Jeremy Saldivar’s scientific contributions

Historically, aseismic fault creep has been measured primarily at <40 m apertures with creep meters, at km’s distances with GPS methods, and with inSAR. In this project, we difference high resolution point clouds recently generated using structure from motion photogrammetry (SfM) against older LiDAR data, to ascertain the three dimensional deformation from creep across ~1km wide zones at 30 m resolution. The approach is applied at two study sites, Dry Lake Valley (DLV) on the central creeping section of the San Andreas Fault, and at Salt Creek (SC) on the Southern San Andreas Fault. Near DLV, creep meters have indicated the near-field creep rate is ~2.3 cm/yr, and nearby GPS stations record ~2.9 cm/yr of fault-parallel motion. In contrast, creep meter motion at SC is ~2 mm/yr, and nearby GPS stations show ~ 6.4 mm/yr displacement. At DLV, we differenced EarthScope Northern California LiDAR, collected in 2007 at 5.2 pts/m2, with an SfM-derived point cloud generated for this study. The SfM point cloud was made in fall 2017 using a Sensefly eBee Plus, a local GNSS base, both on-board photo geolocation and ground control points, and its point density is 422 pts/m2. For SC, work is currently underway to difference Salton Sea LiDAR (collected in 2010, 5 pts/m2) against an SfM point cloud (1500 pts/m2) made in 2016 using a Sony A5100 camera mounted to a DJI Phantom 2, ground control points, and a local GNSS base. Deformation at the 2.75 x 1 km DLV site was estimated by dividing the area into 30 x 30 m squares and finding the rigid-body deformation of each square of the SfM data that produced the closest fit to the older LiDAR point cloud (iterative closest point method). Results show ~30 cm of right-slip and ~8 cm of NE-up vertical motion concentrated within ~40m of the fault trace. Measured right-slip is consistent with creep meter data and site geomorphology, whereas vertical motion may be exaggerated by errors from vegetation and possibly LiDAR processing. Areas of large error are correlated with vegetation. The low slip rate at SC makes detection of creep more difficult; however, the lack of vegetation in the area may improve the resolution of the method, and its application to the site will test the lower limit of detection feasible with this approach.

We compared the accuracies of digital elevation models (DEMs) produced with structure-from-motion photogrammetry using different cameras, unmanned aerial system (UAS) platforms, and georeferencing techniques. The UASs and cameras were a Phantom 4-Pro quadcopter (stock DJI 20Mpixel camera), eBee Plus fixed-wing (20Mpixel SODA camera), and DJI Matrice 100 quadcopter (24.3Mpixel Sony A6000). The Matrice 100 and eBee Plus carried on-board carrier-phase GNSS systems that measured camera positions using post-processed kinetic (PPK) solutions. Point clouds were constructed from photographs taken with each platform at both Rush Valley, Utah, and Dry Lake Valley, California, using ground control points (GCPs) for georeferencing. In addition, point clouds for both sites were made from Matrice and eBee photos using only photo geotags, and using geotags plus GCPs for georeferencing. Point clouds used 1,227 to 3,560 photographs, contained 463,752,119 to 1,327,879,287 points, and covered 0.72 to 3.10 km2. DEMs were created from each point cloud using Agisoft’s built-in binning method, and vertical accuracy was assessed by calculating root-mean-square-error (RMSe) for each DEM relative to bareground checkpoints measured using PPK GNSS. RMSe for the Phantom georeferenced with GCPs only averaged 12.4cm. DEMs georeferenced with camera geotags only averaged 11.9cm higher than checkpoints for the eBee, and 19.0cm for the Matrice, indicating an elevation bias in point clouds and DEMs georeferenced with this method. After removal of the bias, RMSe were as follows: eBee camera geotags only, 4.6cm; eBee camera geotags plus GCPs, 4.9cm; Matrice GCPs only, 6.2cm; Matrice camera geotags only, 5.0cm; Matrice camera geotags plus GCPs, 5.4cm.

Deformed cultural features, creepmeters, and GPS reference-stations show active aseismic creep on the NW-striking central San Andreas Fault (SAF). However, the distribution of creep across the fault is poorly known because surface displacement measurements are spatially sparse. We image creep by differencing a 2017 point cloud generated from unmanned aerial system (UAS) photographs processed with structure-from-motion (SfM) against the 2007 EarthScope Northern California LiDAR Project imagery. We constructed an SfM-based point cloud from 3533 photographs collected in October 2017 with an eBee Plus UAS flown at ~115m above-ground-level. The point cloud contains 1.3x10^9 points, covers a width of ~0.9-1.2km along ~2.8km of the SAF near Dry Lake Valley (36.469, -121.058). The point cloud was georeferenced using 30 ground control points (GCPs) and on-board GPS photograph location measurements. GCP and photograph locations were measured using carrier-phase, post-processed-kinetic GPS solutions, and a local base. 2017 positions were corrected to the IGS2000 2007.3 epoch of the 2007 LiDAR, and uniformly translated to adjust for plate motion. We calculated 3D displacements between the 2007 EarthScope LiDAR and the 2017 SfM imagery using the Iterative Closest Point algorithm with a 25m resolution. The results show ~30cm of right-lateral motion on the SAF with the majority of motion occurring within ~50m of the geomorphic fault trace, and ~2-5cm of SW side down vertical motion. These results are consistent with creepmeter data and show a more complete picture of the spatial distribution of creep along the central SAF.

The Topliff fault (TF), is a west-dipping normal fault extending ~30km along the southeastern side of Rush Valley, Utah. It probably is structurally connected to the Oquirrh-Great Salt Lake Fault, forming Utah’s second longest fault system. It poses significant earthquake risk to Utah County, however the late Quaternary earthquake chronology, slip rate, and seismic hazard are poorly constrained. In Southern Rush Valley, the TF cuts the highstand shoreline of pluvial Lake Bonneville (~18.5Ka) and older Quaternary fan surfaces (40.137, -112.210). Here, we used structure-from-motion photogrammetry to construct a digital elevation model (DEM) and determine the offsets of shoreline and alluvial fan surfaces by surface-rupture on the TF. The 6cm-pixel DEM was made from photographs collected with an eBee Plus unmanned aerial system and georeferenced using geotagged photographs and ground control points, both of which were measured using dual-frequency, survey-grade GPS. Vertical RMSerror of the DEM is 4.4cm We measured shoreline elevation as the intersection of the wave-cut platform and face at 13 locations. Mean elevation on the footwall is 1578.4 ± 0.45m, and 1577.1 ± 0.14m on the hanging-wall indicating 1.3 ± 0.6m offset on the TF since ~18.5Ka. Analysis of the scarp along surfaces below the shoreline also indicates one post-18.5Ka event (1.8 ± 0.3m). Older alluvial fans have 3 to 5 m of additional offset. Slip rate on the TL is probably < 0.1mm/yr. This is broadly consistent with the Oquirrh Fault to the north. However, imperfect scarp and shoreline preservation makes these results inconclusive.