High Resolution Topography of the Northern 16 km of the M6.9 1983 Borah Peak Earthquake Surface Rupture on the Lost River Fault Zone, Idaho, USA

Michael P. Bunds1 (michael.bunds@uvu.edu), Christopher B. DuRoss2, Ryan D. Gold2, Nadine G. Reitman3, Nathan A. Toké1, Richard W. Briggs2, Stephen F. Personius4, Kendra Johnson5, Lia Lajoie5,

Brittany Ungerman1, Ephram Matheson1, Jeremy Andreini1, and Kenneth Larsen1

1 Dept. of Earth Science, Utah Valley University, 800 West University Parkway, Orem, UT 840582 U.S. Geological Survey, 1711 Illinois Street, Golden, CO 80401 3 University of Colorado Boulder, UCB 399, Boulder, CO 80309-03994 U.S. Geological Survey – Emeritus, 1711 Illinois Street, Golden, CO 804015 Colorado School of Mines, 1318 Maple Street, Bldg 6, Golden, CO 80401

IntroductionThis document accompanies a high-resolution topographic data set and orthomosaics of part of the Lost River fault zone (LFRZ), Idaho, USA. The topographic data set covers the northern ~16 km of the surface-rupture that occurred on the LRFZ in the Mw 6.9 1983 Borah Peak Earthquake (Figure 1). Point clouds and digital surface models (DSMs), were generated from low-altitude aerial photographs using Structure-from-Motion and multi-view stereo processing (SfM) (Table 1). The point clouds, DSMs, orthomosaics, and supporting tables of photograph positions and georeferencing information are available free-to-public on OpenTopography. An overview of the dataset and methods used to create it is provided here.

For additional information, please contact Michael Bunds, Utah Valley University (michael.bunds@uvu.edu).

BackgroundThe LRFZ is a major, range-bounding, west-dipping normal fault in the northern Basin and Range province (Figure 1). The Mw 6.9 1983 Borah Peak earthquake occurred on it and created surface rupture along the southern portion of the Warm Spring section and the entire Thousand Springs section of the fault (Crone et al., 1987; DuRoss et al., in press). The Warm Spring and Thousand Springs sections are separated by the Willow Creek Hills, which form a structural boundary. We used unoccupied aerial systems (UAS) and a tethered ‘helikite’ balloon to acquire aerial photographs, which were combined with ground control georeferencing to create two point clouds that model the topography at high resolution (< 10 cm) along the northern ~16 km of surface rupture. One point cloud covers the southern ~9 km of the Warm Spring Section and a second point cloud covers the northern ~7 km of the Thousand Springs Section. DSMs were generated from the point clouds, and orthomosaics of each area were also made from the aerial photographs. The primary motivation for acquisition of the data set was study of the surface offset across the Willow Creek Hills structural boundary from the 1983 earthquake and prehistoric surface-rupturing events (DuRoss et al., in press).

Data Description (Table 1)1. Two separate point cloud models of the imaged terrain, one of the Warm Spring section (WS)

and one of the northern part of the Thousand Springs section (TS), including the Willow Creek Hills strand (also called the Arentson Gulch Fault, e.g., DuRoss et al., in press). The WS point cloud contains 3.909 x 109 points and the TS point cloud contains 2.603 x 109 points, each with a

page 1 of 6

color attribute. Points are unclassified and include both ground and vegetation. The WS point cloud covers approximately 4.46 km2, for an average point density of 876 points/m2 and the TS point cloud covers 6.86 km2 at an average point density of 379 points/m2. The reference frames for both point clouds are NAD83(2011) UTM zone 12 (EPSG 6341), epoch 2010.000, and NAVD88 (i.e., orthometric heights determined using GEOID 12B, units of meters). Note that the WS point cloud is predominantly located within UTM Zone 12, but extends into Zone 11 at its northwestern end (the TS point cloud lies wholly within Zone 12).

2. DSMs, with a 0.10-m pixel size. DSMs are derived from the point clouds and cover the same areas. Rasterization was performed in Agisoft Metashape (v1.5.1), using a ‘binning’ algorithm, wherein each DSM pixel value (elevation) is calculated as an average of the elevations of points located within the pixel area (the WS point cloud density averages ~8.8 points/pixel, the TS averages ~3.8 points/pixel). The DSM reference frames are NAD83(2011) UTM zone 12 (EPSG 6341), epoch 2010.000, and NAVD88 (i.e., orthometric heights determined using GEOID 12B, units of meters).

3. Orthomosaics, with a 0.05-m pixel size. Orthomosaics were constructed from the aerial imagery in Agisoft Metashape, using the DSMs to project the imagery. The reference frame is NAD83(2011) UTM zone 12 (EPSG 6341), epoch 2010.000.

4. Table of camera locations. The WS and TS point clouds were constructed from low-altitude aerial photographs taken from UAS and a tethered balloon. A table of the locations from which aerial photographs were taken is provided as a comma-delimited file. The reference frames are NAD83(2011) UTM zone 12 (EPSG 6341), epoch 2010.000, and NAVD88 (i.e., orthometric heights determined using GEOID 12B, units of meters).

5. Table of ground control point locations. The WS and TS point clouds were georeferenced using ground control points (GCPs), which are markers placed on the ground (visible in aerial photographs) and the positions of which were surveyed with differential GNSS (see georeferencing section below). A table of the locations of the GCPs is provided as a comma-delimited digital file. The reference frame is NAD83(2011) UTM zone 12 (EPSG 6341), epoch 2010.000, and NAVD88 (i.e., orthometric heights determined using GEOID 12B, units of meters).

Data Collection OverviewField work to acquire the high-resolution topography was conducted in May 2015 (Warm Spring section), and May 2016 (Thousand Springs section). The WS survey area was divided into 13 subsections (‘polygons’), and the TS section into 7 polygons. Each polygon contains at least nine ground control points and each was surveyed so as to ensure overlap in aerial photography and imaging of GCPs across polygons.

The WS model was made from 12,586 low altitude aerial photographs that were collected in May 2015 using UAS and a tethered helikite. UAS used for the WS data were DJI Phantom 2 v2 quadcopters, equipped with Sony A5100 cameras (16 mm lens) using a custom, fixed mount (9877 photographs). Data for the WS section were also collected with a Canon PowerShot SX230 that was attached with a fixed mount to a helikite (2709 photographs). For the TS model, 6053 low altitude aerial photographs were collected in May 2016 using UAS. UAS used were DJI Phantom 2 v2 quadcopters, equipped with Sony A5100 cameras (16 mm lens) using a custom, fixed mount (4072 photographs), a Falcon fixed wing also equipped with a Sony A5100 (20 mm lens) and a fixed mount (948 photographs), and 3DR quadcopter with a gimbal-mounted GoPro Hero 3 Black camera (1033 photographs). With the exception of the 3DR,

page 2 of 6

gimbals were not used, so most photographs are not from a near perfect nadir orientation, and vary from nadir by up to ~48o. Average ground sample distance for the WS section is ~2.15 cm, and ~3.07 cm for the TS data. Photograph overlap as calculated by Agisoft is > 9 throughout nearly the entire survey area. An additional measure of overlap is the number of photographs in which GCPs appear, which averages 39.8 for the WS section and > 20 for the TS section. The DJI Phantom and helikite platforms were piloted and navigated manually, whereas the Falcon and 3DR flew automated, calculated flight plans generated by experimental software. A table of positions for the photographs is included. The positions are estimates calculated during SfM processing.

ProcessingProcessing of the aerial photographs to produce point clouds was done primarily with Agisoft Photoscan; some final processing (e.g., DSM generation and export) was done with Agisoft Metashape. Each of the 20 polygons that compose the covered area was initially processed independently, including georeferencing, to produce a sparse cloud and camera models. The sparse clouds and camera models were georeferenced and optimized using locations of GCPs (described below). For the Willow Creek Hills strand of the TS area, points in the sparse cloud with relatively large uncertainty were iteratively removed and the model re-optimized. The polygons for each area (WS and TS) were then merged using both ground control and point cloud information (i.e., all polygons for WS area were merged together into a single model, and the same was done for the TS area), and dense point clouds were generated for the entirety of each of the surveyed areas (WS and TS). The sparse point clouds and camera models were made using the ‘high’ setting in Agisoft (i.e., photos were downsampled x2), and the dense cloud was built using the ‘high’ setting (i.e., photographs were downsampled x2). DSMs were made in Agisoft Metashape, which averaged elevations of points within each DSM pixel to calculate the pixel elevations. Point clouds and DSMs for the WS and TS sections were generated independently of each other.

Georeferencing and AccuracyAll data (point clouds, DSMs, orthomosaics, and accompanying GCP and camera locations), are in NAD83(2011) UTM Zone 12 (EPSG 6341) coordinates, with elevations given as NAVD88 orthometric heights in meters determined using GEOID 12B. All data were processed in epoch 2010.0000.

The point clouds were georeferenced using GCPs. GCPs consisted of markers placed on the ground, the positions of which were measured using differential GNSS (dGNSS). A total of 196 GCPs were used (141 for WS, 55 for TS), and a table of GCP positions is included (reference frame NAD83(2011) UTM Zone 12, GEOID 12B NAVD88 heights in meters, epoch 2010.0000).

dGNSS was done in kinetic mode using three different local reference stations and two rovers. Reference stations consisted of a Trimble 5700 receiver and Trimble Zephyr Geodetic I antenna. Reference station locations were determined using the National Geodetic Survey’s Online Positioning User Service (OPUS) and over 16 hours occupation time at each reference station site. OPUS results in epoch 2010.0000 were used, so effectively the entire data set is in that epoch. Reference station uncertainty is estimated to be approximately +/- 0.003 and 0.006 m horizontal and vertical respectively. The rovers used to occupy the GCPs (and checkpoints) consisted of a Trimble R8 and a Trimble 5700 receiver/Zephyr Geodetic antenna. Rover accuracy generally degrades towards the east, where proximity to the Lost River Range and taller and denser vegetation limited satellite visibility and decreased measurement precision. GCP location uncertainty is estimated to vary from +/- 0.01 (horizontal) and +/- 0.015 (vertical) up to approximately +/- 0.02 (horizontal) and +/- 0.04 (vertical).

page 3 of 6

The position of every GCP was manually marked by hand in every photograph in which it clearly appears using Agisoft Photoscan prior to optimization of the sparse point clouds and camera models, except for the relatively small areas flown by the 3DR quadcopter. In those areas photograph density is very high (up to 200 photos per GCP) and GCPs were placed in 20% to 50% of photos in which they appear.

Final point clouds, DSMs and orthomosaics were cropped prior to export from Agisoft Metashape. Cropping was done to include the area reasonably well covered by GCPs and photographs.

Vertical accuracy of the final DSMs were assessed using checkpoints. Checkpoints are dGNSS measurements taken on bare, relatively flat ground away from GCPs and vegetation. 57 checkpoints were measured in the WS area, 27 in the TS area. The elevation predicted for each checkpoint was extracted from the DSMs using ArcMAP, and a residual calculated as the difference between the DSM elevation and the directly measured elevation for each checkpoint. Root-mean-square error (RMSe; 1-sigma error) for all checkpoints relative to the WS DSM is 7.2 cm, and it is 5.9 cm for the TS DSM. GCPs and checkpoints were measured using the same reference stations as the GCPs, so some additional error could be present in the data set due to error in the relative and absolute positions of the reference stations, however this is likely small relative to the checkpoint RMSe. Note that misfit of the point clouds to GCPs (as measured by Agisoft) is much less than checkpoint error, and reflects the fact that GCP misfit as calculated in Agisoft is not a reliable measure of model accuracy across the survey area.

Although we have endeavored to minimize and quantify error in the data set, no guarantee of accuracy is given nor implied. In addition, DSM and point cloud error should be expected to vary across the data set. SfM processing using GCPs results in a model with good fit to GCPs, and potential warping between GCPs (in our case, probably mostly <5.9–7.2 cm based on checkpoints), as well as significant potential warping in areas not surrounded by GCPs (i.e., areas on the edge of the data set, outside of GCP coverage). Photograph coverage also may affect error, and portions of the model that extend beyond flight lines may contain larger than average error. For these reasons, the tables of GCP and camera locations are included.

page 4 of 6

Figure 1. Location map showing approximate areas covered by point clouds and DSMs (outlined in yellow), the Warm Spring and Thousand Springs sections of the Lost River Fault Zone, Willow Creek Hills, and extent of surface rupture in 1983 earthquake (from DuRoss et al., in press).

page 5 of 6

Table 1. Summary of Topographic Model Parameters

Parameter WS model TS model

Total points 3.909 x 109 2.603 x 109

Coverage area 4.46 km2 6.86 km2

Point density 876 points/m2 379 points/m2

Number photographs 12,586 6053

Average GSD 0.0215 m 0.0307 m

Number GCPs 141 55

Number checkpoints 57 27

Checkpoint RMSerror 0.072 m 0.059 m

DSM resolution 0.10 m 0.10 m

Orthomosaic resolution 0.05 m 0.05 mHorizontal reference frame –

point clouds, DSMs and orthomosaics

NAD83(2011) UTM Zone 12 (EPSG 6341) epoch 2010.0000

NAD83(2011) UTM Zone 12 (EPSG 6341), epoch 2010.0000

Vertical reference frame NAVD88 (GEOID12B) NAVD88 (GEOID12B)

Field data collection date May, 2015 May, 2016

References CitedCrone, A.J., Machette, M.N., Bonilla, M.G., Lienkaemper, J.J., Pierce, K.L., Scott, W.E., and Bucknam, R.C., 1987, Surface faulting accompanying the Borah Peak Earthquake and segmentation of the Lost River Fault, Central, Idaho, Bulletin of the Seismological Society of America, v.77, n. 3, pp 739-770.

DuRoss, C.B., Bunds, M.P., Gold, R.D., Briggs, R.W., Reitman, N.G., Personius, S.F., and Toke, N.A., in press, Complex normal-fault rupture behavior based on detailed surface-displacement measurements of the 1983 Mw 6.9 Borah Peak and prehistoric earthquakes along the northern Lost River fault zone (Idaho, USA), Geosphere.

AcknowledgementsThis project was supported by awards from the Scholarly Activities Committee in the College of Science, Utah Valley University, and the Department of Earth Science, Utah Valley University. We thank Bret Huffaker, Rick Lines, and Alexandria Valenzuela for assistance in the field.

page 6 of 6