Characteristics of debris avalanche deposits inferred from source volume estimate and hummock morphology around Mt Erciyes , central Turkey

Debris avalanche caused by volcano sector collapse often forms characteristic depositional landforms such as hummocks. Sedimentological and geomorphological analyses of debris avalanche deposits (DAD) are crucial to clarify the size, mechanisms, and emplacement of debris avalanches. We describe the morphology of hummocks on the northeastern flank of Mt. Erciyes in Kayseri, central Turkey, likely formed in the late Pleistocene. Using a remotely piloted aircraft system (RPAS) and the structure-from-motion multi-view stereo photogrammetry (SfM), we obtained high-definition digital 20 elevation model (DEM) and orthorectified images of the hummocks to investigate their geometric features. We estimated the source volume of the DAD by reconstructing the topography of the volcano edifice using a satellite-based DEM. We examined the topographic cross sections based on the slopes around the scar regarded as remnant topography. Spatial distribution of hummocks is anomalously concentrated at a certain distance from the source, unlike those that follow the distance-size relationship. The high-definition land surface data by RPAS and SfM revealed that many of the hummocks are 25 aligned toward the flow direction of the debris avalanche, suggesting that the extensional regime of the debris avalanche was dominant. However, some displaced hummocks were also found, indicating that the compressional regime of the flow contributed to the formation of hummocks. These indicate that the flow and emplacement of the avalanche were constrained by the topography. The existing caldera wall forced the initial eastward flow to move northward, and the northside caldera wall forced the flow into the narrow and steepened outlet valley where the sliding debris underwent a compressional regime, 30 and out into the unconfined terrain where the debris was most likely emplaced on an extensional regime. Also, the estimated volume of 12–15  ́ 10 m gives a mean thickness of 60–75 m, which is much deeper than the reported cases of other DADs. This suggests that the debris avalanche must have flowed further downstream and beyond the current DAD extent. Assessments of the DAD incorporating the topographic constraints can provide further insights into the risk and mitigation of potential disasters in the study area. 35

. Such recurrence of sector collapse is often expected in single or multiple directions (Lagmay et al., 2000;Yoshida and Sugai, 2007;Paguican et al., 2012Paguican et al., , 2014. However, since the recurrence interval is not constant and the stationary period can last for more than 10 3 to 10 4 years (Moriya, 1988), the areas around an unstable volcano can be highly urbanized without considering the risk of a sector collapse. In particular, the downstream area of debris avalanche deposits is often flat enough to be urbanized or cultivated. In such urbanized areas, detailed information of 5 the past debris avalanche including its timing, size, and kinematics is crucial for the future hazard assessment.
Debris avalanche caused by a volcanic sector collapse forms characteristic depositional landforms called hummocks, composed of large block facies within a matrix (Siebert, 1984;Ui et al., 1986Ui et al., , 2000Orton, 1996). Block facies are composed of the volcanic edifice transported, potentially showing an intact clast or multiple deformed clasts of a single or multiple rock types (Evans and DeGraff, 2002). Within the domain of debris avalanche deposits (DAD), hummocks often 10 have clear boundaries and are readily identifiable, thus providing a chance to carry out the robust size and shape analysis (Mizuno, 1958;Hashimoto et al., 1979). Such the morphology of hummocks, located on the surface of DADs, can be used to estimate the characteristics of the debris avalanche, i.e., the kinematics of the sliding mass (Dufresne and Davies, 2009;Koarai et al., 2008;Yoshida and Sugai, 2010;Yoshida, 2012Yoshida, , 2013Yoshida, , 2014Yoshida et al., 2012). Although sedimentological investigations of the internal structures of DADs often provide insights into the transport mechanisms (e.g., Glicken, 1996;15 Bernard et al., 2008;Shea et al., 2008), geomorphological or geometrical analyses of the surficial morphology of DADs including hummocks can also provide estimations on the size and processes of the debris avalanche. In particular, because the sedimentary structure is mutually related to the hummock distribution and shape, the distribution and morphology of hummocks formed in DADs provide an opportunity to examine the volumetric and kinematic characteristics of the debris avalanche (Dufrense and Davies, 2009;Yoshida and Sugai, 2010;Yoshida et al., 2012). During the sliding of a sector 20 collapse, the fractured mass of the original volcanic edifice forms collapse structures along with the alternate regimes of extension and compression (Paguican et al., 2012(Paguican et al., , 2014. This often results in the alternate extensional and compressional structures of DADs along the avalanche path, while the compressional regime may result in the denser distribution of hummocks particularly around the flow front . Hummocks are often in the order of tens to hundreds of meters in size, so aerial photographs are often used for their 25 identification and morphological analysis (Glicken, 1996;Yoshida, 2012Yoshida, , 2013Yoshida, , 2014Yoshida and Sugai, 2010;Yoshida et al., 2012). Digital elevation models (DEM) by airborne laser scanning (ALS) within 1-5 m resolution can also be used for such analysis (e.g., Hayakawa et al., in press), but such high-definition data including paired aerial photographs and ALS-DEMs are often unavailable in many areas. High-definition satellite remote sensing imagery and DEMs can also be used for the analysis of hummocks, but the acquisition cost is often high and may not be readily obtained in many cases. Due to 30 availability, details of many known DADs remain unexamined.
Recent developments in the remotely-piloted aerial system (RPAS), as well as that in the structure-from-motion multiview (SfM-MVS) photogrammetry, have enabled on-site, cost-efficient acquisition of high-definition topographic and imagery data (e.g., Westoby et al., 2012;Fonstad et al., 2013;Obanawa et al., 2014;Hayakawa et al., 2016). Such detailed earth surface data are useful for detailed topographic analysis, including topographic feature extraction and landform 35 classifications, for areas of approximately 0.1-10 ´ 10 9 m 2 that cannot be achieved by low-resolution satellite imagery (e.g., Koarai et al., 2008). Moreover, although ordinary aerial photographs taken by manned aircraft have often been utilized to investigate topographic features of DADs in such a small-to medium-sized area (Siebert, 1984;Glicken, 1996;Sugai, 2007a, 2012), high-quality stereo-paired aerial photographs are often unavailable in many areas over the world.
Without the availability of ordinary aerial photographs, RPAS-derived land surface data (topography and imagery) are 40 therefore useful enough even if not fully taking the advantage of its highest resolution. Furthermore, the higher resolution data acquired by RPAS should serve as a potential archive for future studies, because the rapid increase in the use of RPAS will enable further data collection of such high-definition data in many areas.
Using the combined RPAS-and satellite-derived topographic data, we analyze the morphology of hummocks and DAD on the northeast flank of Mt. Erciyes, previously described by Şen et al. (2003). We use RPAS for the acquisition and analysis of detailed, high-definition morphological data for the hummocks formed on the DAD, where ordinary stereo-paired aerial photographs nor ALS-derived topographic data are unavailable. The high-definition topographic data are crucial for the identification and analysis of 10-to 100-m scale hummocks. We also utilize topographic data at 10 m resolution from 5 satellite SAR imagery for the analysis of the surrounding areas. The volume of the DAD crucial for describing the sector collapse, which is hard to know from the extent and depth of the deposits, is obtained by reconstructing the original topography of the source area.

Study area
Mt. Erciyes, located in the south of Kayseri City in central Turkey (Fig. 1), is the highest stratovolcano in this region with an 10 elevation of 3917 m at its summit. The current volcano edifice has been formed after the pre-Erciyes volcanic activities terminated about 3 Ma with an extensive ignimbrite eruption (Innocenti et al., 1975;Şen et al., 2003). The stratigraphy of Mt. Erciyes is composed of basaltic, andesitic, and dacitic lava flows in ca. 2.5-0.2 Ma, followed by pumice and pyroclastic flow deposits (Şen et al., 2003).
The youngest deposit is the DAD, emplaced after 83 ka (Şen et al., 2003). Moraines that formed in the last glacial 15 maximum (21.3 ka) are also present in the valley along the avalanche flow (Üçker Valley, A in Fig. 1) (Sarıkaya et al., 2009). This gives an approximate age of the sector collapse of between 20-80 ka.
The DAD is observed within 2,000-9,000 m downstream along a fluvial valley from the mountain top, covering an area of ~14 ´ 10 9 m 2 (Şen et al., 2003) (B and northern part of C in Fig. 1). The downstream extent, however, is hard to identify due to limited exposure, and erosion and remobilization of DAD after emplacement, particularly in the fluvial valley. 20 Nevertheless, we found several outcrops of the DAD along a newly constructed highway with locations shown as stars in Fig. 1. The farthest outcrop is 16,000 m away from the summit of Mt. Erciyes. Moreover, hummocks appear with in 11,000-13,000 m downstream from the summit. These indicate that the potential extent of the DAD is farther downstream over the area mapped by Şen et al. (2003), with an area greater than 20 ´ 10 9 m 2 .
Based on the collapse scar (Fig. 2a), the debris avalanche was supposed to be flowing to the east (A in Fig. 1). The flow 25 then turned to the north due to confinement by pre-existing caldera walls. Unfortunately, we could not find any outcrops in the field including abutting deposits on the footslopes of caldera wall. However, the present lake, located in the south of the bend of the DAD and now serves as a reservoir (D in Fig. 1), is possibly a remnant of a dammed lake. Based on this, it is clear that the debris avalanche flow did not largely turn to the south, but followed northward along the pre-existed caldera wall. 30 Hummocks, characteristic mound-shaped topographic features of DADs, are densely located in an approximately 2 ´ 10 9 m 2 area around 11,000-13,000 m (along-valley distance) from the summit of Mt. Erciyes. Our topographic measurements, therefore, focus on this domain. As noted, this domain of hummocks is out of the range of the DAD area previously reported (B in Fig. 1; Şen et al., 2003), but the potential extent of the DAD likely covers this domain and farther downstream (E in Fig. 1). As far as we investigate, no hummock is observed in other areas of the DAD along the valley. 35 Mound-shaped features on the eastern side of the DAD (C in Fig. 1) are confirmed to be basaltic or andesitic lava domes.
The left-lateral strike slip Erciyes Fault is mapped to potentially cut across the Mt. Erciyes along the north-northeast direction (Emre et al., 2011). It is a member of the Ecemiş Fault group of the East Anatolian Fault zone. The age and history of the fault activity, however, are not well known.
The climate in the area is warm and dry, with an annual precipitation of ~400 mm, resulting in scarce vegetation cover. 40 Climatic fluctuations including relatively wet periods in the late Pleistocene to Holocene has been dry enough not to grow 4 dense vegetation since the last interglacial (Kuzucuoglu et al., 1999;Bayer Altin et al., 2015;Pickarski et al., 2015). Hence, the slope processes have not been so active to significantly modify the topography. The original topography of the debris avalanche, as well as the landforms surrounding the sector collapse, is therefore supposed to be well preserved, except for areas around valley bottoms where fluvial or glacial modifications have been active and volcanic alluvial fans are well developed. 5

RPAS-based SfM-MVS photogrammetry and hummock mapping 10
To measure the detailed surface morphology of the DAD, we apply the SfM-MVS photogrammetry using low-altitude aerial photographs taken by an RPAS. (Fig. 3) The RPAS includes a small unmanned aerial vehicle (UAV) equipped with a digital camera that obtains low-altitude aerial photographs in the field. The image data obtained are used to capture earth surface condition and morphology data with a relatively broad areal coverage, typically for 0.1-10 ´ 10 9 m 2 . We used two UAVs for the measurement: DJI Phantom 2 on which a Nikon Coolpix A (sensor resolution: 4928 × 3264 pixels, 35-mm equivalent 15 focal length: 28 mm) digital camera or RICOH GR (sensor resolution: 4928 × 3264 pixels, 35-mm equivalent focal length: 28 mm) is mounted, and DJI Phantom 3 Professional with a built-in stabilized camera FC300X (sensor resolution: 4000 × 3000 pixels, 35-mm equivalent focal length: 20 mm).
The UAVs are flown from several locations in and around the hummocky area. The UAVs were manually operated, and flight courses were set to cover the area of interest sufficiently. During flights, the camera shutter automatically records 20 data every 2 seconds. A flight of ~10 to 20 minutes takes 300-600 photographs.
Geographical coordinates of several ground control points (GCP) were obtained, using a post-processed kinematic global navigation satellite system (GNSS) receiver. Trimble GeoExplorer 6000XH was used as the GNSS rover, with log data corrected using the fixed GNSS base station of the International GNSS Service (IGS) network. The GCPs are in the main target area around the hummocks. Characteristic objects that are readily identifiable in the aerial photographs were 25 selected as GCPs such as road intersections, flat stone surfaces on bridges, and large flat-topped boulders. The projection of the geographical coordinates is set to UTM zone 36N on WGS 84 (EPSG:32636).

5
We carried out SfM-MVS photogrammetry using PhotoScan Professional Edition photogrammetric software by Agisoft LLC. From multiple photographs, SfM process provides three-dimensional positions of the stereo-paired photographs, in which the same features are identified among the paired photographs as tie points. Although geographical coordinates of the camera locations can be known from the GNSS records in the built-in camera of the UAV devices, the positional accuracies of the single-source GNSS receiver of the UAVs are low, being on the order of meters. Several GCPs 10 whose coordinates are obtained by the post-processed GNSS receiver are therefore placed in the image data to improve positional accuracies at a decimeter level, which are low enough for the analysis of hummocks in 10 1 -10 3 m 2 scales. The tie point cloud is adjusted to fit the GCPs (bundle adjustments), and the points with large reprojection errors (≥1.0 pixels) are removed to achive decimeter-scale accuracies of the bundle adjustments. By the MVS-photogrammetric process, denser three-dimensional points are further obtained from the aligned and paired images. Based on the resultant point density and 15 ground resolution of the images, the resolution of the raster data including a DEM and an orthorectified composite image is determined.
Slope distribution and topographic contour lines are derived from the DEM using GIS software (ArcGIS Desktop 10.3 by ESRI). Hummock bases were manually traced using the DEM-derived topographic data and orthorectified image.
Although several approaches to automatically derive hummocky mounds from DEMs that may be useful for inaccessible 20 remote areas have been proposed, manual interpretation from stereo-paired aerial photographs and field validation have long been the standard method and is still the robust way (Glicken, 1996;Yoshida et al., 2012;Hayakawa et al., 2017). In this study, we carried out manual identification of hummocks using the RPAS-derived DEM and orthorectified aerial images instead of stereo-paired aerial photographs that are unavailable in this area. The procedures are as follows.
1) To identify local convex mounds in the DEM, the elevation values are multiplied by -1 to generate a flip-side of 25 the DEM, and the local depressions in the inverse surface are highlighted by applying the sink-fill process. For this process, the D8 flow directions, the potential water flow direction simply identified using the lowest elevation in the surrounding eight cells (Jenson and Domingue, 1988), are derived from the flip-side DEM, and the depressions are virtually filled up to match with the local trend surface. The filled areas roughly correspond to the local mounds in the original DEM. This generally results in the interpretation of some large hummocks having several peaks, if 30 those peaks are closely located and elevated compared to the surrounding flat areas. If the surrounding areas are sloping, the mound base can be underestimated due to the insufficient filling of the depressions (flip-side of the mounds). Therefore, these local mounds are used as a supportive material for the manual interpretation of hummocks 2) Together with the image of local mounds, orthorectified ground-surface image and slope maps are used to 35 manually detect boundaries of the actual mounds. Because the flat lands, which are predominantly composed of the DAD matrix, are mostly cultivated whereas the mounds of hummocks are unsuitable for cultivation due to their rocky composition, the boundaries of the hummock mounds can be clearly interpreted in the ground-surface images. Also, the gap of slope angles is clear enough to separate the steep mounds and surrounding flat lands.
3) Topographic contour lines and hillshade image maps derived from the DEM are also supportively used for the manual tracing of mound boundaries. However, boundaries are not always horizontal and the contour lines do not 5 generally follow the mound boundaries except in flat areas. In that case, interpretations by the local mound images, aerial images and slope maps are given priority. 4) After polygon vector data of hummock boundaries are obtained, a field check was done to ensure the reliability and accuracy of interpreted hummock boundaries. In the field, the composition of mounds is mainly checked to confirm whether the interpreted mounds are actually hummocks, and whether the boundaries are correct. 10 The geometric features of hummocks are then examined in GIS and area is calculated for each polygon. The height of a hummock is estimated by constructing a flat basal surface by interpolating elevation values on the polygon peripherals by triangular irregular network (TIN) surface. The maximum difference between the actual surface and the estimated basement surface is regarded as the height of the hummock. Although the actual mass of a hummock should exist in the underground areas, the depth of such the submerged structure is less known, and the surficial height of hummocks have often been utilized 15 as a representative index showing their morphological characteristics (Glicken, 1996). The upper-half volume of a hummock, i.e., the volume of the hummock mound above the basement surface, is also obtained as one of the proximal indicators of the hummock size. The volume is calculated as the sum of the elevation differences between the actual surface and the estimated basement surface for each cell in a hummock polygon.
Relationship between hummock alignments and the flow direction of debris avalanche has been studied. The 20 orientation of hummocks is a key morphological measure to investigate the dynamics of debris avalanche (Glicken, 1996;Paguican et al., 2012;Yoshida, 2014). Here, two directionality indices of the polygons are calculated based on their major axis. One is the direction from the north, and another is the displacement angle against the flow direction of DAD. These two are similar measures because the main direction of the DAD is to the north, but different in that the former has both positive (east) and negative (west) values, while the latter only gives absolute values. Figure 4 shows the definition of the 25 displacement angle of the major axis of a hummock: the absolute angle between the main direction of the valley-filling DAD and the major axis (Yoshida, 2014). The distance from the source along the valley, as explained later, is also assigned to the centroid of each polygon (Fig. 4).

Reconstruction of the source area of DAD using PRISM-DEM
We use satellite-based imagery and topographic data to cover wider areas than the RPAS-derived data: AVNIR-2 (Advanced Visible and Near Infrared Radiometer type 2) and PRISM (Panchromatic Remote-sensing Instrument for Stereo Mapping), both of which were mounted on ALOS (Advanced Land Observation Satellite). The AVNIR-2 sensor has four bands for visible and near-infrared wavelengths, with a typical resolution of 10 m. These data are too coarse for hummock extraction 5 but can be used as a background data. The AVNIR-2 data was to confirm the hummocky topography in and around the DAD. The PRISM optical sensor is a panchromatic radiometer for wavelengths of 0.52-0.77 µm, with a spatial resolution of 2.5 m. The original data was processed to generate a 3-m resolution DEM with a vertical accuracy of ~6 m (Takaku and Tadono, 2007;Habib et al., 2017). This is resampled to 10 m resolution to avoid surficial noises. Although there are still errors particularly around the borders of mosaic images due to mismatch, the DEM can be used as a background topographic 10 condition covering the whole area of the sector collapse, DAD, and surrounding areas. The resolution and accuracy of the DEM are, however, unsuitable for the extraction and analysis of small hummocks.
The PRISM-DEM was further processed to show topographic characteristics in the study site. The distances along flow paths based on the D8 flow directions are generated using the hydrological toolset in the GIS software, providing the position of the hummocks (the distance from the source) based on the flow directions (Fig. 4) linearly-extrapolated lines (Fig. 5). The function of a quadratic Bézier curve is given as: where B(t) is the estimated elevation, and t represents the relative position between the points of P 0 and P 2 (0 ≤ t ≤ 1). Eq. (2) 9 Reconstruction of the original topography by linear and Bézier methods used 3-dimensional TIN models. The differences between the reconstructed TIN surface and the present land surface are calculated as the estimated volumes of the sector collapse.

Morphometric characteristics and distribution of hummocks
There were eight UAV flights that took 2900 photos. Of these, 1572 were used for photogrammetry after excluding low-10 quality or misaligned images. To adjust the geographical coordinates of the tie points and camera locations, GCPs were set in the images and the tie points and camera locations were adjusted to fit the GCP-derived coordinates. Five GCPs were set around the main hummock area. Geographical coordinates taken by the post-processed GNSS provided positional errors of 0.31 m and 0.41 m in horizontal and vertical directions, respectively. The bundle adjustment using these GCPs resulted in the estimated horizontal and vertical accuracies of the topographic model of 0.48 and 0.99, respectively. Although the 15 number of GCPs is not many, and the errors are on the order of decimeters, we assume that these error values are sufficiently low for the 10 1 -10 3 m 2 hummock analysis. After removing the points with large reprojection errors (>0.79 pixels), the number of valid tie points for the paired images was 14,220,242. The multi-view stereo photogrammetry was then performed to generate the dense point cloud of 51,505,810. Based on this three-dimensional point cloud and the original images, the data of land surface imagery and topography were obtained, covering an area of 8.5 ´ 10 9 m 2 . The resolution of the raster-20 type topographic data (DEM) was set to be 36 cm, based on the average point density of 7.8 pts/m 2 . The orthorectified image was given with a resolution of 9.0 cm, based on the average ground resolution of the original image of 8.96 cm/pix. The outline of hummocks was then traced from the RPAS-derived topographic and image data (Fig. 6). In total, 65 hummocks were extracted, all are within ~11,000 to 13,000 m away from the DAD source. Table 1 shows the basic morphological properties of the hummocks. The polygon area ranges over two orders of magnitude from 100 m 2 to 20,000 25 m 2 , with a mean area of 3,315 m 2 . The height shows less variation from 1 to 20 m with a mean of 5.4 m, while the volume largely ranges from 200 to 770,000 m 3 with a mean of 79,429 m 3 . The lengths of major and minor axes are tens to hundreds of meters with means of 69 m and 41 m, respectively. The mean elongation ratio, i.e., the length ratio of the major axis to the minor axis, is 1.7, equivalent to an approximate aspect of 5:3 of hummock bounding box.
The displacement angle (the relative orientation against the flow direction of the debris avalanche) almost fully ranges 30 from 0° to 90°, with an average of 38°. Fig. 7 shows more details of the orientation and displacement of the hummocks in histograms. The north-based orientation shows large fraction toward the north (Fig. 7a), while the displacement angle shows a bimodal distribution in the histogram, with relatively large fraction around 0-40° (Fig. 7b). These indicate that many hummocks tend to follow the direction of the northward flow of the debris avalanche, but some are exceptionally placed with large displacements against the flow direction.    (Fig. 2c). This can be regarded as a knickzone, which has relatively steeper gradient than the adjacent upstream and downstream zones (Hayakawa and Oguchi, 2006).

25
Transverse cross profiles along the DAD extracted from PRISM-DEM show downstream widening of the DAD in elevations of 2800-2200 m, and narrowing in 2200-1600 m (Fig. 9). The potential extent of the DAD in the downstream limit, corresponding to the location of the northernmost outcrop identified in the field (Fig. 1), was highlighted as solid black

10
The area of the polygons of the hummocks was summarized for each 500-m bin of the distance from the source summit to be compared with the previously reported binned average of hummock areas along the distance ( Fig. 10; Yoshida et al., 2012). In Fig. 10, the mean area of all hummocks identified for each 500-m distance is shown as white circles, whereas the mean area for all the hummocks is shown as a black horizontal bar. The limited distribution and concentration of hummocks 15 only in the area 11,000-13,000 m from the summit and the downstream increase in hummock area suggest that the distancesize relationship does not work for Mt. Erciyes DAD.

Volume estimation of DAD by the source area reconstruction
Using two different interpolation methods, linear and Bézier, the original surface of the edifice before the sector collapse were reconstructed (Figs. 11,12). The particular difference between the linear and Bézier methods is the maximum summit elevation of the reconstructed surface. The pre-avalanche edifice summit generated by linear interpolation is at 4100 10 m elevation. This is much higher than the present summit of 3917 m (Fig. 11c). The summit generated by Bézier method do not exceed the present summit elevation (Fig. 11d). In both cases, the summit of the reconstructed edifice is at 1,000 m southeast of the current summit (Figs. 11a, b, 12b, c).
The volume of the missing edifice by the sector collapse is calculated as the sum of the elevation differences between the reconstructed surface and the present landform: 12.75 ´ 10 8 m 3 by linear method, and 10.67 ´ 10 8 m 3 by Bézier method. 15 If we assume the preexisting fluvial dissection of the edifice or a crater on the top (Aramaki, 1963), this reconstructed volume may be reduced by approximately 10% (Aramaki, 1963;Yonechi et al., 1988;Yoshida and Sugai, 2007a). The estimated source volume of the sector collapse is then given as 9.60-11.48 ´ 10 8 m 3 . Since the plan area of the sector collapse source area is 5.2 ´ 10 9 m 2 , the mean depth of the missing edifice is obtained to be 186-223 m. Although the accuracy of this estimation is hard to be quantified due to the lack of any morphological evidence of the sector before the 20 collapse, we consider these values as a feasible, order-of-magnitude estimate of the missing volume and depth.

Characteristics of the debris avalanche inferred from the volume and thickness
Because of the expansion of the original mass and/or sediment entrainment during transport, the volume of DAD is more than that of the missing sector for 25-30% (Siebert, 1984;Yoshida and Sugai, 2007a). Accounting for the estimated source volume of 9.6-11.5 ´ 10 8 m 3 , the potential volume of the DAD would become 12-15 ´ 10 8 m 3 . This estimated 15 volume is relatively larger than the average volume (ca. 5 ´ 10 8 m 3 ) of reported DADs in the Holocene and Pleistocene (Siebert, 1984), but falls within the range of 2-260 ´ 10 8 m 3 (Siebert, 1984). If compared with the known cases, the estimated volume is comparable to the debris avalanche in the historical time period at Mt. Iriga in the Philippines, whose volume is known as 15 ´ 10 8 m 3 (Aguila et al., 1986;Paguican et al., 2012), and more than half of that of the 25-28 ´ 10 8 m 3 Mt. St. Helens case of 1980 (Voight et al., 1983;Siebert et al., 1987;Glicken, 1996). Also, the estimated volume of Erciyes 20 is approximately twice or three times of the well-described case of the 4.9 ´ 10 8 m 3 volume avalanche on northern flank of Mt. Bandai, Japan in 1888 that was triggered by a phreatic eruption (Yonechi et al., 1988;Yonechi and Chiba, 1989;Yoshida, 2012), or the non-eruptive case of Mt. Unzen-Mayuyama, Japan in 1792 (4.4 ´ 10 8 m 3 ) (Inoue, 1999;Takarada and Melendez, 2006).
Although the exact extent of the DAD is difficult to confirm in this study, the rough estimation of the potential extent 25 of the DAD in this study has an area of at least 20 ´ 10 9 m 2 , which is apparently larger than that (14 ´ 10 9 m 2 ) previously mapped by Şen et al. (2003). Dividing the estimated source volume of the DAD, 12-15 ´ 10 8 m 3 , by the potential extent of the depositional area of 20 ´ 10 9 m 2 , the mean thickness of the DAD is ~60-75 m. This thickness of DAD is much larger than those reported in the previous studies: e.g., 25-30 m for Mt. Asama (Yoshida and Sugai, 2010), 45 m for Mt. St. Helens (Voight et al., 1983). The area and volume of the DADs across the world summarized by Siebert (1984) also gives an average thickness of 24 m with a range of 3-58 m, Mt. Erciyes for DAD thickness is out of the range of the reported ones.
This supports that the extent of the depositional area of the DAD of Mt. Erciyes can be much larger than 20 ´ 10 9 m 2 , potentially twice to three-times of the identified area. Based on these analyses, it is reasonable to say that Erciyes debris 5 avalanche deposits could have flown out of the gorge, farther downstream and spread over the relatively flat terrain 18,000 m from the summit (E in Fig. 1).

Debris avalanche and formation of hummocks
Based on the plots in Fig. 10, the distance-size relationship for Erciyes may be comparable to the case of the Onuma avalanche on Komagatake volcano, the Ofujisan avalanche on Nasu volcano, or the Gotenba debris avalanche on Mt. Fuji 10 . However, unlike the other cases, the Erciyes hummocks are densely found only in a limited area (11-13,000 m from the source). The distance-size relationship of hummocks cannot be directly compared to the other cases (Siebert, 1984;Yoshida et al., 2012) in which hummocks are more evenly distributed without significant topographic barriers.
In turn, the absence of hummocks in both upstream and downstream zones indicate a different pattern of debris 15 avalanche flow for this particular case of Mt. Erciyes. In general, hummocks in DADs are primarily formed due to the initial stage of sliding of the original sector of the edifice, with some blocks preserving the original structure that appears as secondary hummocks by floating on the surface of DADs in the later stage (Paguican et al., 2014). In the upstream zone of the present hummocks of Mt Erciyes, the absence of such primary hummocks in the 0-11,000 m area from the source may indicate either that hummocks are not formed due to the high fracturing of DADs, or that hummocks are formed but buried 20 by subsequent sediment covering over the DAD.
On one hand, such high fracturing of the Erciyes DAD is feasible because the debris avalanche could have had low viscosity with high water content with a sufficient amount of ice of glaciers on the original mountain in the glacial period, likely under the cool and wet climate in the late Pleistocene (Sarıkaya et al., 2009;Bayer Altin et al., 2015). It should also be noted that the debris avalanche flow was affected by the pre-existing topography of the former caldera (Fig. 1). The debris 25 avalanche is supposed to have initially flown to the east, and after blocked by the former caldera wall located in the east, the debris avalanche changed its direction to the north (Fig. 1). If the debris avalanche is well fractured, the flow could have shallowly spread over the flat area within the caldera without forming primary hummocks.
On the other hand, even if some small primary hummocks were formed at the initial stage, they could have been hidden by post-collapse modifications and surficial sediment deposition in this upstream area of the DAD (Yoshida and Sugai, 30 2007a), as indicated by the presence of convex-up cross profiles in elevation of 2800-2300 m (Fig. 9, or around the letter B in Fig. 1). Although no significant eruptive activity producing thick tephra or pyroclastic flows has been recognized in the volcanic stratigraphy (Şen et al., 2003), post-collapse eruptive activities might have occurred at a more local scale to cover the upper portion of the DAD. Loose materials including volcanic colluvial or alluvial deposits on the surrounding slopes could also have been reworked to contribute to cover the DAD. However, the thickness of such surficial sediment remains to 35 be examined.
In the downstream zone of the present hummocks, the stream gradient is high, and the valley width is narrow forming a knickzone (Figs. 8,9). This topographic feature could have existed before the sector collapse because this location corresponds to the intersection of the north-oriented valley, which seems to follow the mapped fault line as discussed later, and the former caldera wall of the pre-Erciyes volcanic activity (Fig. 1). The debris avalanche went over the wall at this pre-terminal portion of the caldera floor due to the confinement of the sliding materials. The flow could then have acquired higher speed when flowing through the steepened knickzone. The bimodal distribution of the displacement angle of hummocks would support this hypothesis (Fig. 7b). As suggested by the displacement angles, some hummocks with displaced orientations indicate a partial compressional regime of the debris avalanche: The sliding materials are confined, and hummocks are emplaced with their major axis perpendicular to the flow direction (Paguican et al., 2014;Yoshida, 5 2014). Whereas, the other hummocks without large displacements indicate an extensional regime of the debris avalanche due to the rapid northward flow through the valley: the stretch of the materials forced some of the hummocks aligned to the flow direction (Paguican et al., 2014;Yoshida, 2014). Such a change in the flow regime of debris avalanche can also be supported by progressive changes in flow materials (Clavero et al., 2002). In this case, since the debris avalanche flowed in the caldera, the increase in water content of the avalanche bottom is feasible if there was a caldera lake or rivers prior to the 10 collapse. The lowering of friction at the bottom flow could have enhanced the rapid passage through the valley, while the avalanche surface remained dry enough to emplace the hummocks.
The debris avalanche materials which went over the gorge could have been further deformed and shallowly spread over the relatively flat area after reaching the downstream of the knickzone (Figs. 8, 9). Hummocks may no longer be formed in such highly-fractured materials of the debris avalanche. No data on the detailed geological structure in the downstream areas 15 is available, and it is hard to find the extent of the DAD. Further careful investigations regarding the remnant of this DAD would be required to clarify the actual impact of the debris avalanche in this region. Future studies should therefore include field-based verification of the deposits such as borehole core drilling. Moreover, such a complex case of debris avalanche with both topographic barriers (caldera wall) and passage (valley) has not frequently been reported in the previous studies, except for a few cases (Francis et al., 1985;Glicken, 1996;Yoshida and Sugai, 2007b). The case of Erciyes remarks the 20 necessity of considering complex topographic constraints for the prediction of flow dynamics of debris avalanche.

Potential effects of fault lines on the debris avalanche
The presence of the strike-slip Erciyes Fault (Fig. 1) may have influenced the generation and emplacement of the DAD.
Although the exact age and intensity of the fault activity are unknown, and the presence of the fault itself also needs further careful assessments (Okumura et al., 2016), the regional stress field related to this fault group could have affected the 25 occurrence and the direction of the sector collapse of Mt. Erciyes. In general, sector collapses often occur in the direction perpendicular to the major horizontal compression by faulting beneath the volcano (Moriya, 1980;Siebert, 1984;Vidal and Merle, 2000;Tibaldi et al. 2008), or in parallel to the fault directions (Lagmay et al., 2000;Yokoyama and Nakagaki, 2003;Wooler et al., 2009). In Mt. Erciyes, the major horizontal compression by the strike-slip faults in the north-south direction could have caused the sector collapse toward the east. Furthermore, the northward valley seems to have been formed along 30 the fault line cutting the caldera wall. The debris avalanche, which has flown to the north being blocked by the pre-existed caldera wall, could have spread far downstream by overpassing the former caldera wall along the pre-existed valley. In such a case, the existence of local faults has dual significance, in the occurrence of sector collapse itself, and in the directionality of the debris avalanche. Although further examination of this issue are out of the scope of this study, detailed surveys of active faults and stress field in this region is also highly significant in predicting the future sector collapse of the volcano 35 (Tibaldi et al. 2008;Wooler et al., 2009).

Conclusions
In this study, we utilized the RPAS-based SfM-MVS photogrammetry to map the topography of Mt. Erciyes DAD. Detailed outlines of hummocks were extracted from the high-definition land surface data. Despite the limited availability of highdefinition aerial or satellite images, rapid on-site acquisition by RPAS-based SfM-MVS photogrammetry enabled to obtain new insights regarding the formation of hummocks and DAD at Mt. Erciyes. The volume of the sector collapse was estimated by reconstructing the original topography of the source area. Although the precise extent of the DAD is difficult to trace, the estimated volume of ~12-15 ´ 10 8 m 3 suggests that considerable amount of sediment could have spread downstream over the observed limit of the DAD. Morphological analysis of the extracted hummocks suggests that the geometry of the hummocks in this study area is characterized by pre-existing topographic constraints, which significantly 5 controlled dynamics of debris avalanche emplacement and formation of hummocks. In particular, former caldera walls had a strong effect on the flow direction of the debris avalanche as it confined the flow. The presence of strike-slip fault can also have considerable effects on both the occurrence of the sector collapse and the flow path of the debris avalanche. Such factors as pre-existing topographic and tectonic characteristics should carefully be considered for the hazard estimation of sector collapses, which can occur in many places around the world. 10 The study area is one of the potentially hazardous areas in the East Anatolian Fault zone with densely distributed volcanoes and faults (Koçyiğit and Erol, 2001;Korkmaz, 2009;Okumura et al., 2017). Such activities of active faults (earthquakes) and volcanoes themselves are hazardous, and they can trigger sector collapses and succeeding debris avalanche in the area. The detailed map of the DAD should be incorporated into a regional geomorphological map to further examine other geomorphological features including fluvial valleys, terraces, lakes and active faults (Erol, 1999). In particular, 15 the insights inferred from this study proposes the necessity of the assessment of potential debris avalanche paths. As noted, the studied debris avalanche had been significantly affected by topographic constraints, and the flow path was relatively limited following the pre-existing topographic slot across the caldera. This means that the hazardous areas affected by debris avalanche are rather limited, regardless of the direction of the initial collapse of the edifice. Further investigations such as numerical modeling of the debris avalanche flow based on the present topography are expected to be carried out. At least, 20 several scenarios of the potential sector collapse of the present edifice should be proposed in the future studies.
In addition, although the age of the sector collapse is relatively old (possibly 20-80 ka; Şen et al., 2003;Sarıkaya et al., 2009), subsequent sediment supply from the DAD could also have affected ancient human activities in the Holocene period in the downstream basin areas. Since a lot of archaeological sites are found and investigated in the area (Kontani et al., 2014;Yener et al., 2015), the assessments of the relationships between DAD and surrounding palaeoenvironments including 25 human activities will provide further insights for the potential disaster and its mitigation in the study area, such as floods, volcanism, landslides and fault-induced earthquakes.