*Manuscript Click here to view linked References
1 Structural data collection with mobile devices: 2 Accuracy, redundancy, and best practices
3 Richard W. Allmendinger*, Chris R. Siron, and Chelsea P. Scott1 4 Dept. of Earth and Atmospheric Sciences 5 Cornell University 6 Ithaca, NY 14850 USA
7 Keywords: Smartphone, compass, accuracy, redundancy,
8 Abstract 9 Smart phones are equipped with numerous sensors that enable orientation data
10 collection for structural geology at a rate up to an order of magnitude faster than tradi-
11 tional analog compasses. The rapidity of measurement enables structural geologists, for
12 the first time, to enjoy the benefits of data redundancy and quantitative uncertainty es-
13 timates. Recent work, however, has called into question the reliability of sensors on An-
14 droid devices. We present here our experience with programming a new smart phone
15 app from scratch, and using it and commercial apps on iOS devices along with analog
16 compasses in a series of controlled tests and typical field use cases. Additionally, we
17 document the relationships between a iPhone measurements and visible structures in
18 satellite, drawing on a database of 3,700 iPhone measurements of coseismic surface
19 cracks we made in northern Chile following the Mw8.1 Pisagua earthquake in 2014. By
20 comparing phone-collected attitudes to orientations determined independently of the
21 magnetic field, we avoid having to assume that the analog compass, which is subject to
22 its own uncertainties, is the canonical instrument. Our results suggest that iOS devices
*Corresponding author. Email address: [email protected]
1 Now at: School of Earth and Space Exploration, Arizona State University, Tempe, AZ, USA
-1-
23 are suitable for all but the most demanding applications as long as particular care is
24 taken with respect to metal objects that could affect the magnetic field.
25 1. Introduction 26 Structural geology has always suffered from a relatively small number of record-
27 ed, quantitative measurements. A field geologist working with a traditional analog
28 compass and paper field notebook typically records a few tens of orientation measure-
29 ments per day. Not only are the total measurements small in number, but repeat meas-
30 urements of sufficient quantity to establish statistical uncertainty are extremely rare in
31 the published literature. Why bother to make tens of measurement of a single bedding
32 surface at a single site when doing so would severely limit the number of outcrops we
33 could document in a day? As Ramsay and Huber wrote with respect to strain meas-
34 urements nearly a quarter century ago (Ramsay and Huber, 1983, p. 78), “it is often
35 more rewarding to spend time in the field collecting a lot of data of relatively low de-
36 gree of accuracy at many localities, rather than to concentrate on obtaining a few strain
37 data with an extremely high degree of accuracy.” However, without data redundancy,
38 it is impossible to evaluate the significance and accuracy of a measurement. Thus, the
39 ideal case would be to make lots of measurements of high accuracy and well established
40 uncertainty quickly enough that one can still visit many localities.
41 Smart phone compass/stereonet programs (“apps”) will inexorably replace tra-
42 ditional analog compasses (Brunton, Freiberg, Silva, etc.) because of convenience, cost,
43 and ubiquity but mostly because of their rapidity. The ability to record orientation
44 along with location (latitude and longitude or UTM) and date/time with a single tap of
45 an on-screen button reduces, by nearly an order of magnitude, the amount of time that
46 it takes to make a measurement. If many more measurements can be made in the same
-2-
47 period of time, then field geologists can begin to enjoy the benefits of data redundancy
48 that simply are not feasible with analog instruments. Additionally, in every cash-
49 strapped geology department, the question will be, or is already being, asked: “why
50 should we invest thousands of dollars in analog compasses when every student already
51 has a smart phone at their disposal?”
52 The enthusiasm for smart phone orientation measurement apps was recently
53 dampened somewhat by Novakova and Pavlis (2017) as well as earlier studies (e.g.,
54 Hama et al., 2014) that demonstrate considerable variation in quality of device sensors
55 and accuracy of recorded observation. Most conclude that the least reliable sensor is the
56 device magnetometer which likewise accords with our experience. However, compari-
57 sons in previous studies tend to be incomplete or misleading in assessment of accuracy
58 because: only one operating system (i.e., Android or iOS) is tested, multiple devices are
59 not used in testing, the details of the algorithms used in the apps are seldom well de-
60 scribed, the error in the dip is assessed separately from the error in the strike (rather
61 than using poles to planes), single measurements from analog compasses are regarded
62 as canonical rather than comparing multiple analog compass measurements to multiple
63 smart phone apps, and there is no assessment of accuracy that does not rely on the
64 magnetic field.
65 Here, our purpose is three fold: (1) introduce a new, free iOS app, Stereonet Mobi-
66 le, written by the senior author2 and document the basics of it’s functioning; (2) compare
67 Stereonet Mobile and a popular smart phone app, Fieldmove Clino, to each other and to
68 analog compass measurements on a datum by datum and group by group basis; and (3)
2 Because Stereonet Mobile is available for free from the iOS App Store, the senior author has no finan- cial conflict of interest.
-3-
69 document the accuracy of app measurements independent of analog compass meas-
70 urements by using a massive database of measurements of surface crack measurements
71 and comparing those measurements with the same cracks visible on Google Earth im-
72 agery.
73 Our study, using only Apple® iPhones (iOS operating system), stands in contrast
74 to Novakova and Pavlis (2017) who used only Android devices. Together, these two
75 studies tend to suggest that the four different iPhones devices used here are significant-
76 ly superior in accuracy and reliability compared to the two tested Android devices. This
77 conclusion has also been reached by Midland Valley, the publisher of Fieldmove Clino
78 who state, “We have observed much larger variations in the measured data recorded
79 using Android devices which we suspect is largely down to the quality of the hardware
80 components inside the device” (Midland Valley, 2017). Our data are also consistent with
81 the recent work of (Cawood et al., 2017) who compared remotely sensed surface data
82 (LiDAR, Structure from Motion) to both digital and analog compass readings. However,
83 anyone collecting irreplaceable field data with any electronic device will want to con-
84 duct their own tests and continue to carry an analog compass in the field with them.
85 Even where a device is not reliable for data collection, many apps allow manual data
86 entry, giving the user some of the benefits of the smart phone (e.g., automatic recording
87 of location, time, and date) without the uncertainty in accuracy.
88 2. Stereonet Mobile
89 2.1. Device Sensors 90 Smart phones have a vast array of sensors to determine device orientation in-
91 cluding GPS receivers, accelerometers, gyroscopes, magnetometers, and even barome-
92 ters. From these sensors, it is possible to determine device orientation, position, velocity,
-4-
93 and linear and rotational acceleration. The iOS operating system provides the pro-
94 grammer with this derived information through its CoreMotion routines which handle
95 the translation of the raw sensor data into the orientations that we as geologists want.
96 Foremost among these is magnetometer calibration which attempts to cancel out the ef-
97 fects of local magnetic fields, especially from other components within the device such
98 as the power supply, etc., so that the orientation with respect to magnetic north can be
99 determined. The basic reason that dip measurements collected with smart phones are
100 generally much more accurate than strikes is that the magnetometer is much more sen-
101 sitive to, and local perturbations more common in, the local magnetic field than in the
102 local gravity field. The dip of the device can be determined from the three components
103 of the acceleration due to gravity alone and does not have to depend on the magnetom-
104 eter at all.
105 Sensors in the iPhone, sampled by the Sensor Kinematics Pro app at about 30 Hz,
106 appear to be very stable (Fig. 1) especially in comparison to the Android devices tested
107 by Novakova and Pavlis (2017, their figure 2). Nonetheless, the iPhone magnetometer is
108 easily perturbed by passing even small metal objects within several centimeters of the
109 device (Fig. 2b). This has considerable implications for best practices in the field when
110 using phones as data collection devices.
111 2.2. Device Coordinate System and Determining Orientation 112 The iOS device coordinate system and the rotations about the three axes are
113 shown in Figure 2. One “reads” the face of the device like a right-handed map coordi-
114 nate system: the first axis, X1, is parallel to and in the short, or side-to-side, direction of
115 the face with positive to the right. The second axis, X2, is parallel to the face and the long
116 axis of the device with positive up, and X3, the third axis, is perpendicular to the face
-5-
117 and positive upwards. The change in orientations of the device is determined by the ro-
118 tation of this coordinate system with respect to a reference coordinate system.
119 The change in orientation is supplied to the programmer by iOS in several differ-
120 ent ways. Perhaps most common is using the Euler angles (Fig. 2), the pitch, roll, and
121 yaw (sometimes known as the Tait-Bryan angles), which are familiar to anyone in avia-
122 tion or boating. Determining device orientation using these angles, though, can be sub-
123 ject to an artifact known as gimbal lock where one degree of freedom is lost in certain
124 orientations. Thus, iOS also provides orientation information via a rotation matrix or via
125 quaternions. Stereonet Mobile uses the rotation matrix to calculate the orientation of the
126 device relative to the reference frame. The rotation matrix, r, in terms of the pitch roll
127 and yaw, for iOS is given as:
r11 = cos(roll)cos(yaw)−sin(roll)sin(pitch)sin(yaw)
r12 = cos(yaw)sin(roll)sin(pitch)+ cos(roll)sin(yaw)
r13 = −sin(roll)cos(pitch)
r21 = −cos(pitch)sin(yaw)
r22 = cos(pitch)cos(yaw)
r23 = sin(pitch)
r31 = cos(roll)sin(pitch)sin(yaw)+ cos(yaw)sin(roll)
r32 = sin(yaw)sin(roll)− cos(roll)cos(yaw)sin(pitch)
r33 = cos(roll)cos(pitch)
128
129 The basic form of these equations will look familiar to anyone who has studied how ro-
130 tations are accomplished in stereonet programs (e.g., Allmendinger et al., 2012) because
-6-
131 they represent a single rotation accomplished by performing, in order, the three rota-
132 tions about the three axes (i.e., three matrix multiplications).
133 The iOS operating system provides the programmer with four different potential
134 reference frames. Stereonet Mobile uses the “CMAttitudeReferenceFrameXTrue-
135 NorthZVertical” reference frame. That is, the matrix is equal to the identity matrix when
136 the phone face is horizontal with the short axis (X1) aligned NS. To determine true north
137 the operating system must know the device position on the globe in order to calculate
138 magnetic declination. Thus, reading an orientation must also turn on the device GPS
139 receiver which can, if left on, drain the device battery more quickly than just about any
140 other sensor.
141 The matrix, r, is an orthogonal transformation matrix between the the device co-
142 ordinate system and the North-East-Down (NED) coordinate system familiar to struc-
143 tural geologists. To translate device orientation to geological orientation, we simply cal-
144 culate the orientation of a unit vector parallel to X3 (i.e., the pole to the device) for
145 planes and another unit vector parallel to X2, the long axis of the device, for lines (Fig.
146 2). In terms of direction cosines in a NED coordinate system, the pole to the phone and
147 the geological surface against which it is held is given by:
pole.north = r31
pole.east = –r32
pole.down = –r33 148
149 Likewise, a lineation’s direction cosines are:
-7-
lineation.north = r21
lineation.east = −r22
lineation.down = −r23 150
151 Because Stereonet Mobile uses the pole to the device, the user can place the back of
152 the phone flush on the bedding surface in any orientation to measure the surface of in-
153 terest. We have not noted any significant variation in accuracy when the phone is held
154 in different positions, including upside-down. To measure a line, the long axis or edge
155 of the phone must be parallel to the lineation on the rock but the back of the phone need
156 not be flush against the rock. Stereonet Mobile can simultaneously measure the orienta-
157 tion of a plane and a line it contains by placing the back of the phone flush on the rock
158 with the long axis parallel to the lineation in the plane (Fig. 3a).
159 In cases where one would not want, or cannot, place the phone on the surface to
160 be measured, Stereonet Mobile is also capable of measuring a plane’s orientation by
161 sighting through the device camera (Fig. 3b). When making a sighting measurement
162 with the plane viewed edge-on, the pole to the device is assumed to be parallel to the
163 strike direction and the long axis of the device parallel to the true dip direction. For
164 sighting measurements made down-dip, the trend and plunge of the pole is assumed to
165 be equal to the dip azimuth and dip of the plane.
166 2.3. Redundant sampling 167 Novakova and Pavlis (2017) demonstrated that, at least for Android devices,
168 transients in the sensor data, brief marked excursions from the long term average value
169 of the sensor, are a serious issue but something that can be mitigated by over-sampling
170 to calculate a long term average. While transients appear to be much less of an issue for
-8-
171 iOS devices, Stereonet Mobile nonetheless uses oversampling to avoid any such prob-
172 lems. Before starting sampling, however, the device must be stable. Stereonet Mobile de-
173 termines device stability using the acceleration and rotation rate data provided by the
174 device. Absolute stability is not necessarily desirable as Stereonet Mobile permits the de-
175 termination of orientation by sighting and, whenever the phone is not held against the
176 rock, small motions are inevitable. Thus, stability in Stereonet Mobile is defined as user
177 acceleration rates of < 0.04 m/s2 and rotation rates of < 0.09 radians/s, values that were
178 picked by trial and error. Stability constraints help to avoid inadvertent recording of da-
179 ta while the device is moving.
180 Once the user holds the device stably for 1 s, Stereonet Mobile determines the ori-
181 entation every 100 ms and displays the mean and standard deviation of all measure-
182 ments for as long as stability is maintained. For example, if the user holds the phone on
183 a bedding surface for 5 seconds, the orientation and error displayed (Fig. 3) and record-
184 ed will reflect the average of 40 measurements ((5 s - 1 s wait time)×10 samples/s). If the
185 error in strike or dip exceeds 3° or the device is moved above the stability threshold, the
186 values are deleted and averaging begins anew. The same standards are used for linea-
187 tion measurements and for measurements of planes by sighting with the device camera.
188 Note that this sampling procedure is useful for eliminating random errors in-
189 cluding sensor transients, but it does not eliminate systematic errors such as those that
190 arise from the nearby environment. If the magnetic field is continuously perturbed by
191 the presence of a nearby metal object, data redundancy will not fix the problem. The
192 standard way to attempt to reduce such problems is by magnetometer calibration. For
193 iOS devices, this is achieved by waving the phone in a figure-8 pattern or tilting and ro-
194 tating the phone. In iOS 10 and with recent Apple® devices, one almost never sees the
-9-
195 calibration screen, reflecting the increasing sophistication of the CoreMotion routines
196 and services provided to the programmer. Nonetheless, in our experience, moving the
197 phone in a figure-8 before starting measurements at a new outcrop or after making sev-
198 eral measurements still seems to give better results than just assuming the operating
199 system is giving the best possible orientations.
200 3. Comparison to Analog Compass Readings 201 In this section, we compare both Stereonet Mobile and Fieldmove Clino to meas-
202 urements made by traditional analog compasses. An iPhone 6s and an iPhone 7, both
203 running iOS 10.3 were used for the digital measurements. Two traditional Brunton
204 compasses, each about three decades old, were used to measure the strike of a plane
205 and then the dip as separate measurements. A newer Brunton Geo compass was also
206 used to measure dip azimuth and dip of planes in a single measurement, similar to how
207 a geologist using a Freiberg compass would do. The traditional Brunton provides a
208 more precise measure of dip than the Geo, simply because of the larger scale of the cli-
209 nometer. Note that, for analog compasses, strike and dip are commonly independent
210 measurements whereas the phone measures the pole to the plane in a single measure-
211 ment.
212 3.1. Controlled Test 213 The first set of observations were collected in a highly controlled environment. A
214 thin, heavy, flagstone of Archean Elba Quartzite from the Raft River Mountains of
215 northwestern Utah measuring 96 by 53.5 cm was propped up at different angles in an
216 outdoor setting away from environmental noise (e.g., power lines, metallic structures).
217 Most of the surface roughness of the Elba is at a scale smaller than the area of the back
218 of the compass or phone. One of the authors removed all metal from their person (keys,
-10-
219 pocket knives, metallic wristwatches, writing instruments, etc.). The quartzite slab was
220 tilted at 72°, 45° and 30° and, for each dip orientation, we made 30 measurements each
221 with Stereonet Mobile, with Fieldmove Clino, and with two types of Brunton compasses at
222 different places on the slab (Fig. 4). Because there was no significant difference between
223 the two types of analog compasses, those results were combined. No attempt was made
224 to link individual measurements by different devices; instead, we wanted to calculate
225 the best average determination of slab orientation. Note that, the actual number of indi-
226 vidual measurements by Stereonet Mobile (and perhaps by Fieldmove Clino) is much larg-
227 er because each of the 30 measurements represents the average of many tens of meas-
228 urements at 100 ms intervals.
229 Figure 4 depicts the results of this controlled test on an equal area, lower hemi-
230 sphere projection. The α95 uncertainty cones for the mean vector of each individual
231 group of 30 measurements vary from 0.4 to 0.6°. The angular difference between the
232 mean vectors for each of the devices at each of the orientations is ≤ 2°. This difference is,
233 in some cases, large enough that the uncertainty cones do not overlap. However, in a
234 real world case, it is highly unlikely that a difference of <2° would make the slightest
235 difference in all but the most demanding applications, in which case one would be un-
236 likely to be using either an analog or digital compass!
237 The rose diagram (Fig. 4) shows the difference in strike between the different de-
238 vices for the 72°-dipping data set. Although the measurements are not identical, the
239 maximum of the strikes for the two iPhone apps and the analog compass measurements
240 overlap. In general, as the dips decreases from 90°, the difference in strike will common-
241 ly be larger than the angular difference between the plane. Thus, for anything but near-
-11-
242 vertical planes, the difference in strike is a poor indicator of similarity between planes;
243 the angular difference between poles should always be used instead.
244 3.2. Field Test at Bear Valley, Pennsylvania 245 The Bear Valley strip mine (Fig. 5) in the Anthracite District of the Pennsylvania
246 Valley and Ridge province is a classic locality (Nickelsen, 1979) visited by many genera-
247 tions of Northeastern U.S. geology students. The site, which provides spectacular three
248 dimensional exposures of tightly folded Carboniferous strata, is removed from power
249 lines and metallic structures that might perturb the local magnetic field. Two of the au-
250 thors, both experienced field geologists, visited the site and collected comparative data
251 using both Stereonet Mobile and traditional analog compasses. This represents a more
252 typical field situation as no special attempt was made to remove metal objects from
253 pockets, belts, etc. Because the site was stripped to a single stratigraphic horizon, in
254 most places in the pit, topographic contours lines — constructed from the Pennsylvania
255 state LiDAR survey with a resolution of 2 ft (0.62 m) — parallel strike of the bedding.
256 Thus, one can visually compare strikes measured at the site with local contours to assess
257 accuracy of the iPhones, except at site 1 where the nose of an anticline was excavated to
258 construct an access road, and locally where the device GPS mislocated the measure-
259 ments by ~15 m (Fig. 5). In the latter case, the measurements were made in a narrow
260 valley with limited sky view. The fit is quite good in most cases.
261 Two different types of studies were conducted. In the first, the same spot on the
262 rock was measured both with Stereonet Mobile and with a Brunton compass, a plane-by-
263 plane comparison. In the second, at three different sites, we measured ten strikes and
264 dips using both the phone and the compass and, in one case, also using the sighting ca-
-12-
265 pability of Stereonet Mobile. Both experiments sampled rocks on both limbs of the folds,
266 though not in the same place.
267 In the plane-by-plane comparison (Fig. 6), most measurements are close to each
268 other but certainly not exactly the same. The median mismatch is 2.7° and average is 3.2
269 ± 2.25° with the normal distribution significantly skewed towards smaller angles (Fig.
270 6b). Four out of twenty-four measurements have mismatch angles exceeding 5.5°. While
271 we presume that the analog compass measurements are more accurate, we have no in-
272 dependent means of verifying that assumption. Using the poles to bedding measured
273 by the iPhone, the best-fitting fold axis as determined by the phone differs by just 0.1°
274 from that determined by the analog compass. In other words, either data set would
275 have worked well to determine the fold axis in these folded strata. There is no con-
276 sistency to the difference between phone and compass measurements suggesting that
277 the differences are random.
278 In the second Bear Valley data set, average measurements of about 1 m2 of three
279 different bedding surfaces were determined (Fig. 7). In this case, while observations
280 from iPhone and compass are very similar, the mismatch angles of the mean vectors of
281 the poles (3.5 – 4.3°) are slightly larger than the α95 uncertainty cones (<2°). For the
282 most gently dipping of the three bedding sites measured, the rose diagram (Fig. 7)
283 shows that the strikes of the phone measurements appear consistently rotated clockwise
284 by 5-10°, though the actual angular different between the poles is less than 4°. If we cal-
285 culate the fold axis from bedding poles determined from the iPhone measurements, the
286 result differs by 4.4° with most of that difference in the plunge angle. Although we do
287 not know with great certainty the source of the mismatch between iPhone and compass
288 measurements, the geologist who collected this data set was wearing a metal wrist-
-13-
289 watch, which can produce visible perturbations in the device magnetometer readings
290 when held closer than 10 cm to the face of the phone (Fig. 1b).
291 4. Independent Assessment of Phone Accuracy 292 As the above discussion implies, we generally assume that the analog compass is
293 the canonical measurement and any deviation when compared with a phone measure-
294 ment at the same locality indicates a deficiency of the phone. However, for any single
295 phone-compass measurement pair, we have no independent means of verifying that the
296 analog compass is more accurate. In this section, we compare a large number of iPhone
297 measurements, not to compass measurements, but to the orientations of the structures
298 visible on Google Earth imagery. The almost complete lack of vegetation in the Atacama
299 Desert of northern Chile, combined with unique suites of surface cracks visible from
300 space, make this comparison possible. Along with Chilean colleagues, the senior author
301 and his students have been studying these surface cracks for the last 15 years (Baker et
302 al., 2013; González et al., 2008; Loveless et al., 2005, 2009) and have mapped more than
303 50,000 individual crack traces on Google Earth and IKONOS imagery.
304 In 2014, the Mw8.1 Pisagua earthquake, located just offshore of the northern
305 Chile (Fig. 8), produced a new suite of fresh surface cracks (Fig. 9) which, in virtually all
306 cases, reactivated the long-lived surface cracks visible on the satellite imagery. The first
307 and third authors documented the orientations of more than 3,700 new cracks in 14
308 days of field work soon after the earthquake (Loveless et al., 2016; Scott et al., 2016). All
309 of the original observations are available in the supplemental material associated with
310 Scott et al. (2016). These new cracks were measured using an iPhone 4s and an iPhone
311 5s running the Fieldmove Clino app (Stereonet Mobile did not exist at that time). Of course,
312 not all surface cracks were reactivated during the Pisagua earthquake so our measure-
-14-
313 ments capture only a subset of the cracks visible from space. Nonetheless, this data set
314 gives us a quantitative basis to compare phone measured orientations with orientations
315 in the imagery, independent of analog compass measurements. Though our iPhone
316 measurements were checked periodically with analog compass measurements, it is un-
317 likely that we would have been able to make 1/10th as many measurements without
318 using the phones as primary data collection devices.
319 The surface cracks have the further advantage that, because they are vertical, the
320 determination of the strike depends completely on the magnetometer and thus repre-
321 sents the best test of the most sensitive sensor of the phone. On the other hand, the
322 cracks are not perfectly planar but have significant irregularities at various scales along
323 strike (Fig. 9) such that measurement in the field necessarily involves some visual aver-
324 aging, thus increasing the uncertainty of the measurement. Because of uncertainty in
325 location, commonly on the order of 5-10 m, exists for any GPS receiver, we cannot al-
326 ways relate the orientation of every measured crack on the ground to those visible in
327 satellite imagery. Finally, one should not expect perfect agreement between cracks
328 measured on satellite imagery and those fresh coseismic cracks measured on the
329 ground: the latter constitute only a small subset of the former and at any particular site,
330 only those cracks which were suitably oriented for reactivation opened coseismically.
331 We have chosen four sites out of the 72 visited to show here (Figs. 10-13). One
332 can see the old cracks, in which the fresh coseismic cracks occur (Fig. 9a), clearly at the-
333 se four sites and they demonstrate well the relationship between orientations measured
334 in the field using iPhones and the orientations of cracks in the imagery. The results we
335 show are typical of the data set as a whole. For each site, both a satellite image with
336 strike indicators for the cracks measured in the field and a rose diagram comparing the
-15-
337 old cracks on the imagery to the new cracks measured with phones. All of the satellite
338 images have had some basic image processing applied to them, including contrast and
339 tonal enhancement, sharpening, and denoising, to render them easier to see in the pub-
340 lication. The statistics for each of the sites and data sets are summarized in Table 1.
341 For all of the sites, the circular mean of the coseismic crack orientation is within
342 about 10° of the mean of the long-term crack orientation and in all cases, the uncertainty
343 ranges overlap at the 2σ level (Table 1). Furthermore, the azimuthal bins with the max-
344 imum number of strikes are identical in all but one case. This is a surprisingly good cor-
345 relation considering that the average reflects all of the orientations of the long-term
346 cracks whereas the Pisagua earthquake cracks are represent just a subset of the orienta-
347 tions at each site. This is perhaps clearest at the Caleta Buena site (Fig. 11) where long-
348 term cracks with azimuths between 045 and 090° are common but were not very suita-
349 bly orientated for reactivation during the Pisagua earthquake. Likewise, the Pampa de
350 Tana site (Fig. 13) has numerous NNE to NS striking long-term cracks that were not re-
351 activated.
352 The Punta de Lobos fan (Fig. 10) and the Caleta Junin (Fig. 12) sites have the
353 most impressive correlation between long-term crack orientation as documented with
354 satellite imagery and coseismic crack orientation measured with iPhones. Both visual
355 inspection of the images, the rose diagrams, and the statistics all show that the iPhones
356 captured the orientations very well. Thanks to a previous study at Punta de Lobos
357 (Baker et al., 2013), we also have on-the-ground, long-term crack measurements made
358 with a Brunton compass (Fig. 10). The Brunton measurements are statistically indistin-
359 guishable from the coseismic crack orientations measured with iPhones and Fieldmove
-16-
360 Clino. Overall, these four sites are representative of the coseismic crack data set as a
361 whole in terms of fidelity to local orientations.
362 5. Discussion
363 5.1. Lessons pertinent to using iPhones as data collection devices 364 In most cases, the angular mismatch between measurements made with several
365 different iPhones and those made with traditional analog compasses or measurements
366 on satellite imagery differ by less than 10°. In the case of the satellite imagery, the mis-
367 match is probably much lower than shown because of the inclusion of unreactivated,
368 long-term cracks in the mean orientation calculation. This is reflected in the large and
369 overlapping uncertainty intervals in the data sets. In places where one can with confi-
370 dence visually relate field observations with specific cracks on the ground (e.g., Caleta
371 Junin, Fig. 12), the correlation is surprisingly good. Overall, our 3,700 iPhone measure-
372 ments in northern Chile provide robust proof that iPhones can be used in the field with
373 good results.
374 The data from our controlled test using the Elba Quartzite flagstone (Fig. 4) at
375 various dips demonstrates that iPhone measurements can be very reliable when appro-
376 priate precautions — removal of metal objects from the person making the measure-
377 ments — are taken. When multiple measurements from the phone are averaged, they
378 are just as good and nearly indistinguishable from a similar number of averaged meas-
379 urements from traditional compasses. Both phone and compass have uncertainties and
380 an average of analog measurements should be compared to an average of digital meas-
381 urements.
382 There is no significant difference between the two iPhone programs tested, Stere-
383 onet Mobile and Fieldmove Clino, probably because both use the same iOS CoreMotion
-17-
384 routines to calculate the orientation of the phone and thus the structure. Fieldmove Clino
385 deployed on Android devices (Novakova and Pavlis, 2017) appears to be hostage to the
386 lower quality sensors used in those Android devices; the iOS version performed ex-
387 tremely well in our northern Chile coseismic crack study. It is incumbent upon devel-
388 opers to explain how their software works to scientists who rely on their apps to collect
389 priceless data. It would be useful to know, for example, what averaging schemes, if any,
390 Fieldmove Clino uses and how it calculates orientation as we have done here with Stereo-
391 net Mobile.
392 The field test at Bear Valley (Figs. 5-7) likewise demonstrates that iPhones can be
393 useful in a traditional field setting and, though perhaps not quite as accurate as analog
394 compasses, can in aggregate provide useful information in a fraction of the time re-
395 quired to use the compass. The cylindrical fold axis determined by both phone and
396 compass measurements from the same outcrops differs by less than 5°. The strikes also
397 compare very well to contours of the LiDAR topography of the area (Fig. 5). The Bear
398 Valley data set also highlights the Achilles heel of all smart phone measurements: the
399 sensitivity of the device to local magnetic fields. In one case (Fig. 7), there appears to be
400 a systematic difference in strike of up to 10° which may have been due to the user wear-
401 ing a metal wristwatch which would have been reasonably close to the phone during
402 measurement. However, even in that case, the actual angular difference between phone
403 and analog compass measurements is less than 4°.
404 5.2. Best practices for smart phone data collection 405 The four different iPhones tested here appear to be demonstrably superior for
406 data collection to the two Android devices tested by Novakova and Pavlis (2017). This
407 suggests that anyone contemplating data collection with a smart phone should consider
-18-
408 their device purchases very carefully, prioritizing quality and reliability over economy.
409 However, on both platforms, phone components and operating systems change all the
410 time and no one can guarantee absolutely that the most reliable phone today will be so
411 two years from now. Unfortunately, most phone makers have little incentive to make
412 phones that are ideal for the structural geologist’s purpose. Nonetheless, someone mak-
413 ing a smartphone purchase today with the intention of collecting data appears to have a
414 better chance of success with an Apple® device.
415 Regardless of the device purchased, one should always do careful tests similar to
416 those described here to determine the reliability of their individual instrument before
417 heading out to the field. General reputation of a manufacturer does not guarantee that
418 the individual device will be adequate to the task. In the event that a particular device
419 proves faulty, many apps, including those described here, can still be used as data re-
420 corders, providing the user with automatic time, date, and location tagging of all obser-
421 vations.
422 When using a smart phone in the field, special care beyond that normally used
423 with analog compasses should be taken around metal objects and magnets. There are
424 many phone cases available with magnetic clips or closures that can thoroughly spoil
425 your phone readings. Additionally, the user should be careful to remove from proximi-
426 ty seemingly innocent things like metal wristwatches, pens, hand lenses, pocket knives,
427 etc.
428 Even though the magnetometer calibration screen seldom appears anymore in
429 iOS 10, in our experience, it is a best practice to perform similar motions (figure-8s, tilt-
430 ing the phone in all orientations) before starting on any new outcrop. Additionally, sim-
431 ilar calibrations should be undertaken periodically on a single outcrop or whenever a
-19-
432 reading does not appear to make sense, or where digital and analog measurements dif-
433 fer significantly. Using a program that can display an orientation on a high-resolution
434 satellite image so the user can verify that a measurement agrees with local geologic
435 strike of the feature being measured can provided added confidence. Fieldmove Clino
436 can do this but Stereonet Mobile, at the present time, cannot.
437 It goes without saying that any phone being used for data collection should be
438 protected from dust, water, and other abuse in a ruggedized, non-magnetic case. Our
439 experience in northern Chile suggests that, in some ways, data collection with a phone
440 may actually be more secure than in a notebook: whenever the user has a cell phone or
441 wifi signal, s/he can simply email the current data file to themselves. This facilitates
442 back up of critical field data at more frequent intervals than one would do if they had to
443 wait until returning to town to find a photocopy shop to copy one’s field notes!
444 Our Chile experience also suggest that the user should have, as part of the stand-
445 ard field gear, large capacity, rechargeable lithium ion batteries. Small, portable batter-
446 ies with capacities exceeding 20,000 mAh cost less than $50 and can recharge a smart
447 phone completely 5-7 times. When we were using our phones to make 300 or more
448 measurements/day, the battery would become completely depleted at or even before
449 the end of a ten hour field day. With 4 or 5 days between return trips into town, having
450 such batteries available in one’s camp and backpack is essential.
451 6. Conclusions 452 With some modest precautions, Apple® iPhones can be successfully used by
453 structural geologists as data collection devices in the field. We have no direct experience
454 with Android devices, but the work of Novakova and Pavlis (2017) is less encouraging
455 for those devices. Both studies, however, tested an infinitesimal number relative to the
-20-
456 total number of devices that have been produced of each type and, furthermore, operat-
457 ing systems and device components change frequently. As of today, the structural geol-
458 ogist will still want to take their analog compass to the field with them and verify their
459 phone observations. Anyone contemplating data collection with any smart phone needs
460 to carry out extensive tests of the kind performed here to verify that the data collected
461 are reliable.
462 When testing mobile devices, four best practices should apply: First, because
463 both phone and analog compass measurements have uncertainty and natural surfaces
464 are inherently irregular, one should make multiple measurement of the same surface
465 using each type of instrument and compare the averages of the measurements. Compar-
466 ing one-off measurements, as has commonly been done in the past, fails to acknowledge
467 that uncertainty exists in all measurements with any type of instrument. Second, when
468 evaluating planar orientations, one should always compare the angular difference be-
469 tween poles to the planes and not the difference in strike or dip. At low to moderate
470 dips, the strike can differ significantly between two measurements even though the ac-
471 tual angular difference between the planes as determined by the poles is small. Third,
472 the user should invest in a program that monitors the device sensors over time, note
473 transients, and experiment with the effect of proximity of external metallic objects on
474 the phone magnetometer. Several such apps are available for free or at modest cost in
475 the app stores for both Android and iOS devices. Finally, where possible the user
476 should compare their phone measurements, not only to analog compass measurements,
477 but also to data independent of the magnetic field such as the LiDAR topographic con-
478 tours (Fig. 5) and the Google Earth images (Figs. 10-13) used in this study.
-21-
479 Apps on mobile devices will inevitably replace traditional analog compasses.
480 Smart phones have already replaced compact cameras, video cameras, GPS receivers,
481 music players, digital voice recorders, exercise monitors, and reference libraries, to
482 name a few. In each of these cases, the question was not whether the phone was superi-
483 or to the dedicated device that it replaced but whether it was good enough for the pur-
484 pose at hand. The results presented here suggest that iOS devices have achieved that
485 status for all but the most demanding applications. In most use cases, the accuracy of
486 the iPhone is more than sufficient given the variability of the natural feature being
487 measured. Additionally, because the rapidity and ease of use, iOS devices have a dis-
488 tinct advantage over analog compasses: multiple measurements can be made rapidly
489 and uncertainty instantly assessed. Making a single measurement with an analog com-
490 pass and paper field notebook is so time consuming that geologists seldom make more
491 than one measurement at a site and thus have no way to assess uncertainty. Structural
492 geology is entering the age of data redundancy.
493 7. Acknowledgments 494 The senior author thanks Paul Karabinos, Néstor Cardozo, and Haakon Fossen
495 for beta testing Stereonet Mobile. Comments from Haakon and Néstor have helped to
496 improve the manuscript. We are likewise grateful to ______for reviews of this
497 manuscript. The collection of the northern Chile field data was made possible by a RAP-
498 ID grant from the U.S. National Science Foundation, EAR-1443410.
-22-
499 8. References Cited 500 Allmendinger, R.W., Cardozo, N., and Fisher, D., 2012. Structural Geology Algorithms: 501 Vectors & Tensors, ed. Cambridge University Press, Cambridge, England, 289 p. 502 Baker, A., Allmendinger, R.W., Owen, L.A., and Rech, J.A., 2013. Permanent defor- 503 mation caused by subduction earthquakes in northern Chile. Nature Geoscience 504 6, 492-496. 505 Cawood, A.J., Bond, C.E., Howell, J.A., Butler, R.W.H., and Totake, Y., 2017. LiDAR, 506 UAV or compass-clinometer? Accuracy, coverage and the effects on structural 507 models. Journal of Structural Geology 98, 67-82. 508 González, G., Gerbault, M., Martinod, J., Cembrano, J., Allmendinger, R., Carrizo, D., 509 and Espina, J., 2008. Crack formation on top of propagating reverse faults of the 510 Chuculay Fault System northern Chile: insights from field data and numerical 511 modeling. Journal of Structural Geology 30, 791-808. 512 Hama, L., Ruddle, R.A., and Paton, D., 2014. Geological Orientation Measurements us- 513 ing an iPad: Method Comparison. TPCG, 45-50. 514 Loveless, J.P., Allmendinger, R.W., Pritchard, M.E., Garroway, J.L., and González, G., 515 2009. Surface cracks record long-term seismic segmentation of the Andean mar- 516 gin. Geology 37, 23-26. 517 Loveless, J.P., Hoke, G.D., Allmendinger, R.W., González, G., Isacks, B.L., and Carrizo, 518 D.A., 2005. Pervasive cracking of the northern Chilean Coastal Cordillera: New 519 evidence of forearc extension. Geology 33, 973-976. 520 Loveless, J.P., Scott, C.P., Allmendinger, R.W., and González, G., 2016. Slip distribution
521 of the 2014 MW=8.1 Pisagua, northern Chile, earthquake estimated from coseismic 522 fore-arc surface cracks. Geophysical Research Letters 43, 10134-10141. 523 Midland Valley, 2017. Fieldmove clino: Android users guide, ed. Midland Valley, Glas- 524 gow, Scotland, 38 p. 525 Nickelsen, R.P., 1979. Sequence of structural stages of the Alleghany Orogeny, at the 526 Bear Valley strip mine, Shamokin, Pennsylvania. American Journal of Science 527 279, 225-271. 528 Novakova, L., and Pavlis, T.L., 2017. Assessment of the precision of smart phones and 529 tablets for measurement of planar orientations: A case study. Journal of Structur- 530 al Geology 97, 93-103. 531 Ramsay, J.G., and Huber, M.I., 1983. The techniques of modern structural geology, vol- 532 ume 1: Strain analysis, ed. Academic Press, London, 307 p. 533 Scott, C.P., Allmendinger, R.W., González, G., and Loveless, J.P., 2016. Coseismic exten- 534 sion from surface cracks reopened by the 2014 Pisagua, northern Chile, earth- 535 quake sequence. Geology 44, 387-390.
-23-
536 9. Figure Captions 537 Figure 1. Graphs showing sensitivity of (a) orientation, (b) magnetometer, and (c)
538 accelerometer sensors on one of the iPhones used in this study. The sensors were sam-
539 pled at ~30 Hz using the Sensor Kinematics Pro app available in the iOS app store. Com-
540 pare this figure to Figure 2 of Novakova and Pavlis (2017). In each, green corresponds
541 to the X1 axis, blue to the X2 axis, and red to the X3 axis (see Figure 2 for the coordinate
542 system). The inset diagrams magnify the graph to show the variability of the sensor da-
543 ta. In (b), the device magnetometer was perturbed by passing a metal object ~5 cm from
544 the device at the times indicated.
545 Figure 2. The iOS device coordinate system (X′1, X′2, X′3) and its relationship to a
546 typical structural geology North-East-Down (X1, X2, X3) coordinate system. Device ori-
ˆ 547 entation is determined by transforming the unit vectors p and ℓˆ into the unprimed
548 geographic coordinate system. Selected direction cosines of the transformation matrix
549 are shown. Many programs use the Euler angles pitch, roll, and yaw to determine de-
550 vice orientation but iOS also provides the programmer with access to the rotation ma-
551 trix and quaternions. Stereonet Mobile uses the former.
552 Figure 3. Screen captures of Stereonet Mobile in data capture mode. (a) Data cap-
553 ture by placing the device directly on the rock surface to be measured. The app is capa-
554 ble of capturing both the plane and a line in the plane in a single measurement. (b) Data
555 capture by sighting parallel to the strike direction using the device camera. The green
556 circle in the lower right shows that the direction of sight (pole to the phone) is within 2°
557 of horizontal. In both cases, once stable the phone measures the orientation of the de-
558 vice every 100 ms and reports the uncertainty to the user.
-24-
559 Figure 4. Controlled experiment using a flagstone of Elba Quartzite propped up
560 at different angles. Thirty measurement each were made using the two smart phone
561 apps, Stereonet Mobile and Fieldmove Clino, and two different analog compasses, a tradi-
562 tional Brunton and a Brunton Geo. The difference in mean vector (large circles) of the
563 poles to the plane (small dots) for the three different dips are shown graphically and
564 numerically. Petals in all rose diagrams are filled with alpha channel transparency so
565 that the reader can see where more than one petal overlaps. Additional apparent colors
566 are due to overlapping primary colors and partial transparency.
567 Figure 5. Topographic map of part of the abandoned open pit at Bear Valley,
568 Pennsylvania. Topographic contours constructed from the Pennsylvanian state LiDAR
569 data set, originally at 2 ft. (0.62 m) intervals. For clarity, only 10 ft. (3.05 m) contours are
570 shown. Because the pit was stripped to a single stratigraphic horizon, contour lines
571 should approximate strike of bedding in many parts of the pit, though not at Site 1. The
572 red strikes and dips were measured with an iPhone 7 and are plotted on the stereonet in
573 Figure 6; the blue strikes and dips were measured with an iPhone 6s and plotted in Fig-
574 ure 7. Locations generally have uncertainties of ±5 to ±10 m. All sites also had analog
575 compass measurements but those data are not plotted here.
576 Figure 6. Measurement by measurement comparison of Stereonet Mobile and
577 Brunton compass measurements at Bear Valley, Pennsylvania. (a) Both poles (dots) and
578 planes (great circles) are shown. Difference between the calculated fold axis from the
579 two different data sets is 0.1°. (b) Histogram of the angular mismatch of poles to planes
580 between phone and compass for each of the 24 measurements in (a).
581 Figure 7. Ten phone and ten Brunton measurements of three different irregular
582 bedding surfaces, showing planes (great circles) and poles (dots) as well as mean vec-
-25-
583 tors and α95 uncertainty cones for each population. The delta angles are the difference
584 of the mean vectors of each population at each site. For the lowest dipping strata, ten
585 measurements of bedding using Stereonet Mobile’s sighting routine were also made
586 (blue great circles). The rose diagram of strike directions shown for the lowest dipping
587 site indicate that the phone strikes are rotated clockwise from the Brunton Geo strikes
588 by about 5-10°, although the angular difference of the poles are just 3.7°.
589 Figure 8. Map of Pisagua earthquake and location of the four sites depicted in
590 Figures 10-13 depicted with large yellow stars. Locations of the main shock as well as a
591 significant foreshock and aftershock are shown. The colored “butterflies” show the av-
592 erage and 2σ uncertainty of crack orientations at each of the 72 sites studied. See Scott et
593 al. (2016) and Loveless et al. (2016) for more information.
594 Figure 9. Typical coseismic cracks produced during the Pisagua earthquake. (a)
595 A long-term crack filled with aeolian sand (marked by white arrows) showing fresh re-
596 activated cracks in the middle. (b) Detail of a fresh coseismic crack cutting across faintly
597 visible tire tracks.
598 Figure 10. Punta de Lobos fan surface (Baker et al., 2013) shown in a Google
599 Earth image and the locations and orientations of coseismic cracks shown as short yel-
600 low line segments measured in three transects. Circular histogram (rose diagram)
601 shows the orientations of all long-term cracks measured along the transect (red petals),
602 long-term cracks measured with Brunton compass (green petals) and the orientations of
603 coseismic cracks measured with iPhones (blue petals). There is no significant difference
604 between iPhone and Brunton measurements. The mean vector for the imagery-
605 measured cracks captures a small population of NW-striking long-term cracks that were
606 not reactivated during the earthquake. In all rose diagrams, the numbers along the bot-
-26-
607 tom edge indicate the percent of the total data set. Blue petals are phone measurements
608 and red represent cracks digitized on the imagery. In figure 10, only, green petals are
609 Brunton compass measurements.
610 Figure 11. Caleta Buena site. The Pisagua earthquake preferentially reactivated
611 the NW-striking cracks at this area, with only very minor reactivation of the prevalent
612 NE-striking long-term cracks.
613 Figure 12. Caleta Junin site, located close to the epicenter of the Pisagua earth-
614 quake. Correlation of long-term cracks mapped on Google Earth and coseismic cracks
615 measured with iPhones is particularly good at this site as the longer term cracks are rel-
616 atively unimodal and were well-oriented for reactivation during the earthquake.
617 Figure 13. Pampa de Tana site showing preferential reactivation of the WNW-
618 striking long-term cracks and relatively infrequent reactivation of the prominent,
619 though secondary, ~NS oriented cracks.
620
-27-
621 Table 1. Punta de Lo- Caleta Buena Caleta Junin Pampa de bos Tana
long-term image 354.1 ± 7.1° 313.6 ± 25.3° 347.8 ± 19.0° 297.0 ± 9.2° mean vector & long-term Brunton 004.3 ± 4.9 N/A N/A N/A 2σ uncertainty coseismic 006.9 ± 9.2° 323.1 ± 16.3 352.6 ± 12.5 286.3 ± 12.3°
long-term 001 - 010° 331 - 340° 351 - 360° 281 - 290° Bin azimuth long-term Brunton 001 - 010° N/A N/A N/A w/ max count coseismic 001 - 010° 321 - 330° 351 - 360° 281 - 290° 622
-28- All Figures −1 θ31 = cos (r31) +X′3 θ = cos−1 r +X′2 32 ( 32 ) yaw roll E (X2) pˆ +X′ N (X ) 1 1 ˆ ℓ pitch
D (X3) (a) Stereonet view (b) Sighting view N = 30
SM Δ FMC = 1.6°" Brunton Δ FMC = 0.9°" SM Δ Brunton = 1.1° SM Δ FMC = 0.5°" Brunton Δ FMC = 1.6°" SM Δ Brunton = 1.1° SM Δ FMC = 0.7°" Brunton Δ FMC = 2.0°" SM Δ Brunton = 1.4°
Stereonet FieldMove Brunton Mobile (SM) Clino (FMC) compass 1100
1060
1040 1100 -76.598 ° -76.597 ° -76.596 ° -76.595 ° 40.765 ° 1060
100 0 100 200 300 400 980500 2 ft contour interval feet 100 0 100 meters 940
900 1040
39
34 960 42 28 29 31 20.1 900 1060 44 28 980 930 40.764 ° 970 960 920 13 21 8 31 72 1050 46 28 38 36 70 73 35 19 21 36 920 960 73 73 940 980 57 65.8 36 observations 40 980 38 72 mislocated by 39 1000 1020 64 device GPS 73 73 980 61 61.8 1060 1040
1100
1140 1060 1080
40.763 ° -76.598 ° -76.597 ° 1200 1160 1170
1200 (a) fold axis nearly identical for phone and compass measurements 1
3
2
SM Brunton
(b)
6
5 Variable: Mismatch Angle (°) 4 Mean: 3.22565 Median: 2.67838 3 Std. Dev.: 2.25027 Std. Err.: 0.459335 Frequency 2 Skewness: 1.20171 Minimum: 0.35267
1 Maximum: 9.31927 N: 24
0 -5 0 5 10 15 Range of Mismatch Angle (°) Δ = 4.3°
Δ = 3.7°
Δ = 3.5°
SM SM-sighting Brunton Geo -71.5 ° -71.0 ° -70.5 ° -70.0 ° -69.5 ° -69.0 ° -68.5 ° -17.5 ° -17.5 °
-18.0 ° -18.0 °
–71° –70° –69° -18.5 ° -18.5 ° Bolivia
Location
-19.0 ° –19° -19.0 ° Chile
Fig. 13 -19.5 ° -19.5 °
Mw8.1 (2014-04-01) Fig. 12 Fig. 11
-20.0 ° –20° -20.0 ° Mw6.7 (2014.03.16)
-20.5 ° -20.5 ° Mw7.7 (2014.04.03)
-21.0 ° –21° Fig. 10 -21.0 °
-21.5 ° -21.5 ° -71.5 ° -71.0 ° -70.5 ° -70.0 ° -69.5 ° -69.0 ° -68.5 ° (a)
(b) 200 m N = 71! N = 100
5 10 15 20
100 m N = 99! N = 78
5 10 15 20 25 30
200 m N = 47! N = 118
5 10 15 20 25 125 m