress of the International Society for Rock
Mechanics - the Second Half Century of Rock Mechanics, vol. 1, pp. 201e205.
De Paor, D.G., Whitmeyer, S.J., 2011. Geological and geophysical modeling on virtual
globes using KML, COLLADA, and Javascript. Comput. Geosci. 37, 100e110 .
Dolan, J.F., Haravitch, B.D., 2014. How well do surface slip measurements track slip
at depth in large strike-slip earthquakes? the importance of fault structural
maturity in controlling on-fault slip versus off-fault surface deformation. Earth
Planet. Sci. Lett. 388, 38e47. http://dx.doi.org/10.1016/j.epsl.2013.11.043.
Duffy, B., Quigley, M., Barrell, D.J.A., Dissen, R.V., Stahl, T., Leprince, S., McInnes, C.,
Bilderback, E., 2013. Fault kinematics and surface deformation across a releasing
bend during the 2010 MW 7.1 Darfield, New Zealand, earthquake revealed by
differential LiDAR and cadastral surveying. Geol. Soc. Am. Bull. 125, 420e431.
http://dx.doi.org/10.1130/B30753.1.
Eardley, A.J., 1942. Aerial Photographs: Their Use and Interpretation. Harper and
Brothers, New York.
Farenzena, M., Fusiello, A., Gherardi, R., 2009. Structure-and-motion pipeline on a
hierarchical cluster tree, In: 2009 IEEE 12th International Conference on
Computer Vision Workshops (ICCV Workshops). In: Presented at the 2009 IEEE
12th International Conference on Computer Vision Workshops (ICCV Work-
shops), pp. 1489e1496. http://dx.doi.org/10.1109/ICCVW.2009.5457435.
Favalli, M., Fornaciai, A., Isola, I., Tarquini, S., Nannipieri, L., 2012. Multiview 3D
reconstruction in geosciences. Comput. Geosci. 44, 168e176 . http://dx.doi.org/
10.1016/j.cageo.2011.09.012.
Feng, Q., Sj€
ogren, P., Stephansson, O., Jing, L., 2001. Measuring fracture orientation at
exposed rock faces by using a non-reflector total station. Eng. Geol. 59,
133e146.
Fern
andez, O., 2005. Obtaining a best fitting plane through 3D georeferenced data.
J. Struct. Geol. 27, 855e858.
Fischler, M.A., Bolles, R.C., 1981. Random sample consensus: a paradigm for model
fitting with applications to image analysis and automated cartography. Com-
mun. ACM 24, 381e395.
Fonstad, M.A., Dietrich, J.T., Courville, B.C., Jensen, J.L., Carbonneau, P.E., 2013.
Topographic structure from motion: a new development in photogrammetric
measurement. Earth Surf. Process. Landf. 38, 421e430. http://dx.doi.org/
10.1002/esp.3366.
Furukawa, Y., Ponce, J., 2009. Accurate, dense, and robust multi-view stereopsis.
IEEE Trans. Pattern Anal. Mach. Intell. 32, 1362e1376.
Furukawa, Y., Curless, B., Seitz, S.M., Szeliski, R., 2010. Towards Internet-scale multi-
view stereo. In: 2010 IEEE Conference on Computer Vision and Pattern
Recognition (CVPR), pp. 1434e1441. http://dx.doi.org/10.1109/
CVPR.2010.5539802. Presented at the 2010 IEEE Conference on Computer
Vision and Pattern Recognition (CVPR).
Gee, R.D., Groves, D.I., 1971. Structural features and mode of emplaceme nt of
part of the Blue Tier Batholith in Northeast Tasmania. J. Geol. Soc. Aust. 18,
41e56.
Gillespie, P., Monsen, E., Maerten, L., Hunt, D., Thurmond, J., Tuck, D., 2010. Fractures
in carbonates: from digital outcrops to mechanical models. In: Martinsen, O.J.,
Pulham, A.J., Haughton, P., Sullivan, M.D. (Eds.), Outcrops Revitalized: Tools,
Techniques and Applications. SEPM (Society for Sedimentary Geology).
Gold, P.O., Cowgill, E., Kreylos, O., Gold, R.D., 2012. A terrestrial lidar-based work-
flow for determining Threeedimensional slip vectors and associated un-
certainties. Geosphere 8, 431e442. http://dx.doi.org/10.1130/GES00714.1.
Granshaw, S.I., 1980. Bundle adjustment methods in engineering photogrammetry.
Photogramm. Rec. 10, 181e207.
Groves, D.I., Cocker, J.D., Jennings, D.J., 1977. The geology, geochemistry and min-
eralisation of the Blue Tier Batholith. Geol. Surv. Tasman. Bull. 55, 7e116.
Haddad, D.E., Akçiz, S.O., Arrowsmith, J.R., Rhodes, D.D., Oldow, J.S., Zielke, O.,
Tok
e, N.A., Haddad, A.G., Mauer, J., Shilpakar, P., 2012. Applications of airborne
and terrestrial laser scanning to paleoseismology. Geosphere 8, 771e786.
http://dx.doi.org/10.1130/GES00701.1.
S.P. Bemis et al. / Journal of Structural Geology 69 (2014) 163e178 177

Haeussler, P.J., Schwartz, D.P., Dawson, T.E., Stenner, H.D., Lienkaemper, J.J.,
Sherrod, B., Cinti, F.R., Montone, P., Craw, P.A., Crone, A.J., Personius, S.F., 2004.
Surface rupture and slip distribution of the denali and Totschunda faults in the
3 november 2002 M 7.9 earthquake, Alaska. Bull. Seismol. Soc. Am. 94,
S23eS52. http://dx.doi.org/10.1785/0120040626.
Hagan, T.O., 1980. A case for terrestrial photogrammetry in deep-mine rock struc-
ture studies. Int. J. Rock Mech. Min. Sci. Geomech. Abstr. 17, 191e198. http://
dx.doi.org/10.1016/0148-9062(80)91085-2.
Haugerud, R.A., Harding, D.J., Johnson, S.Y., Harless, J.L., Weaver, C.S., Sherrod, B.L.,
2003. High-resolution lidar topography of the Puget Lowland, Washington - a
bonanza for earth science. GSA Today 13, 4e10.
Hilley, G.E., DeLong, S., Prentice, C., Blisniuk, K., Arrowsmith, J., 2010. Morphologic
dating of fault scarps using airborne laser swath mapping (ALSM) data. Geo-
phys. Res. Lett. 37, L04301. http://dx.doi.org/10.1029/2009GL042044.
Hodgetts, D., 2013. Laser scanning and digital outcrop geology in the petroleum
industry: a review. Mar. Pet. Geol. 46, 335e354. http://dx.doi.org/10.1016/
j.marpetgeo.2013.02.014.
Hudnut, K.W., Borsa, A., Glennie, C., Minster, J.-B., 2002. High-resolution topography
along surface rupture of the 16 October 1999 Hector mine, California, earth-
quake (Mw 7.1) from airborne laser swath mapping. Bull. Seismol. Soc. Am. 92,
1570 e1576. http://dx.doi.org/10.1785/0120000934.
James, M.R., Robson, S., 2012. Straightforward reconstruction of 3D surfaces and
topography with a camera: accuracy and geoscience application. J. Geophys.
Res. 117, F03017. http://dx.doi.org/10.1029/2011JF002289.
James, M.R., Robson, S., 2014. Mitigating systematic error in topographic models
derived from UAV and ground-based image networks. Earth Surf. Process.
Landf. 39, 1413e1420. http://dx.doi.org/10.1002/esp.3609.
Johnson, K., Nissen, E., Saripalli, S., Arrowsmith, J.R., McGarey, P., Scharer, K.,
Williams, P., Blisniuk, K., 2014. Rapid mapping of ultrafine fault zone topog-
raphy with structure from motion. Geosphere. http://dx.doi.org/10.1130/
GES01017.1. GES01017.1.
Krauss, K., 1993. Photogrammetry. In: Fundamentals and Standard Processes, vol. 1.
Dümmlers.
Krishnan, A.K., Saripalli, S., Nissen, E., Arrowsmith, R., 2012. 3D change detection
using low cost aerial imagery. In: 2012 IEEE International Symposium on Safety,
Security, and Rescue Robotics (SSRR), pp. 1e6. http://dx.doi.org/10.1109/
SSRR.2012.6523892. Presented at the 2012 IEEE International Symposium on
Safety, Security, and Rescue Robotics (SSRR).
Kuenen, P.H., 1950. Stereoscopic projection for demonstration in geology, geo-
morphology, and other natural sciences. J. Geol. 58, 49e54. http://dx.doi.org/
10.1086/625694.
Lowe, D.G., 2004. Distinctive image features from scale-invariant keypoints. Int. J.
Comput. Vis. 60, 91e110. http://dx.doi.org/10.1023/B: VISI.00 000296
64.99615.94.
McCaffrey, K.J.W., Jones, R.R., Holdsworth, R.E., Wilson, R.W., Clegg, P., Imber, J.,
Holliman, N., Trinks, I., 2005. Unlocking the spatial dimension: digital tech-
nologies and the future of geosciences fieldwork. J. Geol. Soc. Lond. 162,
927e938.
McCalpin, J., 2009. Paleoseismology, second ed. In: International Geophysics Aca-
demic Press.
Micklethwaite, S., Turner, D., Vasuki, Y., Kovesi, P., Holden, E.-J., Lucieer, A., 2012.
Mapping from an Armchair: rapid, high-resolution mapping using UAV and
computer vision technology. Struct. Geol. Resour. 130e
133.
Moreels, P., Perona, P., 2007. Evaluation of features detectors and descriptors based
on 3D objects. Int. J. Comput. Vis. 73, 263e284. http://dx.doi.org/10.1007/
s11263-006-9967-1.
Morelan, A.E., Stock, J.M., Hudnut, K.W., Akciz, S.O., 2010. Using photogrammetry to
produce high-resolution DEMs of the El Mayor-Cucapah surface rupture. In:
Southern California Earthquake Center Annual Meeting, September, 2010. Palm
Springs, California.
Niethammer, U., James, M.R., Rothmund, S., Travelletti, J., Joswig, M., 2012. UAV-
based remote sensing of the Super-Sauze landslide: evaluation and results. Eng.
Geol., Integration Technol. Landslide Monit. Quant. Hazard Assess. 128, 2e11.
http://dx.doi.org/10.1016/j.enggeo.2011.03.012.
Nissen, E., Krishnan, A.K., Arrowsmith, J.R., Saripalli, S., 2012. Three-dimensional
surface displacements and rotations from differencing pre- and post-
earthquake LiDAR point clouds. Geophys. Res. Lett. 39, n/aen/a. http://
dx.doi.org/10.1029/2012GL052460.
Nixon, C.W., Sanderson, D.J., Bull, J.M., 2011. Deformation within a strike-slip fault
network at Westward Ho!, Devon U.K.: domino vs conjugate faulting. J. Struct.
Geol. 33, 833e843.
Olson, J.E., Pollard, D.D., 1991. The initiation and growth of en 
echelon veins.
J. Struct. Geol. 13, 595e608.
Oskin, M.E., Le, K., Strane, M.D., 2007. Quantifying fault-zone activity in arid envi-
ronments with high-resolution topography. Geophys. Res. Lett. 34, L23S05.
http://dx.doi.org/10.1029/2007GL031295.
Oskin, M.E., Arrowsmith, J.R., Corona, A.H., Elliott, A.J., Fletcher, J.M., Fielding, E.J.,
Gold, P.O., Garcia, J.J.G., Hudnut, K.W., Liu-Zeng, J., Teran, O.J., 2012. Near-field
deformation from the El MayoreCucapah earthquake revealed by differential
LIDAR. Science 335, 702e705. http://dx.doi.org/10.1126/science.1213778.
Peacock, D.C.P., Fisher, Q.J., Willemse, E.J.M., Aydin, A., 1998. The relationship be-
tween faults and pressure solution seams in carbonate rocks and implications
for fluid flow. Geol. Soc. Lond. Special Publ. 147, 105e115.
Philibosian, B., Fumal, T.E., Weldon, R.J., Kendrick, K.J., Scharer, K.M., Bemis, S.P.,
Burgette, R.J., Wisely, B.A., 2009. Photomosaics and Logs of Trenches on the San
Andreas Fault Near Coachella. California (Open-File Report No. 2009-1039). U.S.
Geological Survey.
Pillmore, C.L., 1964. Application of high-order stereoscopic plotting instruments to
photogeologic studies. In: Procedures and Studies in Photogeology, Geological
Survey Bulletin 1043-B. U.S. Department of the Interior, Geological Survey,
Washington, D.C, pp. 22e34.
Pringle, J.K., Howell, J.A., Hodgetts, D., Westerman, A.R., Hodgson, D.M., 2006. Vir-
tual outcrop models of petroleum reservoir analogues: a review of the current
state-of-the-art. First Break 24, 33e42.
Rarity, F., van Lanen, X.M.T., Hodgetts, D., Gawthorpe, R.L., Wilson, P., Fabuel-
Perez, I., Redfern, J., 2013. LiDAR-based Digital Outcrops for Sedimentological
Analysis: Workflows and Techniques. In: Geological Society of London Special
Publications 387. http://dx.doi.org/10.1144/SP387.5.
Remondino, F., 2006. Detectors and descriptors for photogrammetric applications.
In: F€
orstner, W., Steffen, R. (Eds.), International Archives of the Photogram-
metry, Remote Sensing and Spatial Information Sciences: Symposium of ISPRS
Commission III Photogrammetric Computer Vision PCV ' 06, pp. 49e54. Bonn,
Germany.
Seers, T., Hodgetts, D., 2013. Comparison of Digital Outcrop and Conventional Data
Collection Approaches for the Characterization of Naturally Fractured Reservoir
Analogues. In: Geological Society of London Special Publication 374. http://
dx.doi.org/10.1144/SP374.13.
Shipton, Z.K., Cowie, P.A., 2001. Damage zone and slip-surface evolution over um to
km scales in high-porosity Navajo sandstone, Utah. J. Struct. Geol. 23,
1825e1844 .
Smith, M.J., Chandler, J., Rose, J., 2009. High spatial resolution data acquisition for
the geosciences: kite aerial photography. Earth Surf. Process. Landforms 34,
155e161. http://dx.doi.org/10.1002/esp.1702.
Snavely, N., 2010. Bundler: Structure from Motion for Unordered Image Collections.
http://phototour.cs.washington.edu/bundler/.
Snavely, N., Seitz, S.M., Szeliski, R., 20 06. Photo tourism: exploring photo collections
in 3D. In: ACM SIGGRAPH 2006 Papers, SIGGRAPH ’06. ACM, New York, NY, USA,
pp. 835e846. http://dx.doi.org/10.1145/1179352.1141964.
Snavely, N., Garg, R., Seitz, S.M., Szeliski, R., 2008a. Finding paths through the
World's photos. In: ACM SIGGRAPH 2008 Papers, SIGGRAPH '08. ACM, New
York, NY, USA, pp. 15:1e15:11. http://dx.doi.org/10.1145/1399504.1360614.
Snavely, N., Seitz, S.M., Szeliski, R., 2008b. Modeling the world from internet photo
collections. Int. J. Comput. Vis. 80, 189e210. http://dx.doi.org/10.1007/s11263-
007-0107-3.
Stumpf, A., Malet, J.-P., Kerle, N., Niethammer, U., Rothmund, S., 2013. Image-based
mapping of surface fissures for the investigation of landslide dynamics. Geo-
morphology 186, 12e27. http://dx.doi.org/10.1016/j.geomorph.2012.12.010.
Turner, D., Lucieer, A., Watson, C., 2012. An automated technique for generating
georectified mosaics from ultra-high resolution unmanned aerial vehicle (UAV)
imagery, based on structure from motion (SfM) Point clouds. Remote Sens. 4,
1392e1410.
Turner, D., Lucieer, A., Wallace, L., 2013. Direct georeferencing of ultrahigh-
resolution UAV imagery. IEEE Transactions on Geoscience and Remote
Sensing 52, 2738e2745. http://dx.doi.org/10.1109/TGRS.2013.2265295.
van Lanen, X.M.T., Hodgetts, D., Redfern, J., Fabuel-Perez, I., 2009. Applications of
digital outcrop models: two fluvial case studies from the Triassic Wolfville Fm.,
Canada and Oukaimeden Sandstone Fm., Morocco. Geol. J. 44, 742e760. http://
dx.doi.org/10.1002/gj.1196.
Vasuki, Y., Holden, E.-J., Kovesi, P., Micklethwaite, S., 2013. A geological structure
mapping tool using photogrammetric data. In: ASEG Extended Abstracts, pp. 1e4.
Vasuki, Y., Holden, E.-J., Kovesi, P., Micklethwaite, S., 2014. Semi-automatic mapping
of geological Structures using UAV-based photogrammetric data: an image
analysis approach. Comput. Geosci. 69, 22e32. http://dx.doi.org/10.1016/j.
cageo.2014.04.012.
Wackrow, R., Chandler, J.H., 2011. Minimising systematic error surfaces in digital
elevation models using oblique convergent imagery. Photogramm. Rec. 26,
16e31.
Westoby, M.J., Brasington, J., Glasser, N.F., Hambrey, M.J., Reynolds, J.M., 2012.
“Structure-from-Motion”photogrammetry: a low-cost, effective tool for geo-
science applications. Geomorphology 179, 300e314. http://dx.doi.org/10.1016/
j.geomorph.2012.08.021.
Willemse, E.J.M., Peacock, D.C.P., Aydin, A., 1997. Nucleation and growth of strike-
slip faults in limestones from Somerset, U.K. J. Struct. Geol. 19, 1461e1477.
Wilson, P., Gawthorpe, R.L., Hodgetts, D., Rarity, F., Sharp, I.R., 2009. Geometry and
architecture of faults in a syn-rift normal fault array: the Nukhul half-graben,
Suez rift, Egypt. J. Struct. Geol. 31, 759e775.
Zielke, O., Arrowsmith, J.R., Ludwig, L.G., Akçiz, S.O., 2010. Slip in the 1857 and
earlier large earthquakes along the carrizo Plain, San Andreas fault. Science 327,
1119 e1122. http://dx.doi.org/10.1126/science.1182781.
S.P. Bemis et al. / Journal of Structural Geology 69 (2014) 163e178178