Interactions between the accumulation of sediment storage and debris flow characteristics in a debris-flow initiation zone , Ohya landslide body , Japan

Debris flows often occur in steep mountain channels, and can be extremely hazardous as a result of their destructive power, long travel distance, and high velocity. However, their characteristics in the initiation zones, which could possibly be affected by temporal changes in the channel topography associated with sediment supply from hillslopes and the evacuation of sediment by debris flows, are poorly understood. Thus, we studied the interaction between the flow characteristics and the 15 topography in an initiation zone of debris flow at the Ohya landslide body in Japan using a variety of methods, including a physical analysis, a periodical terrestrial laser scanning (TLS) survey, and field monitoring. Our study clarified that both partly and fully saturated debris flows are important hydrogeomorphic processes in the initiation zones of debris flow because of the steep terrain. The predominant type of flow varied temporally and was affected by the volume of storage and rainfall patterns. The small-scale channel gradient (on the order of meters) formed by debris flows differed between the predominant flow types 20 during debris flow events. The relationship between flow type and the slope gradient could be explained by a simple analysis of the static force at the bottom of the sediment mass.


Introduction
Debris flows often occur in steep mountain channels and can be extremely hazardous as a result of their destructive power, long travel distance, and high velocity (Lin et al., 2002;Cui et al., 2011).To lower the hazard of debris flow, field monitoring 25 have been conducted in many torrents in the world including Switzerland (McArdell et al., 2007;Berger et al., 2011b), Italy (Arattano, 1999;March et al., 2002;Arattano et al., 2012), the United States (McCoy et al., 2010;Kean et al., 2013), and China (Zhang, 1993;Hu et al., 2011).Many of these monitoring have been undertaken in the transportation zones of the debris flows, with only a few observations undertaken in their initiation zones, where the unstable sediment start to move (Berti et al., 1999; Nat.Hazards Earth Syst.Sci. Discuss., doi:10.5194/nhess-2017-20, 2017 Manuscript under review for journal Nat.Hazards Earth Syst.Sci.Published: 16 January 2017 c Author(s) 2017.CC-BY 3.0 License.McCoy et al., 2012;Kean et al., 2013).Understanding of the decries-flow characteristics in the initiation zone is needed to estimation of debris-flow volume and timing in down reaches required for improvement of the debris-flow warning system.Debris flows can be classified into various types according to their flow dynamics, solid fractions, and material types (Takahashi, 1991;Coussot and Meunier, 1996;Hunger, 2005).Multiple flows types appear even in the same torrent (Imaizumi et al., 2005;Okano et al., 2012).However, in situ classification of debris flow type is mainly based on monitoring results 5 obtained in the transportation zone.Field monitoring conducted at multiple sites along a debris flow torrent have revealed that flow characteristics (e.g., solid fraction and boulder size) and discharge change as the flow migrates downstream and is affected by erosion and deposition (Takahashi, 1991, Berger et al., 2011b;Arattano et al., 2012).Thus, flow characteristics in the transportation zone possibly differ from those in the initiation zone.In the initiation zone, flows containing an unsaturated layer have also been observed during field monitoring (Imaizumi et al., 2005;Imaizumi et al., 2016b).Such information is 10 needed to explain the sequence of debris flow processes from initiation in the headwaters to termination at the debris flow fan.
The topography along debris flow torrents reflects the characteristics of a debris flow event (Whipple and Dunne, 1992;Coussot and Meunier, 1996;Imaizumi et al., 2016a).The curvature of the terrain can be used to determine the relationship between shear stress and shear strength (Staley et al., 2006).Many observations of the relationships between topography and flow characteristics have been made in the lower part of the debris flow torrents, while such findings are limited in the debris-15 flow initiation zone.In many cases, the channel gradient in the initiation zone is greater than 20° (VanDine 1985;Pareschi et al. 2002;McCoy et al., 2013;Hürlimann et al., 2015).In terms of debris flows caused by the transportation of loose sediment on the floor of a valley, both the volume and grain size of debris flow material (e.g., channel deposits and talus slope) change over time in association with the sediment supply from hillslopes, as well as the evacuation of sediment by debris flows and fluvial processes (Bovis and Jakob, 1999, Imaizumi et al., 2006, Berger et al., 2011a).Although some previous studies have 20 investigated the relationship between characteristics of the debris flow material and initiation condition of a debris flow (Bovis and Jakob, 1999;Jakob et al., 2005;Schlunegger et al., 2009;Chen et al., 2012), only a few have considered the relationship between flow material and flow characteristics (Kean et al., 2013;Imaizumi et al., 2016b).The difficulty in monitoring debris flows in steep and dangerous initiation zones has prevented the collection of the field data needed to clarify the relationship between flow characteristics and flow material.25 In the debris-flow initiation zone at the Ichinosawa catchment within the Ohya landslide, Japan, field monitoring has been undertaken since 1998 (Imaizumi et al., 2005;Imaizumi et al., 2006).This site is suitable for monitoring because of the high debris-flow frequency (about three or four events per year) that occur due to the mobilization of storage (i.e., talus cone and channel deposits) around the channel.In addition, detailed topographic measurements by terrestrial laser scanning (TLS) have been undertaken periodically since 2011 (Hayakawa et al., 2016).30 In this study, we investigated the relationship between the slope gradient and type of sediment transport based on a simple analysis of the static force.Then we considered the relationship between the accumulation conditions of storage (i.e., slope gradient and volume of storage) and debris-flow characteristics based on field monitoring results in the Ohya landslide.Our specific objectives were to explain the relationship between topography and sediment transport type by simple models, clarify Nat. Hazards Earth Syst. Sci. Discuss., doi:10.5194/nhess-2017-20, 2017 Manuscript under review for journal Nat.Hazards Earth Syst.Sci.Published: 16 January 2017 c Author(s) 2017.CC-BY 3.0 License.temporal changes in the channel topography of the debris-flow initiation zone due to the occurrence of debris flows, and clarify the interaction between the characteristics of sediment storage (i.e., volume and slope gradient) and debris-flow type.

Slope gradient and type of sediment transport
In this section, we describe our analysis of the balance of static force at the bottom of a sediment mass to assess the relationship between slope gradient and the type of sediment transport.Shear stress at the bottom of a sediment mass needs to exceed shear 5 strength for the initial movement of a stable mass (Takahashi, 1991;Imaizumi et al., 2016c).Similarly, shear stress needs to exceed shear strength at the bottom of a traveling sediment mass for the continuity of travel (Takahashi, 1991;Watanabe, 1994).Under the assumption that cohesion is negligible, shear stress τ and shear strength at the bottom of sediment mass can be given as follows: where ηw is the ratio of the depth of the saturated zone ( ) to the depth of the sediment mass ( ), hw is the height of the water table, z1 and h are the height at the bottom and surface of the sediment mass, respectively, n is the porosity of sediment, is the force of gravity acting on a unit volume of sediment, is the force of gravity acting on a unit volume of water (or interstitial water for debris flow), S is the degree of saturation in the unsaturated zone,  is the slope gradient, and ϕ 15 is the effective internal angle of friction (Fig. 1).Note that  is the slope gradient of the potential sliding surface when we consider the initial movement of the stable mass, whereas it is the gradient of the surface topography when we consider the migration of the traveling sediment mass.The n in the moving sediment mass is similar to that of stable sediment when dispersion of the sediment particles is not significant, while n in the moving sediment mass greatly exceeds that of the stable sediment when the particle dispersion associated with collision among the particles is significant (e.g., fully saturated debris 20 flow).Based on Eqs.(1) and (2), the critical condition for the movement of sediment mass ( ) can be given as follows (Imaizumi et al., 2016c): ( 3 ) Based on the ηw, three typical sediment transport types were considered: fully unsaturated (ηw = 0), partly saturated (0 < ηw < 25 1), and fully saturated (ηw = 1).In case of ηw = 0, Eq. ( 3) is expressed as: If the slope gradient tan exceeds tanϕ, the sediment mass can move without any saturation (Fig. 1a).In other words, Eq. ( 4) expresses the lowest boundary of the slope gradient for the movement of the fully unsaturated sediment mass (hereafter referred to as 1).There are several types of sediment transport in fully unsaturated conditions, such as rockfall, dry granular flow, and 30 dry ravel, which is the gravitational transport of ground surface materials by bouncing, rolling, and sliding (Carson, 1977; Nat . Hazards Earth Syst. Sci. Discuss., doi:10.5194/nhess-2017-20, 2017 Manuscript under review for journal Nat.Hazards Earth Syst.Sci.Published: 16 January 2017 c Author(s) 2017.CC-BY 3.0 License.Dorren, 2003;Gabet, 2003).Although physical mechanisms between individual particles (i.e., rockfall and dry ravel) and flows of grains (i.e., dry granular flow) are different in terms of the interactions among particles, the slope gradient of talus slopes, which are formed by fully unsaturated sediment transport processes, are usually similar or slightly smaller than ϕ regardless of the transport type (Kirkby and Statham, 1975;Carson, 1977;Mangeney et al., 2007).Pyroclastic flow, which is a fluid composed of a mixture of air and particles, is the exception in the above relationship between  and ϕ.Pyroclastic flow 5 sometimes reaches terrain much smaller than ϕ (Yamashita and Miyamoto, 1993;Takahashi and Tsujimoto, 2000).The pyroclastic flow is an important sediment transport process in areas with many fine particles (i.e., volcanic areas).In this study, we did not consider pyroclastic flow because we focused on sediment transport processes in areas in which gravel, cobbles, and boulders dominated the debris flow material.
The relationship between slope gradient and the volumetric sediment concentration (1 -n) in a saturated sediment mass is 10 given by substituting 1 into ηw in Eq. (3): tan 1 1 tan .
( 5 ) Note that n here becomes larger than the n of storage when sediment particles start to disperse by collision with other particles.
Equation ( 5) expresses the relationship between the solid concentration in the steady-state flow (called equilibrium 15 concentration) and the slope gradient (Takahashi, 1991;Egashira et al., 2001, Takahashi, 2014).A similar trend as in Eq. ( 5) was also found in the in situ measurements of Navratil et al. (2013).By substituting the porosity of the storage into n, Eq. ( 5) expresses the slope gradient needed for the entrainment of a fully saturated sediment mass, with no dispersion of particles (hereafter referred to as 2).When the slope gradient of the terrain ranges between 1 and 2, the transportation of a partly saturated sediment mass occurs (Fig. 1b) (Watanabe, 1994;Imaizumi et al., 2005;Imaizumi et al., 2016c).Equation (5) just 20 considers shear strength and shear stress at the bottom of the sediment mass, but does not consider the conditions of the fluid of the sediment mass.Thus, not only the debris flow, but also the sliding of the sediment mass without any fluid and plug flow, of which the upper layer in the sediment mass is not fluid, also satisfy Eq. ( 5).Nevertheless, Eq. ( 5) is a necessary condition for any type of sediment movement with partly saturated conditions.Partly saturated sediment transport has also been observed in channels gentler than 2 (i.e., channel gradient < 10°) (McArdell et al., 2007;McCoy et al., 2010;Okano et al., 2012).

25
However, the appearance of partly saturated sediment transport in such gentler channels is generally limited at the front of a surge.In torrents gentler than 2, the existence of surface flow over debris flow material is theoretically required for the initiation of debris flow (Fig. 1c) (Takahashi, 1991;Imaizumi et al., 2016c).Once sediment starts to move, it spreads throughout the flow (Fig. 1d).Thus the volumetric solid concentration is lower than that in sediment storage (1 -n).
The explanations above are also applicable to the relationship between the amount of water in the sediment mass from the 30 upper channel reaches (and slopes) and the slope gradient of the terrain formed by that sediment mass (Takahashi, 1991;2014).
If the amount of water in the sediment mass from the upper channel reaches is constant, the slope gradient of the terrain Nat. Hazards Earth Syst. Sci. Discuss., doi:10.5194/nhess-2017-20, 2017 Manuscript under review for journal Nat.Hazards Earth Syst.Sci.Published: 16 January 2017 c Author(s) 2017.CC-BY 3.0 License.
approaches the gradient given by the substitution of ηw, n, and S of the sediment mass into Eq.(3) as a result of deposition and erosion.
In this study, we call any partly and fully saturated sediment transport processes as partly and fully saturated debris flows, respectively.Although some sediment transport processes at our monitoring site were not typical debris flows, such as those composed of a mixture of sediment and water (Takahashi, 1991;Coussot and Meunier, 1996), we believe the processes at our 5 study site directly or indirectly contributed to the initiation of debris flow.We conducted field monitoring within the Ohya landslide in the Southern Japanese Alps (Fig. 2).The Ohya landslide has an estimated total volume of 120 million m 3 , and was initiated during an earthquake in A.D. 1707 (Tsuchiya and Imaizumi, 2010).
The climate at the site is characterized by a high annual precipitation (about 3,400 mm) (Imaizumi et al., 2005).Heavy rainfall (i.e., total rainfall > 100 mm) occur during the rainy season from June to July and the autumn typhoon season (from August to 15 October).The geological unit is Tertiary strata, which is composed of well-jointed sandstone and highly fractured shale.
Unstable sediments have been supplied from outcrops into the channels in the landslide scar and have affected the initiation of debris flows since the original failure.Almost all of debris flows in the Ohya landslide occur in the upper Ichinosawa catchment (Imaizumi et al., 2005).The total length of the channel is 650 m and the south-facing catchment (1,450 -1,905 m a.s.l) has an area of 0.22 km 2 .Most of the basin is characterized by high and steep slopes (40 -65°).Seventy percent of the slope is scree and outcrop, whereas the remaining 30% is covered with forest, shrubs, and tussocks.Anthropogenic influences are absent in the catchment because of the harsh environmental conditions.5 Unconsolidated debris, ranging from sand particles to boulders (Imaizumi et al., 2016c), is located in the channel bed and talus cones, and is the source of debris flow material (Imaizumi et al., 2006).The thickness of debris deposits, including large boulders (> 1 m), exceeds several meters in some sections.Freeze-thaw that promotes dry ravel and rockfalls are the predominant sediment infilling of the channels (Imaizumi et al., 2006).Most of the channel was covered by channel deposits when a large volume of storage accumulated in the bottom of the valley, while the channel bed is alternatively composed of 10 bedrock and sediments when the storage volume is low.Volume of storage displays seasonal changes caused by sediment supply from hillslopes in winter and early spring, and the evacuation of storage due to the occurrence of debris flow in summer and autumn (Imaizumi et al., 2006).The stored sediment has never been completely eroded by debris flows.Changes in the volume of storage at longer time scales (several years) occurs by the timing of large debris flows with a volume > 15,000 m 3 that drastically decrease volume of storage (Imaizumi et al., 2016c).15 Nat. Hazards Earth Syst. Sci. Discuss., doi:10.5194/nhess-2017-20, 2017 Manuscript under review for journal Nat.Hazards Earth Syst.Sci.Published: 16 January 2017 c Author(s) 2017.CC-BY 3.0 License.

Debris-flow Monitoring 5
The monitoring system was installed in the study site in spring 1998 and consisted of a rain gauge, ultrasonic sensors, water pressure sensors, and a video camera (Imaizumi et al., 2005).The video camera, which provided motion images of debris flows, was initially monitored at P4, then moved to P2 in 2000 (Fig. 2a).During the period of 1998 to 2001, the video camera was programmed to take 0.75 s clips of video every 5 min.In April 2001, this interval was shortened to 3 min to capture flow characteristics in more detail.In addition, continuous monitoring cameras, of which recording was initiated by wire motion 10 sensors installed at several cross-sections in the channel, were initially installed at P1 in 2003, then at P3 and P2 in 2004 and 2005, respectively.The video camera system sometimes failed to capture debris flows because of their mechanical problems due to harsh site conditions.We identified timing of such debris flows by field surveys and water pressure sensors.Flow depth  Takahashi (1977Takahashi ( , 2014)).Changes in cross-sectional area of flow were calculated from changes in flow depth and cross-section measurement of the channel topography.Discharge of debris flows were estimated 5 from the cross-section area multi-plied by mean velocity of all layers.Imaizumi et al. (2006) classified the flow phases during debris flow events into two primary types: flows that made of mainly muddy water and those made of mainly cobbles and boulders.The former flows are turbulent and are characterized by black surfaces due to high concentrations of silty sediment sourced from shale in the interstitial water (Fig. 3a).Because the matrix of boulders is filled with interstitial water, this flow is considered fully saturated (Fig. 1d).In contrast, muddy water is 10 not identified in the matrix of the flow surface of the second type of flow (Fig. 3b).Therefore, this type of flow is considered partly saturated (Fig. 1b).We visually identified temporal changes in the flow type during debris flow events from video images based on the existence of interstitial water on the flow surface.Time-lapse cameras (TLCs; GardenWatchCam, Brino, Taipei City, Taiwan) were installed around P1 and P2 in April, 2013, as a backup for video camera monitoring.The number of cameras (1 to 6) differed among the various measurement periods.15 The intervals used for capturing images were also different among the measurement periods and for the various cameras, with a range from 1 s to 10 min.
To identify the arrival of debris flow, a semiconductor type water pressure sensors, which monitored hydrostatic pressures up to 49 kPa with an accuracy of ± 3%, were placed in holes dug in the bedrock of the channel bed.The logging interval was set at 1 min.Because water pressure sensors were sometimes washed away by debris flow, ultrasonic sensors with a measuring 20 range from 120 to 600 cm (accuracy of ± 0.4%) were installed to monitor the surface height of debris flows at 1 min intervals as the backup of water pressure sensors.Water pressure and flow height monitored by these instruments showed intense and abrupt changes during the passage of debris flows, whereas such changes were absent during rainfall-runoff events without occurrence of debris flow.Therefore, we used water pressure and flow depth to identify the occurrence of debris flows.

Topographic Data Obtained by TLS
To clarify temporal changes in the micro channel bed topography associated with occurrence of debris flows, a TLS unit (GLS-1500 Topcon Co., Tokyo, Japan) was used to measure the topography of valley axis (Fig. 2) (Hayakawa et al., 2016).The maximum measurable distance of the TLS unit is 500 m (for target objects with a 90% reflectance), with accuracies of 6" for angle and 4 mm for distance (at 150 m range).From November 2011, field measurements were undertaken in spring, summer, 10 and autumn for 4 years (Table 1).For each measurement, the scanner was set at two positions.One was at the downstream side of the target area (P1 in Fig. 2), where the land had been relatively stable for many years.The other at the upstream side of the target area (around point D in Fig. 2).The location of this scan position varied because it was in the valley bottom where the topography changed significantly due to debris flows.The point clouds measured from different scan positions were registered using at least five reference targets placed between the two scan positions.The two point clouds were registered by the target matching method using these reference targets, with accuracies of 0.5-6 mm.The geographic coordinates (Japan Plane Rectangular CS VIII, EPSG: 2450) of two targets, selected as georeference points, were measured with global navigation satellite system (GNSS) receivers (GeoXH 6000, Trimble, Sunnyvale, CA, USA or GRS-1, Topcon Co., Japan; accurate to a range of 10-63 mm in XY or Z).The baseline solution was calculated using data from GNSS base stations of GEONET, the Japanese GNSS network operated by the Geospatial Authority of Japan.The overall uncertainties of the point clouds, including scanning, registration, and georeferencing by GNSS, were considered in the order of centimeters to a decimeter (Hayakawa et al., 2016).*1 Date of the last debris flow event are not listed when debris flow had not occurred in the year , because the topography was possibly affected by the winter sediment supply rather than the last debris flow in the previous year.
After manually eliminating noise and unnecessary points, the point clouds obtained by the TLS measurement were converted into digital elevation models (DEMs) using a simple triangulated irregular network (TIN) interpolation, because vegetation cover was absent in the site.The resolution of the DEMs was set at 10 cm, which sufficiently covers the average point density of the point clouds.Areas with sparse point clouds (approximately <100 point m -2 ), vegetated areas, and areas affected by the combination of various processes (e.g., talus slopes with the lower part eroded by stream flow) were excluded from the analyses.
The extent of typical geomorphic units in the TLS survey area, including three rock slopes, three talus slopes, and a channel around the monitoring plot P1, was mapped by field surveys to calculate the slope gradient of geomorphic units with various grid sizes (Fig. 4).The mapping was conducted at the same time as each TLS survey because the area changed over time due to the sediment supply from outcrops and transportation of sediment by debris flows.

Estimation of debris-flow volumes
We estimated the volume of storage from photographs taken periodically at sites P1 and P4 (Fig. 2) (Imaizumi et al., 2016a).5 We could not use data obtained by periodical TLS, because the target area of the TLS was a fraction of the storage zone.
Photographs from site P4 cover the entire study site, whereas those from P1 focus on channel deposits at the bottom of the incised main channel, which is shaded in photographs from P4.The area covered by channel deposits and talus slopes in each photograph periods was mapped on GIS based on these photographs.The bedrock topography in the upper Ichinosawa catchment was estimated from terrains by airborne laser scanning (ALS) in the periods when sediment storage was almost 10 absent (i.e., in 2011 and 2012).Thirty-three cross-sectional areas of the storage with spacing of 25 m along the channels (24 cross-sections along the main channel and 9 cross-sections along a tributary) were calculated from the bedrock topography and the location of both ends of the storage along the cross-sections on the GIS storage map under the assumption that surface topography of the storage was an inclined line connecting both ends of the storage.Total volume of the storage was calculated by sum of the cross-sectional area multiplied by the spacing of cross-sections (25 m).The root mean squared error (RMSE) of 15 the procedure, which was obtained from comparison of the volumes of storage by the procedure and that by subtraction of ALS DEMs by estimated bedrock topography, was 5361 m 3 (Imaizumi et al., 2016a).This RMSE is larger than the volume of small debris flows (<2000 m 3 ) but below the sediment supply volume in each year (>10,000 m 3 ).

Debris flow monitoring
From 1998 to 2015, 59 debris flows, which went through or terminated in the section between sites P1 to P3, were observed by video cameras, field surveys, and water pressure sensors in the upper Ichinosawa catchment.In addition, the occurrence of some other debris flows, which terminated above site P1, were found by periodic photography and field surveys.We analyzed 5 video images of seven flows that were clearly captured by video cameras.In addition, we analyzed photographs of one debris flow taken by TLCs with an interval of 1 s (Table 2).We failed to obtain clear video images of other debris flows because of darkness at night and fog.Most of the rainfall events triggering debris flows in the study site can be classified into two groups: long-lasting rainfall events with high total rainfall, caused by typhoons and stationary fronts (rainfall duration >5 h and total rainfall >50 mm), and short-duration convective rainfall events of high intensity (rainfall duration <5 h, total rainfall <50 mm; 10 Table 2).
As noted above, two types of flow have been observed in the Ohya landslide: flow mainly composed of cobbles and boulders (partly saturated flow) and flow mainly composed of muddy water (fully saturated flow).The duration of each flow phase differed between events.For example, a large part of the debris flow surges (88% in terms of total duration) was composed of partly saturated flow during a debris-flow event on 5 August 2008, while 90% in terms of the total duration of surges was 15 classified as fully saturated flow during a debris-flow event on 30 August 2004 (Fig. 5).Although the camera locations, which could potentially affect flow characteristics, differed among events, the proportional duration of the partly saturated debris flow in overall debris flow surges had a weak positive relationship with the volume of storage estimated from periodical photography (R 2 = 0.37, p-value = 0.06; Fig. 6).Rainfall was also a potential factor controlling flow type.The duration of fully saturated flow as a proportion of the whole debris flow surges was generally high during long-lasting rainfall events 20 associated with typhoons and stationary fronts, while the duration of partly saturated flow as a proportion of the whole surges was high during convective rainfall events of short duration (Fig. 6, Table 2).The p-value for the difference in the duration of the different flow types between the two rainfall patterns was below 0.01.

Slope gradient of geomorphic units
Mapping of the three geomorphic units (rock slopes, talus slopes, and the channel) in the TLS survey area showed that the 5 location of each unit was largely similar throughout the study periods, while the size of each unit changed due to sediment supply and transport processes.Some talus slopes completely disappeared in several periods because of erosion by debris flows.The roughness of the ground surface attributable to boulders, at a scale on the order of decimeters (Imaizumi et al., 2016c), was clearly visible on the slope gradient map, with a grid size of 0.1 m obtained by TLS (Fig. 7).
Step-pool-like topography with a height of several meters formed by large boulders (>1.0 m) and bedrock morphology was visible in the 10 channel on the slope gradient maps with a grid size of 1.0 m (Fig. 7).
Average and standard deviation of slope gradient in the three geomorphic units (rock slopes, talus slopes, and the channel) throughout the TLS monitoring period were calculated from DEMs with various grid size.Regardless of the grid size, the average slope gradient of rock slopes and the channel were highest and lowest among the three geomorphic units, respectively (Fig. 8a).The average slope gradient of each geomorphic unit becomes gentler with an increase in the grid size (Fig. 8a).15 However, the relationship between grid size and slope gradient was slightly different among geomorphic units.For a grid size of 1.0 m, a histogram of talus slopes was concentrated in a narrow range around the average value (i.e., >60% in the range between 34-40° for a grid size of 5.0 m; Fig. 8b).This slope angle is considered to represent 1, as reported in previous studies (Kirkby and Statham, 1975;Carson, 1977;Obanawa and Matsukura, 2008).Wide distribution of the slope gradient for a grid size < 1.0 m was mainly attributed to the roughness of the topography by cobbles and boulders.The standard deviation of the 20 slope gradient in the rock slope decreased with an increase in grid size when the grid size was smaller than the width of alternative strata in the study area (about 4.0 m; Imaizumi et al., 2015), and was almost constant when the grid size was larger than that.The standard deviation of the slope gradient in the channel also decreased with an increase in grid size when the grid size was smaller than 4.0 m, and was almost constant when the grid size was larger than 4.0 m.This inflection point of the trend (4.0 m) was probably affected by the largest boulder size in the channel (Imaizumi et al., 2006;Imaizumi et al., 2016c).

Longitudinal channel topography
To clarify the effects of debris flow type on the characteristics of the channel topography, temporal changes in the longitudinal channel profiles were analyzed in the section between points A and E (total length of 100 m; Fig. 2).The channel topography on 21 November 2012 was mainly formed by a debris flow on 30 September 2012, which was dominated by fully saturated flow (Fig. 8).In contrast, the channel topography on 23 August 2015, was mainly formed by a debris flow on 6 August 2015, 5 which was dominated by partly saturated flow.A small debris flow on 17 August 2015, which transported storage with a volume of <3,000 m 3 , also affected a part of the channel topography measured on 23 August 2015; however, the affected area was limited (mostly a width of <5 m) because of its small discharge.
As reported by Hayakawa et al. (2016), the channel bed deformation differed between channel sections (Fig. 9a).When we compared channel bed profiles on 21 November 2012, and 23 August 2015, the channel bed level was clearly different (>3 m 10 in depth) in the sections within 25 m and from 75 to 90 m upstream of point A. In contrast, there was almost no change in the longitudinal channel profile (<1 m) in the section within 30 m upstream of point C.Such differences in channel bed deformation could also be identified in the cross-sectional profile.The cross-sectional profile at points B and D was clearly different, while difference in the profile was not apparent at point C (Figs. 9e, 9f, 9g).
The longitudinal profile of the channel gradient differed among debris flow events.The distribution of the channel gradient 15 calculated for each 0.1 m section (the scale of a boulder) on November 21, 2012 (mainly formed by fully saturated flow), was clearly wider than that on 23 August 2015 (mainly formed by partly saturated flow; Fig. 9b, Table 3a).The p-value obtained by an F-test of the difference in the dispersion of the channel gradient between the two periods was <0.001.In contrast, the average slope gradient in the section between A and E was not significantly different between the two periods (23.1° and 25.4° for 21 November 2012, and 23 August 2015, respectively).20 The difference in the amplitude of the channel gradient calculated for each 1.0 and 5.0 m section was also clear (Figs.9c, 9d, Table 3a).The p-values obtained by an F-test of the difference in the dispersion of the channel gradient between the two periods was <0.001 and <0.05 for each 1.0 and 5.0 m section, respectively.Such temporal changes in the channel gradient was more evident in the sections with active channel bed deformation (from 0 to 25 and from 75 to 90 m from point A) compared to those without active channel bed deformation (from 25 to 75 and from 90 to 100 m from point A) (Tables 3b, 3c).Wide 25 distribution of the slope gradient for each 5.0 m section on 21 November 2012 was mainly attributed to the step-pool topography formed by large boulders and the bedrock step-pool, while that for each 1.0 m section was attributed to such smallscale topographies together with the particle size of large boulders.
Periodic photography and field surveys after eight debris flows, which were successfully monitored by video cameras and TLCs, indicated that small-scale topography after debris flows mainly consisting of fully saturated flow are rugged compared 30 to that after debris flows mainly consisting of partly saturated flow.This trend agrees with that observed by TLS surveys (Fig. 9, Table 3a).
The standard deviation of the channel gradient in the active channel bed deformation sections had a negative relationship with the volume of the storage (Fig. 10a).The p-values for the linear regressions with section lengths of 0.1, 1.0, and 5.0 m were <0.1, <0.05, and <0.05, respectively.As noted above, the theoretical channel gradient dividing partly and fully saturated debris flows (α2) can be obtained from 10 Eq. ( 5).Because the dispersion of boulders in the partly saturated debris flows was not evident in video images, porosity (ratio of liquid and vapor phases, to be exact) in the partly saturated flow may not be higher than that in the saturated channel deposits.
By assuming ϕ = 37.3° (average slope gradient of the talus slope for a grid size = 1.0 to 7.0 m), n = 0.3 (same as the porosity of channel deposits), γ = 26000, and γ = 9800, the α2 in the Ohya landslide was 22.2°.When we focused on the channel sections with active channel bed deformation (from 0 to 25 m and from 75 to 90 m upstream of point A), the proportional 15 amount of all channel sections in which the gradient was in the range between α2 and α1 (theoretical channel gradient for partly  3b).Changes in the channel gradient were less clear in sections without clear channel bed deformation compared to those with active channel bed deformation (Table 3c).
The proportional amount of all channel sections, in which the gradient was in the range between α2 and α1, throughout the 5 entire channel section with active channel bed deformation was higher when the volume of storage was large (Fig. 10b).The relationship was clear for all 0.1, 1.0, and 5.0 m section lengths (p-value < 0.05).In contrast, the proportional amount of all channel sections, in which the channel gradient was gentler than α2, throughout the entire channel section decreased with an increase in the volume of sediment storage.The gradient of the regression line for the section length of 0.1 m was gentler than that for the section lengths of 1.0 and 5.0 m. 10

Factors controlling the debris flow type
In contrast to debris-flow transportation zones in other torrents, in which partly saturated flow dominates just at the front of surges (McArdell et al., 2007;McCoy et al., 2010;Okano et al., 2012), our observations in the debris flow initiation zone of the upper Ichinosawa showed that partly saturated flow sometimes dominates the entire debris flow (Table 2).As with many 15 sections in the upper Ichinosawa catchment (Fig. 9), channel gradients in the debris flow initiation zones in many other torrents are greater than 22. 2° (e.g., VanDine 1985;McCoy et al., 2012), which is the theoretical channel gradient needed for a partly saturated debris flow in the Ichinosawa catchment.Although the α2 is different among torrents affected by soil parameters such as the internal angle of friction, debris flow initiation zones in other torrents possibly satisfy the conditions for the occurrence of partly saturated flow.20 The proportional duration of partly saturated flow in the overall surges had a positive relationship with the volume of storage (Fig. 6).Thus, the volume of storage was a factor controlling not only the initiation of the debris flow, particularly in the supply limited basin (Bovis and Jakob, 1999;Jakob et al., 2005), but also the debris flow characteristics.The water content in the storage, which is considered an important factor controlling flow characteristics (Takahashi, 1991;Hürlimann et al., 2015), may affect such relationships.When a large volume of storage is present in a basin, a large amount of water is needed for 25 saturation of the storage, which can be difficult to attain.In such cases, even if the sediment mass that initially moved in the channel was saturated, storage along the channel eroded by the debris flow would not be fully saturated when the channel gradient >α2.Consequently, partly saturated flow can easily dominate when a large volume of storage accumulates in the channel.Such erosion of unsaturated sediment has also been reported in other channels (Berger et al., 2011b;McCoy et al., 2012).In contrast, storage can be easily saturated by smaller amounts of rainfall when there is only a small volume of storage 30 in the basin.Thus, even when debris flows were not fully saturated in their initial stages, as they traveled downstream they easily became saturated by the water supply from tributaries as well as the erosion of saturated deposits.
Our monitoring results also revealed that the proportional duration of partly saturated flow in a debris flow event was low during long-lasting rainfall events (e.g., typhoons and stationary fronts), while the duration was high during short-lasting rainfall events (e.g., convective rainfall; Table 2, Fig. 6).The relationships between rainfall patterns and flow characteristics have also been reported in other debris flow basins (Okano et al., 2012;Kean et al., 2013;Hürlimann et al., 2014).Okano et al. (2012) obtained a similar trend to that found in Ichinosawa catchment based on observations in the transportation zone.The 5 front of a surge was completely saturated when rainfall intensity over a long duration (24 h) was high, while partly saturated flow occurred at the front of surges during short-term, intense rainfall events.Such a relationship between flow type and rainfall pattern can also be explained by the water content in the storage.During long-lasting rainfall events with high levels of cumulative rainfall, the amount of water in the storage can be high, resulting in a predominance of saturated flow.Water supply from tributaries during such long rainfall events also increases the water content in the debris flow (Okano et al., 2012).10 In contrast, the amount of water in the storage may be low during short-duration rainfall events with low levels of cumulative rainfall, resulting in the domination of partly saturated flow.

Slope gradient of geomorphic units
The peak of the histogram for each geomorphic unit was steep and high when the grid size was sufficiently large to eliminate the influence of small-scale roughness attributed to particle size (Fig. 8b).A large part of the histogram of the rock slopes was 15 steeper than 40° for a grid size of >4.0 m (e.g., 80% for a grid size of 5.0 m; Fig. 8b), indicating that fully unsaturated sediment transport, which occurs on slopes steeper than α1 (Eq.( 4)), is dominant on rock slopes.In contrast, most of the channel was gentler than α1 for a grid size of >4.0 m (e.g., >85% for a grid size of 5.0 m; Fig. 8b), indicating that fully and partly saturated debris flows are dominant in the channel.Such a trend was also apparent in the longitudinal profile of the channel (Fig. 9b).
Based on the statistics of the channel gradient for each 5.0 m channel section, the sum of the length ratio of the channel sections 20 in the theoretical range of fully and partly saturated debris flows was almost 1.0.
The longitudinal channel gradient changed with time and was affected by the predominant type of flow during the preceding debris flow events (Fig. 9).The proportional amount of channel sections between α1 and α2 (theoretical range of the channel gradient for partly saturated debris flow) among all active channel deformation sections was higher when a larger volume of storage was present in the catchment (Fig. 10b).Because partly saturated flow dominated when a large volume of storage was 25 present in the catchment (Fig. 6), partly saturated debris flow likely formed channel topographies with a theoretical channel gradient for the partly saturated flow (Fig. 11).Similarly, the length of the channel sections in the theoretical range for fully saturated flow (<α2) was high when small amounts of storage accumulated in the channel (Fig. 10b), during periods when fully saturated flow dominated (Fig. 6).Thus, fully saturated flow tended to form channel sections gentler than α2 (Fig. 11).The standard deviation of the channel gradient after a debris flow mainly consisting of fully saturated flow was higher than that 30 after a debris flow mainly consisting of partly saturated flow (Fig. 8, Table 3).In addition, the standard deviation of the channel gradient was high when small volumes of storage accumulated in the basin (Fig. 10a), implying that fully saturated flow tended to form a step-pool topography.As a result of field surveys and periodic topography, step-pool topography formed by large Nat. Hazards Earth Syst. Sci. Discuss., doi:10.5194/nhess-2017-20, 2017 Manuscript under review for journal Nat.Hazards Earth Syst.Sci.Published: 16 January 2017 c Author(s) 2017.CC-BY 3.0 License.boulders and the bedrock was generally seen after fully saturated flows.Since temporal changes in the channel gradient were not apparent in the larger-scale topography (i.e., channel gradient for entire sections with an active riverbed deformation with a length of 40 m), the decrease in the channel gradient due to fully saturated flow did not occur throughout the channel but was occasionally interrupted by steep sections (Fig. 11).Periodic photography from P1 showed that the proportional amount of coarse sediment generally increased after the passage of debris flows mainly consisting of saturated flow.The selective 5 removal of fine sediment, which was monitored in the Ichinosawa catchment (Imaizumi et al., 2006;Hayakawa et al., 2016), was evident when fully saturated debris flows occurred, resulting in formation of step-pool topography by large boulders after fully saturated debris flows.
Because we did not consider the dynamic mechanisms of the debris flow, including the collision of particles and the dynamic pressure of interstitial water (Coussot and Meunie, 1999;Takahashi, 2014), our analysis does not explain the unsteady nature 10 of debris flows, such as temporal changes in the flow type during a debris flow event (McArdell et al., 2007;McCoy et al., 2013;Okano et al., 2012).Nevertheless, our analysis enables an understanding of the overall relationship between the predominant types of debris flow and slope gradient.gradient, α1 is the theoretical boundary of the channel gradient between fully unsaturated and partly saturated sediment transport, and α2 is the theoretical boundary of the channel gradient between partly and fully saturated sediment transport.

Summary and conclusion
We investigated the interaction between flow characteristics and topography in the debris flow initiation zone at the Ohya landslide, Japan, using various methods, including a physical analysis of static force, field monitoring, periodical TLS surveys, 20 and an estimation of the volume of storage using GIS.Our study revealed that both partly and fully saturated debris flows are important hydrogeomorphic processes in the initiation zone because of the steep terrain.The predominant type of flow (partly or fully saturated) is affected by the volume of storage as well as rainfall patterns, which control the amount of water in the storage.In addition, small-scale channel gradients (on the order of meters) formed by erosion and deposition during debris flow events reflect the dominant flow type in the flow.Such interactions between flow characteristics and topography could 25 be explained by a simple analysis of static force at the bottom of the sediment mass.

Figure 2 :
Figure 2: Map of the Ohya landslide and upper Ichinosawa catchment: (a) Ohya landslide, (b) Ichinosawa upper catchment.Gentler and steeper terrains are expressed as light and dark colors, respectively.

Figure 4 :
Figure 4: Photograph of the typical geomorphic units seen from P1.The image shows the area of rock slopes (surrounded by red lines), talus slopes (surrounded by yellow lines), and a channel (surrounded by blue lines), for which the slope gradient was calculated from DEMs with various grid sizes.The numbering of the area corresponds to that used in Fig. 7.

Figure 5 :
Figure 5: Hydrograph and the duration of each flow type during debris flow events.(a) Debris flow on August 30, 2004, mainly consisting of fully saturated flow.(b) Debris flow on August 5, 2008, mainly consisting of partly saturated flow.

Figure 6 :
Figure 6: Comparison between the volume of storage and the duration of partly saturated flow as a proportion of the overall debris flow surges.

Figure 7 :Figure 8 : 10 DEMs
Figure 7: Slope gradient map calculated from terrestrial laser scanning (TLS) data on 14 May 2014, with various grid sizes.The location of the analysis area is shown in Fig. 2.5

Figure 9 :
Figure 9: Channel profiles and channel gradient mainly formed by fully saturated flow (measured on 21 November 2012) and partly 5 Earth Syst.Sci.Discuss., doi:10.5194/nhess-2017-20,2017 Manuscript under review for journal Nat.Hazards Earth Syst.Sci.Published: 16 January 2017 c Author(s) 2017.CC-BY 3.0 License.Table 3 Standard deviation of the channel gradient and proportional amount of channel sections in which the gradient was in the range between 2 and 1 throughout the entire channel section.Sections with active channel bed deformation (from 0 to 25 and from 75 to 90 m from point ASections without active channel bed deformation (from 25 to 75 and from 90 to 100 m from point Earth Syst.Sci.Discuss., doi:10.5194/nhess-2017-20,2017 Manuscript under review for journal Nat.Hazards Earth Syst.Sci.Published: 16 January 2017 c Author(s) 2017.CC-BY 3.0 License.

Figure 10 :
Figure 10: Comparison between the volume of storage and channel topography.(a) Comparison of the volume of storage with the standard deviation of the channel gradient for each 0.1, 1.0, and 5.0 m section in the active channel bed deformation zones (from 0 to 25 m and from 75 to 90 m upstream of point A).(b) Comparison of the volume of storage with the length of the channel sections, in which the channel gradient ranged between 1 and 2 (theoretical channel gradient for partly saturated debris flow) and gentler than 2 (theoretical channel gradient for fully saturated debris flow), in the active channel bed deformation zones.Topographic data 5 Earth Syst.Sci.Discuss., doi:10.5194/nhess-2017-20,2017 Manuscript under review for journal Nat.Hazards Earth Syst.Sci.Published: 16 January 2017 c Author(s) 2017.CC-BY 3.0 License.saturated flow) in the entire channel section on 23 August 2015 (mainly formed by partly saturated flow) was higher than on 21 November 2012 (mainly formed by fully saturated flow), regardless of the length of channel section used to calculate the channel gradient (Table

Table 1 : Date of TLS survey
Latter half of the debris flow event was not analyzed because of the darkness affected by the sunset.Therefore, latter half of the event is not included in this data. *a