Thermal Photogrammetric Imaging: A New Technique for Monitoring Dome Eruptions

Abstract

Structure-from-motion (SfM) algorithms greatly facilitate the generation of 3-D topographic models from photographs and can form a valuable component of hazard monitoring at active volcanic domes. However, model generation from visible imagery can be prevented due to poor lighting conditions or surface obscuration by degassing. Here, we show that thermal images can be used in a SfM workflow to mitigate these issues and provide more continuous time-series data than visible counterparts. We demonstrate our methodology by producing georeferenced photogrammetric models from 30 near-monthly overflights of the lava dome that formed at Volcán de Colima (Mexico) between 2013 and 2015. Comparison of thermal models with equivalents generated from visible-light photographs from a consumer digital single lens reflex (DSLR) camera suggests that, despite being less detailed than their DSLR counterparts, the thermal models are more than adequate reconstructions of dome geometry, giving volume estimates within 10% of those derived using the DSLR.

Significantly, we were able to construct thermal models in situations where degassing and poor lighting prevented the construction of models from DSLR imagery, providing substantially better data continuity than would have otherwise been possible. We conclude that thermal photogrammetry provides a useful new tool for monitoring effusive volcanic activity and assessing associated volcanic risks.

Full text

http://dx.doi.org/10.1016/j.jvolgeores.2017.03.022
Thermal Photogrammetric Imaging: A New Technique for Monitoring
1
Dome Eruptions
2
Samuel T. Thielea,b*, Nick Varleya & Mike R. Jamesc
3
4
aColima Intercambio e Investigación en Vulcanología, Universidad de Colima, av. Universidad 333, Las
5
Viboras C.P. 28040, Colima, México
6
bSchool of Earth, Atmosphere and Environment, Monash University, Clayton VIC 3800, Australia
7
cLancaster Environment Centre, Lancaster University, Lancaster LA1 4YQ, UK
8
*Corresponding Author: sam.thiele01@gmail.com
9
Abstract
10
Structure-from-motion (SfM) algorithms greatly facilitate the generation of 3-D topographic models
11
from photographs and can form a valuable component of hazard monitoring at active volcanic domes.
12
However, model generation from visible imagery can be prevented due to poor lighting conditions or
13
surface obscuration by degassing. Here, we show that thermal images can be used in a SfM workflow
14
to mitigate these issues and provide more continuous time-series data than visible-light equivalents.
15
We demonstrate our methodology by producing georeferenced photogrammetric models from 30
16
near-monthly overflights of the lava dome that formed at Volcán de Colima (Mexico) between 2013
17
and 2015. Comparison of thermal models with equivalents generated from visible-light photographs
18
from a consumer digital single lens reflex (DSLR) camera suggests that, despite being less detailed than
19
their DSLR counterparts, the thermal models are more than adequate reconstructions of dome
20
geometry, giving volume estimates within 10% of those derived using the DSLR.
21
Significantly, we were able to construct thermal models in situations where degassing and poor
22
lighting prevented the construction of models from DSLR imagery, providing substantially better data
23

http://dx.doi.org/10.1016/j.jvolgeores.2017.03.022
continuity than would have otherwise been possible. We conclude that thermal photogrammetry
24
provides a useful new tool for monitoring effusive volcanic activity and assessing associated volcanic
25
risks.
26
Key words: Lava Dome, Photogrammetry, Thermal Imaging, Volcán de Colima
27
1. Introduction
28
Lava domes are known to pose significant volcanic hazards, due to their tendency to generate collapse
29
related pyroclastic flows and their association with explosive eruptions (Fink and Anderson, 2000). For
30
example, successive dome collapses at Soufrière Hills on the island of Montserrat, starting in 1995,
31
caused the evacuation and eventual abandonment of the capital Plymouth and surrounding areas
32
(Wadge et al., 2014), while the 1994 collapse of Mount Merapi (Indonesia) resulted in 95 deaths and
33
damage to several villages (Abdurachman et al., 2000). A similar event at Volcán de Colima in 2015
34
generated pyroclastic flows that travelled ~10 km, fortunately causing only minor damage.
35
Monitoring of dome geometry (e.g. volume and height), growth rate and deformation is key to
36
forecasting such dome collapse events (Voight, 2000), and photogrammetry and structure from
37
motion (SfM) are increasingly being used for this purpose (e.g. Herd et al., 2005; Ryan et al., 2010;
38
Diefenbach et al., 2012; James and Varley, 2012; Diefenbach et al., 2013). Using these techniques,
39
morphological and geometric data can be safely and inexpensively acquired, and used to track
40
eruption progress, identify signs of instability or changes in effusion rate, and forecast changes in
41
volcanic risk. These methods, however, rely on clear viewing conditions and so are highly sensitive to
42
degassing, cloud and poor lighting conditions.
43
Thermal imaging techniques are also widely used for monitoring purposes (Spampinato et al., 2011),
44
as they allow quantitative evaluation of heat flux from volcanic vents (e.g. Harris and Stevenson, 1997;
45
Sahetapy-Engel et al., 2008), domes (e.g. Hutchison et al., 2013; Pallister et al., 2013), flows (e.g.
46
Calvari et al., 2003; James et al., 2006) and fumaroles (e.g. Stevenson and Varley, 2008; Harris et al.,
47

http://dx.doi.org/10.1016/j.jvolgeores.2017.03.022
2009). Importantly, the spatial distribution of heat flux can reveal features that are difficult to detect
48
using reflected visible light, such as fumaroles, fractures and rock fall traces (Hutchison et al., 2013;
49
Mueller et al., 2013).
50
Changes in the distribution and intensity of thermal anomalies can also precede volcanic eruptions or
51
changes in eruptive style (Spampinato et al., 2011) and thus have potential for hazard forecasting.
52
However, to facilitate inter-survey comparisons, thermal data need to be spatially referenced, and
53
producing orthorectified thermal maps usually requires additional topographic data, knowledge of the
54
camera location and viewing direction (e.g. James et al., 2006; James et al., 2009; Lewis et al., 2015).
55
This study demonstrates a method for deriving topographic data and georeferenced thermal maps
56
directly from oblique thermal imagery using SfM techniques and imagery captured during an episode
57
of dome growth at Volcán de Colima (Mexico) between 2013 and 2015. We suggest that the resulting
58
three-dimensional thermal models provide intuitive and georeferenced representations of dome
59
surface temperature and valuable measurements of dome geometry. Furthermore, we demonstrate
60
that despite the lower spatial resolution of thermal images, dome volume estimates are comparable
61
to those estimated using SfM reconstructions deriving from visible-light digital single lens reflex (DSLR)
62
photographs, and that unlike the DSLR models, the thermal models can be constructed during periods
63
of poor lighting and extensive degassing.
64
Volcán de Colima is an andesitic and frequently erupting stratovolcano, located at the western limit
65
of the Trans-Mexican Volcanic Belt. During the most recent eruptive periods, six episodes of dome
66
growth have been observed at the volcano (1998–1999, 2001–2003, 2004, 2007–2011, 2013–2015
67
and an ongoing episode initiated in February 2016). This represents the most active period at the
68
volcano since its last catastrophic eruption in 1913. A range of effusion rates have been estimated,
69
with the longer-lived eruptions associated with rates as low as 0.01 m3 s-1. During the current eruption
70
the volcano has exhibited the continuous generation of small Vulcanian explosions with a frequency
71

http://dx.doi.org/10.1016/j.jvolgeores.2017.03.022
of the order of hours. Larger magnitude explosions usually follow periods of dome emplacement,
72
which re-excavate the summit crater.
73
The episode of dome growth investigated in this study began in January 2013 when lava erupted into
74
the base of a ~150 m wide and ~50 m deep crater formed (by several large explosions that same
75
month) on top of a previous (2007 to 2011) lava dome. The new dome proceeded to fill this crater and
76
by April 2013 overflowed to form several lava flows and eventually fill the entire summit crater (~300
77
m across). Several partial collapses (accompanied by increased volcanic activity) resulted in dome
78
destruction on 10 – 11 July 2015; pyroclastic density currents generated by these collapses travelled
79
up to ~10.6 km along the ravine of Montegrande, threatening several ranches and the town of
80
Quesaría (pop. 8611 in 2010). This eruption was the largest (by volume) at Volcán de Colima since
81
1913.
82
2. Methods
83
2.1. Image capture and pre-processing
84
Images (Fig. 1) were acquired using a consumer DSLR (Nikon D90) and a thermal camera (Jenoptik
85
VarioCAM HiRes) from a light aircraft during 30 observation flights, conducted at intervals of
86
approximately one month. The DSLR had an 18–105 mm zoom lens (most images were captured using
87
the 105 mm setting), while the thermal camera used a 75 mm fixed-focal lens. Thermal images had an
88
order of magnitude lower resolution than the DSLR images (640×480 pixels and 4288×2848 pixels
89
respectively).
90
Observational flights involved 2–3 circuits around the crater at a slightly higher elevation than the
91
summit. Typical viewing distances varied between ~1–3 km, corresponding to ground sampling
92
distances of ~5-15 cm/pixel for the DSLR camera (at full zoom) and ~25-75 cm/pixel for the thermal
93
camera. Both cameras were operated by hand, with DSLR photographs captured every ~5–10 seconds
94
and the thermal camera programmed to take an image every 3.5 seconds.
95

http://dx.doi.org/10.1016/j.jvolgeores.2017.03.022
Blurry and poorly exposed images were manually removed from the resulting image sets (of ~100
96
DSLR images and ~200–400 thermal images) prior to photogrammetric processing. Normally, ~50–75
97
DSLR images and ~100–200 thermal images were considered usable, though this varied substantially
98
with viewing conditions.
99
The thermal images were converted from Jenoptic’s proprietary IRB format to JPEG (using a colour
100
scale selected to maximise the amount of detail visible on the dome and volcanic flanks) before
101
photogrammetric processing. A second set of JPEG images were additionally created from the thermal
102
images using a fixed colour scale, and later projected onto the photogrammetric model to create a
103
thermal texture map that can be compared between models.
104
105
Figure 1. Examples of typical DSLR (left) and thermal (right) images from two different observation
106
flights. Both views are looking to the north-west, and the summit region is ~300 m across.
107
2.2. Structure from Motion
108

http://dx.doi.org/10.1016/j.jvolgeores.2017.03.022
Photogrammetric processing of both the DSLR and thermal datasets was performed using Agisoft
109
Photoscan Professional Edition (v1.2.3). Prior to 3D reconstruction, both photosets were masked to
110
remove degassing plumes, aircraft parts and unnecessary background, ensuring that only the edifice
111
region was reconstructed. SfM methods were applied to estimate camera locations, orientations and
112
internal parameters and produce a ‘sparse point cloud’ containing the location of tens of thousands
113
of automatically detected features. These data were then used to constrain a detailed reconstruction
114
of the volcanic edifice, producing a ‘dense point cloud’ typically containing 10 – 20 million points for
115
the DSLR models and ~0.5 million points for the thermal models.
116
Finally, a continuous triangulated surface model was derived from the dense point cloud for image
117
rendering. For practical reasons, we limited the model to 1 million triangles, prior to texturing by
118
projecting the original images onto its surface. For the thermal models, the photoset used to construct
119
the model was exchanged with the photoset with a consistent colour-scale prior to the texturing step.
120
2.3. Georeferencing and Alignment
121
Due to difficult access and high risk, ground control points were not available for any of the models.
122
Instead, similar to the approach used by James and Varley (2012), models generated from the DSLR
123
images were georeferenced (within Agisoft Photoscan) by minimising the distance between features
124
identified on the models and equivalent features located in Google Earth imagery. Here, we
125
additionally used 1-arc second SRTM (Shuttle Radar Topographic Mission) data from February 2000 to
126
derive elevations. As the morphology of the summit area changed substantially over the study period,
127
it was necessary to use Google Earth imagery from different dates for some models, causing relative
128
translations of the results (reflecting the georeferencing error within Google Earth; coordinates of
129
some static features changed by >20 meters between imagery from different dates).
130
To improve the registration between models and facilitate direct comparisons, the relative
131
georeferencing of each model was optimised by aligning to one reference model (from 11 January
132
2013) using the iterative closest point (ICP) alignment implementation in Cloud Compare (Girardeau-
133

http://dx.doi.org/10.1016/j.jvolgeores.2017.03.022
Montaut, 2015). Model location, orientation and scale was allowed to vary during this step, during
134
which areas known to have changed (i.e. the dome and associated flows) were manually excluded.
135
Models constructed using thermal images could not generally be georeferenced from the Google
136
Earth imagery due to difficulties identifying corresponding features in the thermal data. Instead, they
137
were aligned to the DSLR model from the same flight (or from a previous flight if the DSLR model had
138
failed), using a manual 3-D point-matching approach in Meshlab (Cignoni et al., 2008) to achieve an
139
initial alignment that was then optimised using ICP.
140
Where possible, the similarity (and alignment) of the DSLR and thermal models was assessed by
141
comparison with DSLR models generated from the same flight. As the ICP alignment algorithm only
142
applies a scaling and rigid body transformation, similarities between the DSLR and thermal models
143
suggest that the photogrammetric reconstructions converge on a consistent surface shape, adding
144
confidence to the results. Note that while this assessment provides an indication of uncertainty in the
145
overall model shapes, it cannot evaluate the full geospatial uncertainty because the thermal models
146
are not independently georeferenced.
147
2.4. Volume Calculation
148
Dome volume was estimated by determining the difference between each photogrammetric model
149
and the pre-dome reference model created photogrammetrically using data from a flight on 11
150
January 2013. The difference calculations (performed using a Java implementation of the signed
151
tetrahedral method; Zhang and Chen, 2001) determined the volume between the surfaces in up to
152
four ‘regions of interest’ (ROI; Fig. 2). In this instance, a ROI containing the lava dome was defined for
153
each of the photogrammetric models (both DSLR and thermal), and the volume of the dome estimated
154
by comparison with a reference surface representing the pre-dome topography. Where the dome
155
overflowed the crater (and transitioned into a lava flow), a consistent (but visually estimated) ‘dome
156
boundary’ was defined (the boundary between regions a and b in Fig. 2), and the volume of the upper
157
portion of a lava flow was also determined.
158

http://dx.doi.org/10.1016/j.jvolgeores.2017.03.022
In order to better evaluate the uncertainty of the volume estimates, change within two stable
159
reference areas on the flanks of the volcano was also calculated; because these areas should not vary,
160
detecting volume change within them suggests greater uncertainty in the topographic models or their
161
relative registration. These changes were expressed as mean vertical offsets that could then be used
162
to estimate the dome volume (positive or negative) that likely resulted from alignment errors. Note
163
however, that these reference areas were always located on the eastern flanks of the volcano, as the
164
western flanks changed substantially over the study period (due to lava flows), and hence are not
165
equally sensitive to all types of alignment error (e.g. translations or rotations) in dome area.
166
167
Figure 2: Oblique view of the ‘regions of interest’ defined for the 27/4/13 photogrammetric model.
168
Region (a) contains the growing lava dome, and (b) the incipient lava flow. Regions (c) and (d) are the
169
reference areas. The colour map represents the vertical distance between the comparison and
170
reference surfaces. The dome region (a) is ~140 m across.
171
3. Results and Discussion
172
The 30 survey flights allowed the construction of 19 usable models from visible imagery and 22 models
173
from the thermal imagery, although thermal data was only available for 23 flights. These datasets
174

http://dx.doi.org/10.1016/j.jvolgeores.2017.03.022
provide a reconstruction of the summit lava dome geometry at ~monthly intervals for the entire
175
dome-forming eruption.
176
3.1. Comparison of Thermal and DSLR models
177
Both sets of photogrammetric models (thermal and DSLR) reconstructed the crater and dome complex
178
on Volcán de Colima with varying degrees of completeness, detail and accuracy (Fig. 3). It is clear that,
179
in general, models constructed using the thermal images were substantially less detailed than DSLR
180
equivalents. This will be due to a combination of the thermal images having a lower spatial resolution
181
than the DSLR images and a lack of high-frequency image texture, due to low thermal contrast on the
182
volcano flanks.
183
184
185
Figure 3. Selected DSLR and thermal photogrammetric models illustrated by hillshade (top and middle),
186
and associated thermal orthomosaics (bottom). Grid cells are 50×50 m and oriented north-south and
187

http://dx.doi.org/10.1016/j.jvolgeores.2017.03.022
east-west. The model shown in (a) was captured before lava dome growth and was used as the
188
reference model in volume calculations.
189
Nevertheless, 3D reconstruction using the thermal images was found to be far more robust to poor
190
photography conditions than the DSLR models. In particular, thermal models could be constructed in
191
situations where degassing made useful reconstruction from the DSLR images impossible. This is
192
because water droplets in the degassing plume cause near complete scattering of visible light (and
193
hence the plumes appear white), whilst the thermal infrared radiation (7.5-14 μm) is less affected (Fig.
194
4). Of all flights for which both thermal and DSLR data was available, ~30% of the DSLR surveys failed
195
to generate a model while only ~5% of the thermal models failed, even though image locations and
196
overlap were approximately the same. Hence, in addition to providing a useful map of estimated
197
temperature across the crater complex, the thermal models provide greater data continuity than the
198
models from the DSLR.
199
200
Figure 4. Thermal (a) and DSLR (b) images captured at approximately the same time and location
201
(looking towards the east), under strong degassing conditions. The dome is generally resolvable in the
202
thermal image, but is substantially obscured in the DSLR image. A photogrammetric model of the dome
203
was successfully reconstructed from the thermal images, and is of particular importance as it was
204
captured on 5/7/15, days before the major July 2015 eruption. A model was not attempted using the
205
DSLR data due to the degassing.
206

http://dx.doi.org/10.1016/j.jvolgeores.2017.03.022
Shortest distance comparisons between associated DSLR and thermal models show generally good
207
agreement (Fig. 5a and b). As thermal models tend to be smoother than the DSLR models (Fig. 3),
208
differences tend to be focused around sharp topographic features such as the crater rim. However,
209
in a few cases, the thermal models did differ significantly from their DSLR counterparts (Table 1). The
210
largest dome volume difference was observed at the time when the dome area was largest (Fig. 5b),
211
but the second largest observed dome volume difference resulted from the thermal model locating
212
the dome surface ~5 m higher than the DSLR model (Fig. 5c). The reason for this difference is
213
unclear, but highlights our ability to identify uncertainty by comparing the different datasets. A few
214
of the thermal models also contained substantial error (±10 m; Fig. 5d), which was mostly apparent
215
in areas of low thermal contrast, where image alignments and surface reconstructions are likely to
216
be weakest. Although the active lava surfaces were not directly influenced by this effect, the noisy
217
surfaces did impair the ICP process and probably increased registration error.
218
219
Figure 5. The shortest distance between corresponding DSLR and thermal models. Regions where the
220
thermal model is above the DSLR model are yellow-red, while areas where the DSLR model is on top
221
are shaded green-blue. Histograms showing the distribution of the difference values are shown below
222
each map. Examples of typical models are shown in (a) and (b), with few large differences except along
223
sharp features (e.g. the crater rim) and towards model boundaries. Examples of models showing
224
greater differences are presented in (c) and (d), where reconstructed dome geometries do not match
225
well (c) or where substantial error is present (d).
226

http://dx.doi.org/10.1016/j.jvolgeores.2017.03.022
Table 1: The five largest differences between the thermal and DSLR volume estimates. Volumes and
227
differences are in million m3. Percentages are relative to the DSLR volume estimate.
228
Model
DSLR Volume
Thermal Volume
Difference
% Difference
7/06/2015
1.16
1.05
0.11
9%
4/02/2015
1.10
1.20
0.10
9%
20/06/2013
0.51
0.56
0.05
9%
14/02/2014
0.51
0.55
0.04
7%
2/12/2013
0.54
0.52
0.02
3%
229
Using consumer cameras, and in the absence of ground control points sufficient to help constrain
230
photogrammetric processing, SfM-based data have been previously shown to provide topographic
231
data with an overall precision of ~1/1000 of the viewing distance (James and Robson, 2012). Thus,
232
over viewing distances of ~1-3 km, the 1-5 m differences between models are in line with this rule of
233
thumb. These results are reasonable given the low resolution of the thermal camera and the relatively
234
narrow angular field of view of both cameras (12° for the DSLR camera at full zoom and 10° for the
235
thermal camera), which can cause difficulties for precise photogrammetric reconstruction. It is likely
236
that the orbital flight paths play a strong role in helping to reduce error by naturally providing
237
convergent imagery, which mitigates systematic model deformation effects when ground control
238
points cannot be incorporated into the photogrammetric processing (Wackrow and Chandler, 2008;
239
James and Robson, 2014).
240
3.2. Dome volume calculations
241
Dome volumes calculated independently using the DSLR and thermal models generally correspond
242
well (Fig. 6), and differ by <10%. Likewise, the volume difference within the reference areas tended to
243
be small, averaging 5% of the dome volume estimates.
244
While the volcanological significance and the implications of these results for understanding the 2013
245
– 2015 eruption are beyond the scope of this paper, it is clear that they provide valuable information
246
on phases of dome growth (and volume loss) at Volcán de Colima between 2013 and 2015 (Fig. 6).
247

http://dx.doi.org/10.1016/j.jvolgeores.2017.03.022
Average effusion rates could also be estimated from the rate of dome volume change, although the
248
effect of volume loss through explosive activity and lava flows would need to be accounted for.
249
Finally, where both DSLR and thermal models were successful, the two independent reconstructions
250
also provide a valuable indication of uncertainty in model shape. Future studies could extend this
251
approach and use GPS devices to “geotag” image locations at the time of capture, allowing additional
252
evaluation of georeferencing uncertainty as the thermal models would no longer rely on ICP
253
registration against a similar visible-light model for their georeferencing. For high quality camera
254
position data, this ‘direct georeferencing’ approach has been shown capable of delivering decametric
255
accuracies (Nolan et al., 2015). Alternatively, where sufficient topographic features are recognisable
256
in the thermal models, a single georeferenced model or high resolution digital elevation model of
257
known accuracy could be used for georeferencing, avoiding the need for closely associated DSLR
258
models.
259
260
Figure 6. Volume of the lava dome (dotted) and lava flow top (dashed) between initiation of dome
261
growth in January 2013 and dome collapse in July 2015. Where both DSLR (squares) and thermal
262

http://dx.doi.org/10.1016/j.jvolgeores.2017.03.022
(circles) models were available, the lines represent an average estimate. It is clear that there is
263
generally good agreement between volumes calculated with the DSLR and thermal models. Reference
264
area volumes (which would be zero under error-free conditions) are shown in grey to give an indication
265
of relative accuracy.
266
4. Conclusions
267
We have successfully used SfM techniques and oblique thermal images to produce a time-series of
268
georeferenced, three-dimensional thermal models of an active lava dome at Volcán de Colima.
269
Comparisons between these models and equivalents derived from DSLR images suggest that, while
270
less detailed, the thermal models provide a valuable representation of dome geometry. Estimates of
271
the lava dome volume correspond well between the DSLR and thermal datasets.
272
The thermal models were found to be substantially more robust to the adverse effects of degassing
273
and poor lighting. Because degassing is common at Volcán de Colima (as at many other volcanoes)
274
thermal imaging provided important data continuity at times when DSLR image quality was restricted.
275
Where both DSLR and thermal models were available, the thermal models provided a useful
276
complementary geometry estimate, helping to identify uncertainty in the models, and a
277
georeferenced map of temperature distribution that allows identification of thermally active regions
278
on the dome surface.
279
The combined DSLR and thermal datasets provided detailed information about the evolution of the
280
dome on Volcán de Colima between 2013 and 2015. It is possible that, if employed as a monitoring
281
technique (rather than retrospectively), the rapid change in dome volume, morphology and
282
temperature distribution documented by the models in the months leading up to July 2015 may have
283
provided prior warning of the dome collapse.
284
Acknowledgements
285

http://dx.doi.org/10.1016/j.jvolgeores.2017.03.022
The authors would like to acknowledge the multitude of past CIIV students who participated in data
286
collection for this study and helped to finance flights. Some flights were financed by NERC Urgency
287
Grant NE/L000741/1 (PI: Paul Cole). NV was supported by Universidad de Colima FRABA grants. Paul
288
Cole and an anonymous reviewer are thanked for their useful feedback during the review process.
289
References
290
Abdurachman, E., Bourdier, J.-L., Voight, B., 2000. Nuées ardentes of 22 November 1994 at Merapi
291
volcano, Java, Indonesia. Journal of Volcanology and Geothermal Research 100, 345-361.
292
Calvari, S., Neri, M., Pinkerton, H., 2003. Effusion rate estimations during the 1999 summit eruption
293
on Mount Etna, and growth of two distinct lava flow fields. Journal of Volcanology and Geothermal
294
Research 119, 107-123.
295
Cignoni, P., Corsini, M., Ranzuglia, G., 2008. Meshlab: an open-source 3d mesh processing system.
296
Ercim news 73, 45-46.
297
Diefenbach, A.K., Bull, K.F., Wessels, R.L., McGimsey, R.G., 2013. Photogrammetric monitoring of lava
298
dome growth during the 2009 eruption of Redoubt Volcano. Journal of Volcanology and Geothermal
299
Research 259, 308-316. http://dx.doi.org/10.1016/j.jvolgeores.2011.12.009.
300
Diefenbach, A.K., Crider, J.G., Schilling, S.P., Dzurisin, D., 2012. Rapid, low-cost photogrammetry to
301
monitor volcanic eruptions: an example from Mount St. Helens, Washington, USA. Bull Volcanol 74,
302
579-587.
303
Fink, J.H., Anderson, S.W., 2000. Lava domes and coulees. Encyclopedia of volcanoes, 307-319.
304
Girardeau-Montaut, D., 2015. Cloud Compare—3D Point Cloud and Mesh Processing Software. Open
305
Source Project.
306
Harris, A.J., Lodato, L., Dehn, J., Spampinato, L., 2009. Thermal characterization of the Vulcano
307
fumarole field. Bull Volcanol 71, 441-458.
308
Harris, A.J.L., Stevenson, D.S., 1997. Thermal observations of degassing open conduits and fumaroles
309
at Stromboli and Vulcano using remotely sensed data. Journal of Volcanology and Geothermal
310
Research 76, 175-198. http://dx.doi.org/10.1016/S0377-0273(96)00097-2.
311
Herd, R.A., Edmonds, M., Bass, V.A., 2005. Catastrophic lava dome failure at Soufrière Hills volcano,
312
Montserrat, 12–13 July 2003. Journal of Volcanology and Geothermal Research 148, 234-252.
313
Hutchison, W., Varley, N., Pyle, D.M., Mather, T.A., Stevenson, J.A., 2013. Airborne thermal remote
314
sensing of the Volcán de Colima (Mexico) lava dome from 2007 to 2010. Geological Society, London,
315
Special Publications 380, 203-228.
316
James, M.R., Pinkerton, H., Applegarth, L.J., 2009. Detecting the development of active lava flow fields
317
with a very‐long‐range terrestrial laser scanner and thermal imagery. Geophysical Research Letters
318
36. 10.1029/2009GL040701.
319
James, M.R., Robson, S., 2012. Straightforward reconstruction of 3D surfaces and topography with a
320
camera: Accuracy and geoscience application. Journal of Geophysical Research 117.
321
10.1029/2011jf002289.
322
James, M.R., Robson, S., 2014. Mitigating systematic error in topographic models derived from UAV
323
and ground‐based image networks. Earth Surface Processes and Landforms 39, 1413-1420.
324
10.1002/esp.3609.
325
James, M.R., Robson, S., Pinkerton, H., Ball, M., 2006. Oblique photogrammetry with visible and
326
thermal images of active lava flows. Bull Volcanol 69, 105-108. 10.1007/s00445-006-0062-9.
327
James, M.R., Varley, N., 2012. Identification of structural controls in an active lava dome with high
328
resolution DEMs: Volcán de Colima, Mexico. Geophysical Research Letters 39. 10.1029/2012gl054245.
329

http://dx.doi.org/10.1016/j.jvolgeores.2017.03.022
Lewis, A., Hilley, G.E., Lewicki, J.L., 2015. Integrated thermal infrared imaging and structure-from-
330
motion photogrammetry to map apparent temperature and radiant hydrothermal heat flux at
331
Mammoth Mountain, CA, USA. Journal of Volcanology and Geothermal Research 303, 16-24.
332
http://dx.doi.org/10.1016/j.jvolgeores.2015.07.025.
333
Mueller, S., Varley, N., Kueppers, U., Lesage, P., Davila, G.Á.R., Dingwell, D.B., 2013. Quantification of
334
magma ascent rate through rockfall monitoring at the growing/collapsing lava dome of Volcán de
335
Colima, Mexico. Solid Earth 4, 201.
336
Nolan, M., Larsen, C., Sturm, M., 2015. Mapping snow depth from manned aircraft on landscape scales
337
at centimeter resolution using structure-from-motion photogrammetry. The Cryosphere 9, 1445-
338
1463.
339
Pallister, J.S., Diefenbach, A.K., Burton, W.C., Muñoz, J., Griswold, J.P., Lara, L.E., Lowenstern, J.B.,
340
Valenzuela, C.E., 2013. The Chaitén rhyolite lava dome: Eruption sequence, lava dome volumes, rapid
341
effusion rates and source of the rhyolite magma. Andean Geology 40, 277-294.
342
Ryan, G., Loughlin, S., James, M., Jones, L., Calder, E., Christopher, T., Strutt, M., Wadge, G., 2010.
343
Growth of the lava dome and extrusion rates at Soufrière Hills Volcano, Montserrat, West Indies:
344
2005–2008. Geophysical Research Letters 37.
345
Sahetapy-Engel, S.T., Harris, A.J., Marchetti, E., 2008. Thermal, seismic and infrasound observations of
346
persistent explosive activity and conduit dynamics at Santiaguito lava dome, Guatemala. Journal of
347
Volcanology and Geothermal Research 173, 1-14.
348
Spampinato, L., Calvari, S., Oppenheimer, C., Boschi, E., 2011. Volcano surveillance using infrared
349
cameras. Earth-Science Reviews 106, 63-91.
350
Stevenson, J.A., Varley, N., 2008. Fumarole monitoring with a handheld infrared camera: Volcán de
351
Colima, Mexico, 2006–2007. Journal of Volcanology and Geothermal Research 177, 911-924.
352
Voight, B., 2000. Structural stability of andesite volcanoes and lava domes. Philosophical Transactions
353
of the Royal Society of London A: Mathematical, Physical and Engineering Sciences 358, 1663-1703.
354
Wackrow, R., Chandler, J.H., 2008. A convergent image configuration for DEM extraction that
355
minimises the systematic effects caused by an inaccurate lens model. The Photogrammetric Record
356
23, 6-18.
357
Wadge, G., Voight, B., Sparks, R., Cole, P., Loughlin, S., Robertson, R., 2014. An overview of the
358
eruption of Soufriere Hills Volcano, Montserrat from 2000 to 2010. Geological Society, London,
359
Memoirs 39, 1-40.
360
Zhang, C., Chen, T., 2001. Efficient feature extraction for 2D/3D objects in mesh representation, Image
361
Processing, 2001. Proceedings. 2001 International Conference on. IEEE, pp. 935-938.
362
363