A UAS, commonly known as a drone, is an aircraft without human pilot on board that flies autonomously and is being used in many areas because of its convenience, high resolution, etc. (Huang and Chang, 2014; Fernandez Galarreta et al., 2015; Giordan et al., 2015; Tokarczyk et al., 2015; Bühler et al., 2016; Deffontaines et al., 2017). The UAS used in this study is a modified version of the already-available Skywalker X8 delta-wing aircraft reinforced by carbon fiber rods and covered partially by Kevlar fiber sheets. The drone is launched by hand, flies and takes photos autonomously, then glides back down to the ground by using a pre-programmed flight plans organized by a ground control system; it is controlled by the ground control station and remote controller. The autopilot system used in this study is composed of and modified from the open-source APM (ArduPilot Mega 2.6 autopilot) firmware and open-source software Mission Planner, transmitted by ground–air XBee telemetry. In order to generate a high-resolution DSM (defines the terrain data including buildings and tree canopy) and mosaic orthorectified images, a total of 3767 photos were gathered by Sony ILCE-QX1 camera mounted on the crafts, during five fly missions, for a total coverage area of 33 km² with 8 cm ground sampling distance. The coverage of the adjacent photos is kept at 85 % overlap and 45 % side lap. The raw images are processed by ContextCapture and Pix4Dmapper software mutually; the data sets generated in this study are orthoimages and DSMs, with a grid spacing of 10 cm. Prior to the morphotectonic analysis (e.g., Deffontaines and Chorowicz, 1991; Deffontaines et al., 1993, 1994, 1997, 2017; Pubellier et al., 1994), the quality of the data set (see whole data set in Fig. 2) needs to be evaluated. Eighteen ground control points are extracted with grid size of 2 m resolution of airborne light detection and ranging (lidar) data set and from the airborne lidar-associated 25 cm resolution orthorectified image. Most of the ground control points are situated on crossroads, targeted and georeferenced from orthorectified images, and the elevation is from the airborne lidar data. The comparison of the UAS DSM with airborne lidar data gives a root-mean-square deviation of 4.1 cm with maximum error of 42.5 cm from 26 sites of open, bare ground area, e.g., roads, school playgrounds, unvegetated terrains and parking lots. The elevation of the check points is averaged from an area of 4 m², equal to the grid size of airborne lidar data. The distribution of the ground control points and checked points are located in Fig. 2b. Figure 3 highlights the precision and resolution difference on the available topographic data sets. The 40 m grid DEM was generated in 1986 by the Aerial Survey Office from aerial photos, the 5 m grid DEM was generated by several aerial photogrammetric agencies in 2004, whereas most of the aerial photos were acquired in 2003. The project to create a 5 m DEM data set was funded by the Taiwan Ministry of the Interior. The airborne lidar data were acquired in 2011 and used to generated a 2 m grid DTM (defines the geomorphologic elevation after removing the buildings, trees and vehicles) and DSM, but this data set is not authorized to be published even though it was generated by the authors. The ground control points used in this study were extracted from the airborne lidar data set we acquired some years ago. Figure 3 highlights visually the difference of quality, precision and resolution of each digital topographic data set. One can see the low quality of the 40 m DEM in contrast to the precision and resolution of our UAS DSM. Figure 4 illustrates some case examples of the quality of both the UAS DSM as well as the orthorectified image of the Hengchun Fault. The morphostructures analyses based on photo interpretation are conducted accordingly (Fig. 4c, d, g, h, k, and i).

In order to deduce the regional tectonic activity of the Hengchun Fault, we processed 13 radar images (ALOS PALSAR images acquired in L-band; λ= 23.6 cm, average incidence angle = 34.3^∘) in ascending mode (flight direction 350^∘ E of N) between 17 January 2007 and 28 January 2011, thus providing 4 years of monitoring, through StaMPS software (Hooper et al., 2007). A series of 12 interferograms has been generated, all with respect to a unique “super master” as a reference image situated in the mid-time series that minimizes both spatial and temporal decorrelation (Pathier et al., 2003; Champenois, 2011, Champenois et al., 2011, 2012, 2014, Fig. 8). Note that their perpendicular baselines situated between −1194 and 2417 m are significantly below the critical limit where all interferometric coherence is lost (Hooper, 2009). We used the 40 m ground-resolution Taiwan DTM in order to remove both the topographic and orbital components of the interferometric phase. All interferograms shows an excellent coherence in the Hengchun Plain as well as in the hilly area. The method identifies 20 133 PS pixels with an average density on the whole Hengchun Peninsula of 120 PS km⁻² (see Fig. 9), characterized by their phase stability over the whole period. PS analyses lead us to extract a map of mean LOS surface velocities for each PS and also to catch their displacements through time series. Note that these LOS data are not projected onto the vertical component. The PS-InSAR base (fixed GPS station GS59 corresponds to the black and white star see its location close to Checheng and north of the Hengchun Valley in Figs. 1 and 9) is chosen as it presents a stable (to very low) continuous deformation during the monitoring time period. Nevertheless, this location might be subject to small continuous uplift or small continuous subsidence that may explain local discrepancies with the GPS average annual displacement. We compare our PS-InSAR results (Champenois, 2011; Champenois et al., 2011, 2014) to three fixed GPS stations (HENC, GS57, GS59; see Fig. 9) and two leveling lines acquired in the same monitoring time period (see Fig. 10). For each of the three fixed GPS stations (HENC, GS57, GS59), an average displacement has been calculated, projected into the radar LOS by taking into consideration the various local incidence angles along the distance axis (Hanssen, 2001; Cigna et al., 2011).

Figure 8Available ALOS SAR images and their associated perpendicular baselines. All slave images are linked to the best master image chosen in the middle of the time series that maximizes the sum correlation of all interferogram (Hooper et al., 2007).

Download

Figure 9Mean LOS velocities of PS-InSAR between January 2007 and January 2011, superimposed onto the shaded 40 m ground-resolution Southern Taiwan digital terrain model. The large black circles are the GPS points and the color in the circles indicates their average LOS velocity component during the monitoring time period. The black and white star (situated close to Hăikŏu – see location in Figs. 1 and 2) corresponds to the GPS ground fixed station GS59 and is the base of this PS-InSAR data set. Blue and yellow arrows give their horizontal and vertical velocity. Red line: Hengchun Fault. Location of profiles 1 and 2 (see Fig. 11). Note the active surrection and folding east of the Hengchun Fault in contrast to the subsidence of its western part (Hengchun Valley).

Download

The results (Fig. 9) show an impressive density of PS (more than 140 PS km² in the surroundings of the Hengchun Fault) in this area, where both human activity and luxuriant vegetation prevail. The PS map reveals a clear continuous difference in LOS velocities across the Hengchun Fault where the Hengchun Valley (western side of the Hengchun Fault) subsides, with values ranging from −5 to −10 mm yr⁻¹, in contrast with the westernmost Hengchun ridge, which is uplifting. This is evidenced in Fig. 9 by a rapid transition from blue/green to orange/red PS dots. Taking into account the PALSAR ascending satellite orbit and the NNW–SSE Hengchun Fault, trending of both geometries is close to the best configuration to detect and measure the fault displacements (Champenois, 2011). Consequently, Fig. 9 led us to draw, with much more accurate detail than Shyu et al. (2005) or Lin et al. (2009), a new place for this active Hengchun Fault within the Hengchun Plain (Champenois, 2011). The eastern side of the Hengchun Fault is characterized by the rapid increasing uplift of the Kenting Mélange up to a maximum that seems to parallel the Hengchun Fault. Farther east the PS LOS velocities decrease with the distance to the Hengchun Fault so the eastern Hengchun Peninsula appears to be slightly tilted. We observe in the Hengchun Peninsula a similar dissymmetric active vertical behavior as in the coastal range (eastern Taiwan): the western part of the relief is highly actively uplifting and folding in contrast to the eastern part, which is characterized by a decreasing uplifting and which is slightly tilted to the east (see also Deffontaines et al., 2017). We note in Figs. 10b and 11 the coherence of the PS LOS velocities and the GPS measurements (colored large dots that we put into the same LOS radar geometry). Consequently, those PS-InSAR results offer a precise mapping of the active interseismic splays of the Hengchun Fault and allow the quantification of the tectonic displacements.

Figure 10(a) Location of the leveling lines on the Hengchun Peninsula. (b) and (c) Associated PS mean velocity with error bars (blue dots), leveling data (red square and in pink: error bar) and GPS stations data (yellow triangle), projected on LOS. Leveling lines highlight the vertical component of the deformation. The easternmost and westernmost reference points are on an enclosed curve line of the class 1 level reference net defined by Taiwan's Satellite Survey Center of Department of Land Administration (see SSCDLA, 2018). As leveling line 2 parallels the Hengchun Fault (points 6 to 8), it is represented as a red quadrangle. Note the general agreement of both interseismic leveling and PSI vertical topographic displacements on those two transverse profiles across the whole Southern Hengchun Peninsula. They show subsidence in the western part of the Hengchun Fault (HeF) contrasting with uplifting due to active folding on its eastern side. In addition, there is little significative vertical interseismic displacement on the Kenting Fault location on both profiles during the monitoring time period.

Download

Regarding the Kenting Fault, with the data used in this study we see no gradient of PS LOS velocity nor noticeable differential displacements on the GPS stations situated around the Kenting Fault (see Fig. 9 and profiles of Fig. 10a–b). Moreover, from the geological mapping analysis (Chang et al., 2003), the Kenting Fault has a sinusoidal planimetric geometry that follows more or less the topographic isocontour of the relief. So the Kenting Fault is a low-dipping thrust fault. Consequently, this fault may be either locked at the surface or an inherited inactive fault, or it may have a tectonic activity that we do not observe on this PS-InSAR data set during the monitoring time period.

We note a great coherence in between the PS and the available GPS measurements (colored large dots in Fig. 9) provided on two stations. For comparison purposes, the average annual GPS displacements acquired in the InSAR monitoring time period have been compared with the radar LOS. We compared also these PS results with leveling data acquired during the same period of time (between March 2008 and April 2011, during three campaigns of measurement). Figure 10 compares the leveling data, the mean LOS velocities of PS along the leveling lines and GPS data for the two stations situated close to the leveling lines. Note the agreement between all of them. The local heterogeneity of the displacements (e.g., small subsidence and small uplift) within the Hengchun Valley is highlighted by both the HENC GPS station that gives the absolute uplift displacement and the PS-InSAR data with various colored dots, ranging from blue to orange, situated on both sides of the O value.

Figure 11E–W projection of mean LOS velocity in mm yr⁻¹ (blue dots with error bars, E on the right) on two profiles transverse to the northern part of Hengchun Valley (close to the cities of Checheng and Hengchun; see profile location in Fig. 9). Black line: topography; yellow triangle: GPS point; HeF: Hengchun Fault; KeF: Kenting Fault. Note that Figs. 10 and 11 both show similar topographic displacements: the LOS offsets close to 10 ± 2.5 mm yr⁻¹ associated with the Kenting Mélange uplift east of the Hengchun Fault, which contrasts with the generally subsiding Hengchun Valley. Kenting Fault has no vertical interseismic offset during the monitoring time period.

Download

Moreover, in order to better characterize the Hengchun Fault activity and quantify the LOS velocity variation on both sides of Hengchun Fault, we computed the LOS velocity offset across the fault, using 31 profiles perpendicular to the fault line. Each 10 km long PS profile is perpendicular to the Hengchun Fault and characterized by an equidistance of 400 m. In addition, each PS data point situated on both sides (200 m vicinity) is projected on each PS profile. Each data point is only represented on only one profile. The number of PS dots per profile varies between 400 and 1200 (Champenois, 2011). This offset characterizes the slip rate of the fault. We present two specific profiles in Fig. 11 (see location Fig. 9). The two PS profiles, which run transverse to the Hengchun Fault, one north of the Hengchun Valley close to the Checheng city and the second one southward (close to Hengchun city), show LOS offset less than 10 ± 2.5 mm yr⁻¹. Note the high number of PS reflectors situated in the cities of Checheng and Hengchun that confirm that on both profiles we evidence a clear jump in the PS magnitude that corresponds to the aseismic displacement of the Hengchun Fault (Champenois, 2011).

Figure 12Hengchun Fault along-strike variation of LOS velocity offset (perpendicular to the fault, in mm yr⁻¹) including Hengchun and Checheng. Note the active interseismic displacement and its along-strike variations which confirms that Hengchun Fault is an active interseismic creeping fault of Taiwan. Furthermore the LOS velocities' variations are important in terms of seismic hazards as they reveal the location of slowly creeping areas where stress may accumulate at depth, such as those areas close to both Checheng and Hengchun and higher active tectonic areas such as in between Checheng and Hengchun or south of Hengchun Valley. Part of the deformation may also be distributed in the eastern part of the Hengchun Fault (e.g., the Kenting Fault and/or within the intrusion of the Kenting Mélange in between both Hengchun and Kenting faults; see Fig. 14).

Download

So thanks to a series of more than 30 profiles perpendicular to the fault, we computed the interseismic LOS velocity offset along the fault, from north to south, with LOS values (from min 4.9 mm yr⁻¹ to max 10.2 mm yr⁻¹), reaching an average 8 ± 2.5 mm yr⁻¹ during the monitoring time period, which reveals, using two linear regressions calculated on both side of the Hengchun Fault, the clear tectonic activity of the Hengchun Fault (Fig. 12). Note the good coherence between slip rates derived from PS analysis and the GPS measurement (projected in LOS by taking into account each local incidence angle along the distance axis) represented by the yellow triangle in see Fig. 12. This good fit between the fixed GPS stations validates our PS results. This is a major result from this study as it demonstrates the along-strike interseismic activity of the fault (Champenois, 2011; Champenois et al., 2011, 2012; this study) and allows us to obtain the average annual interseismic LOS displacement of the Hengchun Fault and its spatial variation along the fault. Variations of the along-strike velocity offset of Hengchun Fault need to be carefully monitored and analyzed in order to better characterize potential seismic hazards, especially the low interseismic creeping areas where stress may accumulate (see Fig. 12) and may correspond to a partially locked part of the fault source of future earthquakes (close to Hengchun Township, for instance).

StaMPS also gives access to time series of displacement, from 2007 to 2011 (see Champenois, 2011; Champenois et al., 2011, 2012). We can thus derive at each image acquisition date the displacement related to the fault (a cumulated slip value) and then monitor its slip variation through time. The time series of slip values (Fig. 13) does not show any clear seasonal accelerations or decelerations, but only expresses a rather linear behavior of the fault during the monitoring time period. This time series should be updated with new SAR images in order to get a longer monitoring period (Champenois, 2011; Champenois et al., 2011). It will help to know whether the Hengchun Fault has a linear displacement through time (or not).

Figure 13Time series of cumulated slip (on profiles 1 and 2) where no particular seasonal effects prevail.

Download

Figure 14(a) Simplest model of active interseismic tectonic deformation of the Hengchun Fault during the monitoring time period. It is compatible with the oblique geodynamic convergence (large red arrow on the right), GPS, leveling, interferometric active displacements, and the detailed UAS DSM photointerpretation. In addition, it is compatible with Rokkaku and Makiyama (1934) and CGS and CPC geological mappings (Chang et al., 2009b; Deffontaines et al., 2016). 1: Fault; 2: Thrust; 3: Fault plane; 4: Taiwan Strait/South China Sea; 5: Left-lateral component; 6: Transpressive left-lateral component; 7: Fold (anticline and syncline). The GPS data and PS mean LOS velocities imply that Hengchun Fault acts as a left-lateral transpressive strike-slip fault with an actively folding and uplifting anticline (hanging wall) contrasting with the aligned active subsidence of the Hengchun Valley that acts as a trending parallel to the Hengchun Fault. One can see the Hongtsai canyon at sea (see Deffontaines et al., 2016). (b) Simplified lithostratigraphic column of the Hengchun formations in the western Hengchun Peninsula. Al: alluvial deposits; HcL: Hengchun limestone; Ma: Ma'anshan formation; Mt and Mtg: different facies of Mutan formation (Mtg: Loshui sandstone – CGS geological map); KeM: Kenting Mélange; HeF: Hengchun Fault; KeF: Kenting Fault; Mio: Miocene; Pli: Pliocene; Ple: Pleistocene; Hol: Holocene strata (modified from CGS, CPC and previous references).

Download

Many different tectonic models have been proposed on the active Hengchun Fault zone (thrust, left-lateral strike-slip, transpression and transtension; see Chang et al., 2003, 2009b). However, they miss both structural and active tectonic arguments to better understand the Hengchun Fault (e.g., Liew and Lin, 1987; Sung, 1991; CPC, 1993; Chen and Liu, 1993, 2000; Chang et al., 2003, 2009a, b; Chen et al., 2005; Vita-Finzi and Lin, 2005).

The simplest model that we propose herein (Fig. 14) that fits with the oblique NW-trending Taiwan geodynamic convergence (Suppe, 1984), the E–W displacements measured by GPS data (Yu et al., 1997), the geological data set and the geological mapping (CPC, 1993; CGS in Sung, 1991; Chang et al., 2003, 2009b; Giletycz et al., 2015), our detailed high-resolution UAS DSM photointerpretation (Fig. 7) and our interferometric 2009–2011 PS LOS displacements (Champenois, 2011; Champenois et al., 2011, 2014, Fig. 9) is given for the western part of the onshore Hengchun Peninsula (Fig. 14).

From all used geodetic techniques (GPS, leveling lines and interferometric), if we generalize the two leveling sections, which are coherent to both the three fixed GPS stations and the PS-InSAR results, the western Hengchun Peninsula is slowly uplifting (less than +10 mm yr⁻¹), contrasting the slow subsidence of the eastern part of the Hengchun Peninsula (0 to −2.5 mm yr⁻¹). So as we reveal a continuous active differential displacement across both sides of Hengchun Fault, which is deduced from leveling lines, GPS and PS-InSAR, Hengchun Fault shows the existence of a clear interseismic creeping component during the monitoring time period, with an average value of 8 mm yr⁻¹ along the LOS, confirmed by the leveling monitoring (see leveling line 1: 8 mm offset – Fig. 10b). The variation of displacements on both sides of the Hengchun Fault zone also reveals the progressive folding of the Hengchun Fault zone: the eastern hanging wall is progressively shortening and uplifting (as revealed by GS57 and the two leveling lines) contrasting to the progressive subsidence of the Hengchun Basin (footwall situated west of Hengchun Fault – see Figs. 10b, c and 11, profiles 1 and 2). These displacement results (Figs. 9, 10 and 11) combined with the structural photointerpretation (Fig. 7) show that the fault is only locally outcropping and that active folding occurs, increasing progressively with time, stress and strain at depth.

Consequently, due to the rectilinear structural geometry, the Hengchun Fault appears to act as an almost vertically active left-lateral transpressive strike-slip fault with an active uplifting and folding (anticline) on its eastern hanging wall, in contrast to the active subsidence of the western footwall (which corresponds to the easternmost part of the dissymmetric Hengchun Basin, Fig. 14). Moreover, the Hengchun Fault is not outcropping along one small, straight, continuous north-to-south line but is instead a rather large “fault zone” made of little active fault segments (see location mapped in details, Fig. 7). One may note that the thrusting and left-lateral strike-slip motion have been deduced from both from the morphostructural analysis of the high-resolution DTM (see Fig. 4c, d, g and h) and the obliquity of the GPS arrows already evidenced in the bibliography (see blue arrows in Figs. 1 and 9, and Chang et al., 2003). Consequently, Fig. 14 proposes a simple model where the Hengchun Fault acts as a 160^∘ E N-trending transpressive left-lateral strike-slip fault zone. Its active hanging wall is overthrusting and is actively folding the topographic surface that tilts and folds the paleosurfaces: the Hengchun limestone and marine terraces (see Liew and Lin, 1987, and Chen and Liu, 1993). The Hengchun Basin (Hengchun Fault footwall and west of it) under the transpression is also folded and is actively subsiding with a parallel to the fault syncline axis that create the lagoonal/marine deposition of the Pleistocene–Holocene Hengchun Basin (Chen et al., 1991).

Thus, the Hengchun Fault, as many faults in central and southern Taiwan, presents both left-lateral strike-slip and thrusting components such as the Chelungpu Fault reactivated during the Chichi earthquake (21 September 1999) as well as any Taiwan foothill transfer fault zones (e.g., Deffontaines et al., 1997).

Regarding the amplitude and the components of the active Hengchun Fault deformation, the two leveling lines reveal the vertical offset of the Hengchun Fault, consistent with interferometric displacement toward the LOS, with a value of 8 ± 2 mm yr⁻¹, which corresponds to the vertical geodetic interseismic slip rate. Note that this short-term value is a bit different than the vertical long-term slip rate deduced from the marine terraces dating 6.3 mm yr⁻¹ at Hăikŏu and 3.8–6.1 mm yr⁻¹ at Nanwan (Liew and Lin, 1987; Chen and Liu, 1993). From this slight difference in between the long-term and the vertical geodetic slip rates, we may deduce the fact that the Hengchun Fault may be not only a creeping fault. But due to the folding component of the eastern side of the Hengchun Fault, it is needed to locate much precisely the in situ dated samples (Chen and Liu, 1993) to reconstruct and inverse correctly the long-term deformation history.

Chang et al. (2003) characterized the active horizontal slip rate of the western Hengchun Peninsula by comparing the directions and amplitude of the three fixed GPS stations. Note that these values take into account both Hengchun and Kenting faults without individualization.

The ALOS SAR images were obtained from the Japan Aerospace Exploration Agency. UAS data were acquired by the authors. The UAS data can be obtained upon request to the corresponding author (epidote@ntut.edu.tw).

The authors declare that they have no conflict of interest.

This article is part of the special issue “The use of remotely piloted aircraft systems (RPAS) in monitoring applications and management of natural hazards”. It is a result of the EGU General Assembly 2016, Vienna, Austria, 17–22 April 2016.