Headwater sediment dynamics in a debris flow catchment constrained by high-resolution topographic surveys

Abstract. Debris flows have been recognized to be linked to the amounts of material temporarily stored in torrent channels. Hence, sediment supply and storage changes from low-order channels of the Manival catchment, a small tributary valley with an active torrent system located exclusively in sedimentary rocks of the Chartreuse Massif (French Alps), were surveyed periodically for 16 months using terrestrial laser scanning (TLS) to study the coupling between sediment dynamics and torrent responses in terms of debris flow events, which occurred twice during the monitoring period. Sediment transfer in the main torrent was monitored with cross-section surveys. Sediment budgets were generated seasonally using sequential TLS data differencing and morphological extrapolations. Debris production depends strongly on rockfall occurring during the winter–early spring season, following a power law distribution for volumes of rockfall events above 0.1 m3, while hillslope sediment reworking dominates debris recharge in spring and autumn, which shows effective hillslope–channel coupling. The occurrence of both debris flow events that occurred during the monitoring was linked to recharge from previous debris pulses coming from the hillside and from bedload transfer. Headwater debris sources display an ambiguous behaviour in sediment transfer: low geomorphic activity occurred in the production zone, despite rainstorms inducing debris flows in the torrent; still, a general reactivation of sediment transport in headwater channels was observed in autumn without new debris supply, suggesting that the stored debris was not exhausted. The seasonal cycle of sediment yield seems to depend not only on debris supply and runoff (flow capacity) but also on geomorphic conditions that destabilize remnant debris stocks. This study shows that monitoring the changes within a torrent's in-channel storage and its debris supply can improve knowledge on recharge thresholds leading to debris flow.


Introduction
In steep mountain catchments, rainfall intensity and duration (including snowmelt) are insufficient to predict debris flow occurrence, even though the initiation of runoff-generated debris flows requires significant water inflow (Van Dine, 1985;Decaulne and Saemundsson, 2007;Guzzetti, 2008). In many cases, the properties of the channel reach which determine the amount of debris that can be entrained can be often more important than the mechanisms of initiation induced by the hydrological or meteorological conditions prior ards (Rickenmann, 1999;Jakob et al., 2005). However, the difficulty results in quantifying sediment processes and rates and volumes from hillslopes and in-channel debris storage (Peiry, 1990;Zimmermann et al., 1997).
The quantification of the overall sediment production and transfer rate has increasingly relied upon multi-temporal digital stereophotogrammetry (Coe et al., 1993;Chandler and Brunsden, 1995;Veyrat-Chavillon and Memier, 2006) and elevation difference from high-resolution digital elevation models (HRDEMs) (Smith et al., 2000;Wu and Cheng, 2005;Roering et al., 2009;Theule et al., 2012). In terrain dominated by steep slopes, traditional aerial-derived digital elevation models (DEMs) are typically inappropriate to study geomorphic processes. Limitations include the poor rendering of small topographic changes (Perroy et al., 2010), the poor representation of steep terrain with small curvature radii and data gaps in vertically oriented and overhanging topography. Even on gentler gradients, the sharp breaks in slope, encountered in erosion scars for instance, are often insuffi-ciently modelled by airborne HRDEMs, leading to erroneous volume estimations (Bremer and Sass, 2011). This represents a serious drawback in estimating the sediment budget of steep terrain, where sediment activity comes mostly from rock walls and rugged gullies. Because of these issues, many hillslope and rock slope process studies have used terrestrial laser scanner (TLS) data to build the topographic model . The recent development of longrange TLS devices provides an effective means of acquiring high-resolution topographic information that can adequately reflect the morphology of steep bedrock-dominated areas. The practical disadvantages in data acquisition inevitably related to ground surveys can be compensated for by flexibility in transport, ensuring a full coverage with minimal zones of topographic shadowing.
This paper presents a quantitative study of sediment recharge and channel response leading to debris flow events, using 3-D digital terrain models acquired by TLS. This is illustrated on the Manival (French Alps), a torrent that experi- Figure 2. Geological map of the catchment headwater (production zone) after Gidon (1991) and location of first-order debris flow channels (thick blue line) and their respective watersheds (white lines). For the ease of analysis, the Roche Ravine and Col du Baure subcatchments in the east side were further subdivided according to their gully complex (dotted white lines). ences runoff-generated debris flow almost every year (Péteuil et al., 2008). The surveys captured hillslope processes and sediment dynamics occurring throughout the system including the tributary channels down to the main torrent and were performed periodically over 16 months. The spatio-temporal variability of debris production and subsequent transport and storage of sediment are analysed on a seasonal timescale, in order to discuss the debris supply dynamics and the implications in debris flow initiation. This study also complements a parallel investigation regarding the controls on debris flow erosion and bedload transport in the Manival torrent (Theule et al., 2015).

General setting
The 3.9 km 2 Manival catchment located at the edge of the Chartreuse Massif (France) (Fig. 1) has a rugged, 1200 m relief watershed, resulting from deep headward entrenchment (Gidon, 1991). The topography consists of a narrowly confined head and a steep-sided colluvium-filled valley, delimited in the west by a series of rock walls and screemantled deposits separated by rock couloirs and in the east by steep rock and talus slopes divided by gullies. The lithology ranges in age from Late Jurassic to Early Cretaceous ( Fig. 2) (Charollais et al., 1986). In the heart of the basin, thick sequences of calcareous marl interbedded with layers of marl predominate. Towards the ridge, the bedrock evolves progressively from stratified to more massive limestone. The valley sides are formed by the fold limbs of an anticline, where secondary folding and minor faults induce local variations in structure (Gidon, 1991). This tectonic setting and the varying stratigraphic competence have strongly influenced the topographic development of the catchment, providing a dynamic geomorphic environment producing considerable runoff as a response to heavy, frequent rainstorms (Fig. 3).

Characteristics of the headwater sediment dynamics
The contemporary geomorphic activity contributing to the torrent's recharge with debris is concentrated exclusively in the headwater, where no remnant glacial deposits are found (Gruffaz, 1997). In the upper catchment, large old rock deposits flooring the west side hillslope (Fig. 4) have dramatically influenced the bottom topography, and thus the channel network, resulting in a conjunction of four first-order debris flow channels deeply incised down to the bedrock in several reaches. The upper catchment can therefore be subdivided into five subcatchments in terms of sediment recharge (Fig. 2). Bed entrenchment is now constrained by check dams. However, lateral erosion still occurs episodically by flooding and debris flow scouring.
The style of sediment production and delivery is somewhat different throughout the headwater, according to the local morphology and the lithologic and structural setting. The major geomorphic processes, identified preliminarily by observations from aerial photographs and field investigations, were initially characterized in a map (Fig. 4) describing the spatial distribution of geomorphic features and sediment

N o data
Bedload transport -1 Figure 3. Maximum rainfall intensity over the monitoring period measured by a rain gauge located at the top of the torrent (see Fig. 4) and calculated for a 5 min time interval. The mean annual precipitation is about 1500 mm in the headwater of the Manival (modified from Loye, 2013).
transfer processes contributing to debris recharge in the firstorder channels. The west and upper sides are dominated by rockfall. Large rock collapses delimited by persistent joints occur due to the progressive degradation of the slope underneath (Loye et al., 2011). Where the slope gradient allows scree and soil development, erosion scars can be observed; sediment sources are remobilized from discrete shallow landslides. Depending on the location and size, rockfall can reach the channels directly or accumulate on slopes or in ravines, before being subsequently routed to high-order segments by a combination of gravitational and hydrological processes. Towards the east, the erosion seems to be more progressive through the formation of gullies . Near the ridge, the slopes display mostly talus and scree deposits lightly covered with vegetation, whereas the hillside below exposes steepened rock slopes. Many active erosion scars can be observed. They contribute debris into gullies and talus slope deposits that are subsequently entrained in channels downslope. Historical records of debris flows since the 18th century show a frequency of 0.3 events per year that reached the apex of the fan (Brochot et al., 2000). The largest event deposited approximately 60 000 m 3 . However, the torrent experiences smaller fluxes of debris (< 1000 m 3 ) usually not reported in archives. Such events can occur 2-3 times per year, when initiated by intense runoff (Veyrat-Charvillon, 2005). Volumes of debris deposited in the sediment trap for the last 25 years are on average 2200 m 3 yr −1 , reaching a maximum of 7000 m 3 yr −1 in 2008 (RTM service, National Forests Office (France)).

Topographic monitoring using TLS
The terrain was surveyed with an ILRIS-3D laser scanner (Optech Inc.). This device provides a range of up to 1.2 km for 80 % reflectivity surface, and the instrumental precision is about 7 mm/100 m range for both distance and position (Optech Inc.). The overall coverage of the upper catchment with TLS point clouds required 50 scans using a 20 % surface overlap. These scans were collected over a 5-day period from nine individual viewpoints to ensure a full 3-D rendering of the topography. Particular attention was given to irregular regions and major breaks in slope, such as rock couloirs and deep-cut gullies. Using multiple scanning locations allowed us to limit shadow zones and increase the point cloud density of the scanned area. A series of four surveys was performed for each season during 2009, and one extra survey was performed in July 2010 to analyse the effect of the preceding winter period (Table 1). The monitoring setup remained similar for all surveys. Post-processing of the TLS raw data was done using Polyworks (InnovMetric). Erroneous points and vegetation were filtered manually, ensuring a total control of the removed data to preserve a high density of points in topographic features with small radii curvature. Although this procedure is time consuming, box (semi-)automatic approaches to filter vegetation accurately still remain in a stage of development for dissected mountain morphology (Brodu and Lague, 2012). Each of the multiple scans of a survey was merged with another one using common tie points of permanent topographic features and the dataset was processed as 12 standalone sub-datasets, rather than all processed together. Given the size of the monitored area, dividing the point cloud  . Geomorphic process map (contour interval: 20 m) illustrating the spatial pattern of sediment sources and transfer in the first-order channel complex. Note the impressive rock collapse deposits now crossed by four first-order debris channels. Their bed incision is strongly constrained by a series of check dams (marked as black "T" on the map), but erosion scars all along the deposit suggest that the reaches are still subject to lateral erosion.
into smaller datasets avoids the propagation of inaccuracy through large co-registered scan series. ICP (iterative closest point) algorithms (Besl and McKay, 1992), which minimize the distance between two point clouds, were used to determine the best alignment of subsets surveyed at different times in order to obtain the best co-registration within a time series. The same procedure was applied between subset point clouds and a point cloud derived from a commercial airborne laser scanner (mean density: 6.9 pts m −2 ) and acquired in June 2009 to place the TLS data into the standard Lambert projection coordinate system used in France. The initial survey point cloud data were set as the surface model of reference. Each successive survey was georeferenced onto this reference using ICP. The topographic change occurring between two successive surveys is too localized to influence the global co-registration within two survey data subsets consisting of millions of data points, hence the alignment accuracy. More details about multiple scan registration techniques and point cloud time series comparison can be found in Oppikofer (2009). The generated surface produced by the above  procedure has a point spacing ranging from 2.5 to 18 cm according to the distance of acquisition. A maximum range of about 800 m was reached on the top peak of the catchment with a point cloud density of 25 pts m −2 . The surface coverage of our surveys represents 84 % of the deforested area under investigation (Table 2).

Topographic change identification and characterization
The active geomorphic features within two successive datasets were identified on a point-by-point basis using the short-distance neighbouring point search algorithm (Bitelli et al., 2004) that computes, in 3-D, the shortest difference vectors between the points of two datasets. The vector sign indicates the net change direction of topography, i.e. surface of erosion or deposition. A set of points (cluster) was considered active if at least eight adjacent points of similar sign displayed an absolute difference above the limit of detection (LoD, see Sect. 3.4). Each active feature was outlined visually using the point cloud of difference (Fig. 5a). The point clusters of both survey datasets, which correspond to the topography of the active features, were extracted according to their spatial extend coordinates and each detected geomorphic feature was labelled as follows: 1. rock slope erosion, characterized by rockfall or rockslides; 2. hillslope erosion, specifically the reworking of loose or compacted debris on slope, in gullies, and in channels; 3. deposition, including material aggradation initiated by both rock slope failure (new production) and remobilization of debris.
Using the images captured by the TLS integrated camera, clusters of points not corresponding to geomorphic process activity, such as snowmelt, were ignored.

Volume computation of each geomorphic feature
As the volume of active features cannot be directly computed by differencing TLS point datasets, the active features of two successive point clouds must be interpolated into continuous surfaces (DEM). Gridded model (or raster) is regarded as being the most effective type of model to use for irregularly distributed datasets, which sometimes contain few or no points (El-Sheimy et al., 2005), as can be the case for rockfall and erosion scars. The algorithm chosen for the interpolation of the DEM has little influence on the final result, as TLS data provide an extremely dense coverage of the detected objects (Anderson et al., 2005). Therefore, they were interpolated using linear inverse distance weighting (Burrough and McDonnell, 1998) and generated in a regular grid separately. The grid spacing and direction of interpolation were designed in a specific way for each feature: the coordinate system of reference was replaced by a local orthogonal system where the x-y axes represent the average plane of topography nearby (Fig. 5b). This new reference frame was defined using eigenvalue decomposition of the covariance matrix of the point cloud of reference (Shaw, 2003). Interpolating the surface elevation in the direction of local topography allows the generation of a realistic DEM independent of slope steepness and, thus, a close realistic representation of topography in the case of overhanging features. The cell size was defined according to the point spacing distribution of both datasets. A series of tests revealed that setting the grid spacing at 68 % of the cumulative frequency distribution of point spacing provides a continuous surface reconstruction while keeping a high degree of detail from the point cloud. This ensures an accurate volume computation of geomorphic features. The volume was computed as the sum of the cell difference in elevation (both positive and negative) between the successive DEMs. Absolute cell differences lying below a given threshold (see Sect. 3.4) were not considered. This volume computation using a local deterministic method of interpolation and an adaptive gridding approach was developed in the Matlab numerical computing environment. µ σ min max Figure 6. Distribution of the distance between two survey point clouds after the process of georeferencing using the ICP procedure. The distance approaches normal distribution with a zero mean, showing that errors generated by multiple scan registration and point cloud survey georeferencing are Gaussian, random, and independent. Data are given in metres. Table 3. Registration and georeferencing standard deviations (in centimetres) of the position uncertainty on a point by point basis that was used to derive the LoD at 95 % confidence interval and subsequently to detect topographic changes down to a certain minimum volume of geomorphic features.

Point cloud accuracy and limits of detection of the geomorphic features
A reliable identification of erosion and deposition features requires the definition of a LoD, where the change in elevation between successive point clouds can be considered real as opposed to noise. Each TLS data point theoretically has a unique precision depending on the range and laser incidence angle (Buckley et al., 2008). In practice, the individual point precision of a scan can be assumed to model a surface with a global uniform uncertainty, considering the very high point density (Abellàn et al., 2009). Given the homogeneity of surface error and considering that the distance between sequential points at a position (x, y) should tend to 0, the accuracy of TLS data can be estimated by substituting the precision of each data point by a singular measurement of the error associated with the entire point distribution across the surface (Lane et al., 2003). Hence, the uncertainty related to both scan registration and point cloud georeferencing, the instru-mental error included, was defined by the standard deviation of the distance (σ d ) between the points (Fig. 6). The LoD was therefore set at 2σ of the co-georeferencing and corresponds to the 95 % confidence limit (Table 3). Comparison with the approach considering the error propagation for all uncertainties associated with each point cloud and assuming a normal distribution of the error in distance (Taylor, 1997) shows that the uncertainties considered here are consistent.
In the case of volume computation, information on elevation uncertainty associated with each point cloud survey needs to be extended to the DEM on a cell-by-cell basis. For any grid cell (i, j ) generated by the interpolation of adjacent points p with independent elevation, the uncertainty of a cell elevation can be considered the standard deviation (σ e ) of the data points elevation, where σē i, j = σ e p / √ n according to the equation of standard error of the mean, n being the number of points to define the cell elevation. The elevation uncertainty for each cell in a DEM of difference is then expressed by The volume uncertainty is then calculated by summing up the derived volume uncertainty of each cell of the feature as follows: with a = cell area. The smallest detectable volume is about 10 −3 m 3 (10 × 10 × 10 cm) ( Table 3) but can reach up to 0.006 m 3 (25 × 25 × 10 cm) depending on the point spacing at maximum range. Topographic change detection and volume computation accuracy depend not only on the quality of the TLS data, such as point density and post-processingrelated inaccuracy. They also depend on the complexity of the surface geometry, like in our case, by integrating the range in position of all data points defining each grid cell value of a feature. Monitoring the hillslope activity is also limited by the ability of the process to create a distinct topographic change. Consequently, the deposition of individual small rockfalls was not always detected, as detached rock masses fragment into smaller pieces that are below the LoD. A similar issue was observed for erosion processes within debris. Nevertheless, most of the material accumulation could be related to upslope landslides or scouring. The sediment budgets were therefore kept in volumetric units, as they are commensurate for a consistent analysis. They were not converted to mass, although this would make more sense for comparing hillslope processes and rock slope yields. Such conversion requires an accurate density value of each surface process, whose approximations introduce additional unknowns. Deposition related to rock failures may therefore be slightly overrepresented in the sediment balance, although this could be partly compensated for by a limited detection of small features.

Sediment budgets of the Manival torrent
Monitoring of the coarse-sediment transfer has been performed all along the main torrent channel to the sediment trap located downstream on the alluvial fan. The in-channel storage change was established after every noticeable flow event, using the morphological approach based on crosssection survey techniques (Ashore and Church, 1998), and the volume of sediment deposited in the sediment trap was measured by TLS survey differencing. Sequential volumes of recharge enable us to study the influence of debris supply from the production zone through the seasons. The characteristics and observational analysis of this event-based monitoring were documented in detail in Theule et al. (2012Theule et al. ( , 2015 and are therefore not described any further.

Estimation of debris production rate
A rate of debris production for the study period is obtained from the total volume of rock slope erosion. An objective estimation can be deduced by characterizing the cumulative distribution of rockfall volumes with a power law as follows (Gardner, 1970): N is the rockfall frequency for a volume ν greater than V and a and b are constants. a depends on the study size and on rock slope properties, whereas b tends to be rather site independent Dewez et al., 2011).
Considering that rock slope process activity causing rockfall does not fluctuate much over time, the inventory analysis can be used to infer the frequency of the occurrence of larger events. This is done by integrating the rockfall frequency derivative n(v) = dN dV over the range of potential volumes. The estimation of the total volume V t per unit time that can be expected on average over a longer period of observation is therefore expressed by (modified from Hantz et al., 2002) The goodness of fit of the power law was evaluated with the χ 2 test (Taylor, 1997) and the standard deviation of values a and b was determined with the maximum likelihood estimate (Aki, 1965). The erosion rates are assessed by dividing V t with the surface prone to rockfall.
4 Results: hillslope process activity monitoring

First monitoring period (April-August 2009)
The topographic changes recorded from July to August 2009 did not show any relevant geomorphic activity (only a few small rockfalls). These results were therefore merged with the preceding monitoring period. Rock slope activity is dominated by individual small rockfalls distributed throughout the upper catchment. Only few events exceed 1 m 3 , such that contributions in terms of debris production are marginal in most parts of the catchment (Fig. 7). The most significant geomorphic activity was located almost exclusively in the major gullies of Baure and Grosse Pierre ravines and consists essentially of debris scouring of a few 100 m 3 redeposited further down. Material reentrainments were also observed in several other smaller gullies, but their volumes are relatively small. The rock couloirs of the Genièvre subcatchment and the scar of the old rock deposit barely showed any geomorphic activity. The channels displayed a net incision (−636 m 3 ± 43) in the upper reaches. Bedload aggradation remains very low (+90 m 3 ± 6). Below the upper confluence, the channel trunk exhibits a mixed pattern of zones of erosion (−60 m 3 ± 2), such as gravel-wedge scouring, and zones of redeposition of entrained material (+80 m 3 ± 4) induced by bedload transport.

Second monitoring period (September-November 2009)
Rock slope activity remains similar in spatial extent and volumes to the previous survey period, but rockfall frequency is higher (Fig. 8). Hillslope process activity was more widespread on the east side but more localized on the western valley walls, while the rock couloirs showed no geomorphic activity. In the upper headwater, material reworking was concentrated almost exclusively in the steep tributary gullies. They displayed scouring of a relatively large volume (−357 m 3 ± 12). Deposition features along the thalweg were almost inexistent (+18 m 3 ± 1.3). In the southeast, not only the Baure Ravine (net erosion: −61 m 3 ± 8) but the whole series of hillside gullies exhibited signs of activity, such as erosional segments alternating with deposition. On scree slopes, several minor areas with erosional rills and their associated debris deposits were observed, some of them reaching the channel trunk (+42 m 3 ± 2). Such small hillside debris flows were probably triggered by sediment entrainments within the rills, as no evidence of sliding at their head was observed. The channels show a net erosion upstream (−482 m 3 ± 18), whereas continuous incisions were more pronounced in the Manival channel (−443 m 3 ± 16) and also in the Roche Ravine (−40 m 3 ± 3). Deposition zones were almost completely absent (15 m 3 ± 1.3). Towards the upper confluence, the lower segments of the Manival channel exhibited continuous zones of aggradation (97 m 3 ± 6) that were scoured on one side. This morphology is characteristic of closed-process debris flow levees and run-up zones beside the incised channel bed. Below the upper confluence, channel bed cut (−40 m 3 ± 2) and fill (+16 m 3 ± 1) was sparse and concentrated at the junction with hillside gullies. Such a pattern of bed reworking demonstrates the connectivity of the Baure gully series with the channel trunk.

Third monitoring period (November-July 2010)
This period showed an important increase in rock slope erosion, both in frequency and magnitude, resulting from the occurrence of large slope failures and enhanced localized rockfall activity, for instance in rock walls made of calcareous marl situated directly above the Manival (2035 m 3 ± 39) and the Roche Ravine (256 m 3 ± 17) channels (Fig. 9). Most of the debris collapses supplied the channel directly; the rest was temporarily deposited in breaks in the slope. The lower headwater part showed a great fluctuation as well (Genièvre: 116 m 3 ; Grosse Pierre: 145 m 3 ). At the top of the Baure Ravine, 816 m 3 ± 25 of rock fragments contributed substantially to recharge the sediment storage at the gully head. Below, debris infilling was continuously scoured. A 1170 m 3 ± 18 rockslide is responsible for a large channel infill in the Manival subcatchment. Several other smaller rockfalls contributed to the recharge of tributary gullies and scree hollows. In the Roche Ravine, debris deposits were sparse because rockfall remained of a low magnitude on average (571 events < 1 m 3 ), although frequency was high (578 events). The large debris infill at the channel head was caused by two erosion scars in the gullies (270 m 3 ± 14 and 65 m 3 ± 4). In the rock couloirs of the Genièvre subcatchment, a significant accumulation of material from landslides and rockfalls was observed (remnant volume: 204 m 3 ± 13), taking into account that the hillslope erosion represents 450 m 3 (±14). In the Grosse Pierre Ravine, 343 m 3 ± 17 of debris were accumulated at the rock couloir outlet, recharging the scree slope above the channel head. In the Col du Baure, relatively large aggradation in the lower part of tributary gullies was observed (remnant volume: +142 m 3 ± 2), resulting from material entrainment. Several debris slides were also detected on scree slopes, without any contact with the channel trunk.
The upper channel reaches were clearly depositional, as a consequence of large slope failures. The Manival channel showed a continuous zone of remnant accumulation of 948 m 3 (±18) of which a portion was carried along downstream as bedload. Towards the confluence, erosion dominated (−487 m 3 ± 19) over deposition (+25 m 3 ± 3). In the Roche Ravine, a continuous zone of erosion in the scar of the old rock deposit produced debris accumulation mostly on the slope. But a landslide of 190 m 3 ± 9 reached the channel. Overall, aggradation was observed all along the channel head (+148 m 3 ± 18) and scouring was limited (−65 m 3 ± 4). From the confluence downstream, the channel behaviour is dominantly erosional (−97 m 3 ± 4) almost without any aggradation (+3 ± 0.3 m 3 ).

Rock slope production inventory
Over the 16 months, 1866 rockfalls with volumes ranging from 10 −4 to 10 3 were recorded. This yields a total of 3575 m 3 ± 30 and an erosion rate of 3.1 mm yr −1 , given the topographic surface area of rock faces. The inventory follows a power law (Fig. 10) with a 99 % confidence level for events larger than 3 m 3 (χ 2 value = 17.3). For events larger than 1 m 3 , the power law is accepted at the 95 % confidence level (χ 2 value = 5.89). Both threshold volumes provide a b value close to 0.81 ± 0.06. Considering only the volumes above 10 m 3 (25 events) gives a b value of 0.76. Below 0.1 m 3 , the observed frequency deviated clearly from the power law regime until the rollover reached an approximately constant rate for the smallest volumes. According to our inventory, rockfall of more than 1 m 3 is expected 153 ± 11 times per  year on average. The largest event (1170 m 3 ) occurs every 2 years, and the 1-year return period rockfall has a volume of approximately 465 m 3 . Considering only these classes of volumes of the inventory (see Table 6), the rock slope production reaches a rate of 3678 m 3 yr −1 ± 210 (4 mm yr −1 ± 0.3).

Torrent in-channel storage changes
Two debris flows with multiple surges and several remarkable bedload transport events were observed in the main torrent during the survey period . A debris flow occurred on the 25 August 2009, caused by a short-duration rainstorm. The volume of sediment eroded in the torrent (5232 m 3 ± 136) is equivalent to the volume that was redeposited in both the torrent itself and the sediment trap (5072 m 3 ± 125), suggesting that the majority of entrained material was stored in the torrent (Table 4). Sediment input from the headwater can be considered marginal. Before that, no significant torrent activity was observed, despite a series of rainfall events with low to moderate intensity. In September 2009, a long period of moderate rainfall intensity caused material reworking by bedload transport all along the torrent. However, no sediment was supplied to the sediment trap. For each dataset, the power law is fitted for volumes larger than 0.1 m 3 . Below this threshold volume, the distribution exhibits a rollover that progressively reaches an almost constant frequency for the smallest detected volumes.
net gain of storage in the headwater was therefore inferred. In October, a succession of low-intensity rainfall events triggered sediment transport in the torrent that accumulated in the sediment trap with a volume of at least 302 m 3 ± 36. The sediment budget indicates clearly a recharge of 229 m 3 ± 31, a transfer of debris that was stored mostly in the distal part of the torrent. Throughout the winter, a gradual incision was observed all along the torrent, resulting from frequent periods of low-intensity rainfall as well as snowmelt. Due to maintenance (dredging), the sediment trap was disturbed and no reliable data were available. In any case, no sign of significant sediment activity was detected. A new debris flow on 6 June deposited 3320 m 3 ± 176 in the sediment trap. This time, a certain supply of sediment from the headwater was observed (∼ 270 m 3 ). This event was followed by a series of intense rainfall events without much reworking in the distal part, suggesting that any significant transfer occurred into the torrent downstream. The in-torrent storage changes and estimated recharge budgets are shown for each monitoring period in Fig. 11.

Synthesis
The overall transfer dynamics, from debris source zone to the apex of the fan, are illustrated in Fig. 12. The volumes detected during the 16-month study period reveal a net export of 3378 m 3 ± 361 of sediment from the headwater to the main torrent (Table 5). The overall rock slope yield is 3575 m 3 ± 30 for a volume of erosion reaching 3129 m 3 ± 150 on the hillside and 1809 m 3 ± 92 in the channel complex. The volume of deposition, induced by both debris production and material reworking, yields a total volume of 5135 m 3 ± 251, of which only 1382 m 3 ± 56 (27 %) is linked to the channel complex. In the main torrent, the sediment transfer was relatively large (∼ 20 000 m 3 ; net storage change −4950 m 3 ± 118) and essentially related to the occurrence of two debris flows , depleting significantly the in-torrent sediment storage of the distal parts (entrainment zone). Material deposited in the sediment trap for the survey period yields 6075 m 3 ± 45. During the  Table 4 for details on values) (modified from Theule et al., 2012).
autumn, bedload transport of hundreds of square metres contributed to sediment recharge throughout the torrent.
In the spring-midsummer period, the hillside sediment budget yields a total rock slope production of 99 m 3 ± 6, for a volume of erosion of −547 m 3 ± 50 and deposition of +408 m 3 ± 35 (Table 5). This suggests that about 238 m 3 ± 61 of material was supplied the channel complex, originating almost exclusively from material re-entrainment in gullies (Fig. 13). The sediment budget of the channels indicates a significant reduction in storage (−487 m 3 ± 44), comprising large and continuous incisions (−636 m 3 ± 43) in the upper reaches and material aggradation (+149 m 3 ± 11) in the lower reaches resulting mostly from zones of transient redeposition. This results in a recharge of the torrent of +726 m 3 ± 103 for this survey period. During the late summer-autumn season, the total volume of hillside erosion is −640 m 3 ± 27, due to a widespread scouring of the tributary gullies located east and south-east of the headwater (Fig. 14). The total volume of rock slope production (50 m 3 ± 3) and deposition (+182 m 3 ± 12) remained low. Overall, the sediment budget indicates that the hillslope contributed about 510 m 3 ± 30 of sediment to the channel reaches (Table 5). The sediment budget of the channels yields −522 m 3 ± 20 of erosion for +127 m 3 ± 13 of deposition. This is characterized by bedload reworking in both low-order and trunk channels and a progressive transfer of +904 m 3 ± 51 of material into the torrent. During winter-spring 2010, a total deposition volume of +3163 m 3 ± 147 is recorded on the hillside for an eroded volume of −3129 m 3 ± 150. A relatively large production of debris (3424 m 3 ± 89) is observed ( Table 5). The net sediment balance on the hillside yields a supply of +2203 m 3 ± 187 of sediment into the channels, and the net sediment balance for the channel complex indicates an increase in in-channel sediment storage of +455 m 3 ± 47 for a total volume of deposition of 1105 m 3 ± 36 and erosion of 651 m 3 ± 29 due to large bed scouring zones in the downstream reaches. Sediment transfer into the torrent is 1749 m 3 ± 199 (Fig. 15).

Debris supply through rock slope production
Debris production from rock walls shows a strong seasonal pattern. The great majority of recorded rock instabilities in both magnitude (95 %) and frequency (75 %) occurred during the cold period. Previous studies of the calcareous cliffs near Grenoble, which have a similar morphotectonic context, revealed that freeze-thaw cycles are the main triggering factor of rockfall (Frayssines and Hantz, 2006). Ice jacking can cause microcrack propagation, leading to failure (Matsuoka and Sakai, 1999). Along the eastern ridge, the bedrock surface is often highly fractured, suggesting frost shattering. The spatial pattern of rockfall also strongly suggests a tectoniclithological influence that can be explained by differential erosion between the successive limestone and marl beds. In the rock wall series on the west side, the monoclinal configuration of the bedding, combined with a strong difference of competency between stratigraphic sequences, gives rise to an overhanging formation highly susceptible to failure. On the east side, the bedding is mostly cataclinal and approaches dip slope, depending on the slope. Rock failures initiated by planar sliding on bedding planes were observed. The observed debris production follows a power law distribution in a range covering at least 3 orders of magnitude [10 0 -10 3 ]. The exponent b is slightly higher than the average value reported for the Grenoble cliffs ([0.4-0.7]; Hantz, 2011) but is in agreement with other short inventories covering a lower range of volume ([10 −2 -10 2 ]; Hungr et al., 1999;Dussauge et al., 2003). Inventories dominated by small volumes tend to increase the b value, compared to the ones covering rather large volumes (Stark and Hovius, 2001). Above 100 m 3 , the deviation from the power law may be attributed to the short period of sampling for events of such a large magnitude. The rollover encountered towards small volumes results most likely in the under-detection of the number of events. This sampling bias is far above the minimum volume of detection (0.006 m 3 ); therefore, another behaviour characterizing the failure of small volumes cannot be excluded. This may take the form of a physical erosion process that differs from the one influencing larger instabilities, which are controlled primarily by the geometrical and geomechanical properties of the rock mass (Selby, 1993;Sauchyn et al., 1998), and tectonic weakening (Cruden, 2003;Coe and Harp, 2007). As observed here, low-magnitude rockfall events represent a low proportion of overall debris supply, even though they vary locally from 1 or 2 orders of magnitude in volume over time. The total amount of sediment available is only significantly influenced by high-magnitude instabilities (Fig. 16).
Previous sediment budgets derived from topographic measurement using stereophotogrammetry estimated the highest erosion rates over an average of 40 years to range from 10.8 to 17.8 mm yr −1 in the headwater (Veyrat-Charvillon and Memier, 2006). Given the large uncertainty of the approach, and the fact that they measured the hillslope and thalweg geomorphic activity, these values are broadly consistent with the erosion rate derived here from a short-period rockfall inventory by assuming the possible occurrence of rockslide magnitudes [10 6 -10 7 ]. Considering that the power law is valid for larger slope failures, a 7500 m 3 event can be expected every 10 years and a 120 000 m 3 event every 100 years. The average debris production ranges between 5587 ± 241 and 12 903 ± 305 m 3 yr −1 , assuming a maximum potential erosion of 10 5 and 10 7 m 3 respectively over several centuries (Table 6). No historical Manival rockslide exists to support this estimation. The large old rock deposit (∼ 6.1 Mm 3 ) of the upper catchment is the largest detected event, but it may have formed from several rock collapses. The rockfall inventory of the Grenoble cliffs reports volumes smaller than 10 5 m 3 for the last century and 10 7 m 3 since the 17th century . Such a magnitude is also likely at the Manival. A mean rate of rock slope erosion of approximatively 10 mm yr −1 . 10 000 m 3 yr −1 can be therefore expected in the upper catchment over the century.
Upstream from the Manival channel, the scouring of debris slopes and scree hollows triggered by rock slope production accounted for about 40 % of the net erosion recorded during the autumn period and 25 % in the Baure Ravine over the entire study period. The spatial pattern of geomorphic work showed that hillslope process activity was observed principally in gullies and scree slopes situated directly below active rock walls. The dominant mode of debris supply in the Manival headwater is therefore highly episodic, implying a great spatial heterogeneity in sediment recharge rates.

Debris supply through hillslope activity
As rock slope activity was very limited from spring to autumn, hillslope geomorphic activity dominated sediment recharge during this period. Until the end of August, hillside gullies and low-order channels remain almost inactive in terms of sediment delivery. Conversely, the autumn period was characterized by a general increase in the intensity of geomorphic activity. Continuous scouring and the relative paucity of deposition features from hillside gullies as well as clear incisions and micro debris flows in channel reaches indicate that mobilized material was almost entirely entrained downstream by runoff. For the entire area, the hillside contribution represents on average a volume 5 times larger than the volume that was observed in spring and summer, and channel bed reworking was of a much larger magnitude as well.
During winter-spring 2010, the total volume of deposition recorded on the hillside significantly exceeds the rate of deposition recorded so far, resulting from the huge increase in debris production that can be attributed to the winter according to observations carried out in the preceding spring. Hillslope and gully erosion remain on average comparable to the volumetric transfer of sediment observed in the preceding autumn, implying a clear connectivity.
These negative sediment balances in all sediment cascade components suggest a very high degree of connectivity between hillside and channels in autumn, and hillside fan deposits observed in early spring along low-order channel banks reflect an effective hillslope-channel coupling. This differs from effective sediment transfer occurring mostly during the summer (e.g. Berger et al., 2011;Cavalli et al., 2013).

Sediment recharge of the torrent
The sediment input, back-calculated from the in-torrent storage changes, is consistent with the net sediment output recorded from the headwater for the first two survey periods. In the torrent, the morphological monitoring that started in July revealed almost no sediment recharge (< 70 m 3 ) and is coherent with observations made in the summer in the upper catchment. The headwater sediment output must have accumulated before, probably mobilized as bedload by common runoff events in spring. In autumn, both budgets are approximately equal (1018 ± 84 m 3 against 904 m 3 ± 51), considering that few segments between both entities are missing and that both budgets were in volumetric units, despite having different sediment densities. The morphological budget indicates that the torrent experienced a net recharge in the distal part and emphasizes the clear connectivity from the production zones to the torrent, as mentioned before. In the third survey period, the headwater sediment balance indicates a net export of debris (1749 m 3 ± 199), whereas the morphological monitoring detected no significant volumes of debris entering the main torrent. Even the recharge (sediment input, Fig. 11) measured during the June debris flow events (< 600 m 3 ) remains far below the transfer of sediment recorded upstream in the headwater. This discrepancy may result from material deposition occurring in the nonmonitored segments at the headwater outlet. But field studies did not confirm this. The analysis of past series of sediment budgets performed in the upper Manival catchment V 1 = 3600 m 3 V 2 = 5000 m 3 V 3 = 10 000 m 3 V = 25 000 m 4 3 V 5 = 50 000 m 3 Volume maximum -1 -1 Figure 16. Continuous lines: erosion rate as function of size of events for a certain volume of production (potential maximum volume V 1...9 ), considering that rockfall volume distribution observed at Manival follows power law behaviour (Table 6). Dashed lines: contribution of each class of volumes to the erosion rate showing the significant effect of large slope failures. For a maximum eroded volume of 3600 m 3 yr −1 (V 1 ), the 1000 m 3 rockfall event contributes 60 %, while events less than 100 m 3 induce less than 20 % of erosion, although they are of a much higher frequency; a 100 000 m 3 rockslide would generate 70 % of total eroded material of 500 000 m 3 (V 7 ) over a century.
(Veyrat-Charvillon, 2005) reveals that the spring-early summer time currently exhibits a period of recharge following a phase of discharge within a short time lapse depending on the hydrometeorological and snowmelt conditions. The most reasonable explanation is therefore the relatively long time interval between measurements, such as the successive reworking of bedload transport suppressing the cut and fill pattern and masking the short-term behaviour of sediment transfer in the torrent. This is a well-known issue when working with channelized hillslope processes (Fuller and Marden, 2010). Although this monitoring aspect concerns the topographic changes recorded by TLS in the headwater as well, geomorphic activity, such as micro debris flows and contin-uous channel bed degradation, strongly suggests phases of sediment recharge preceding the debris flow events, which would be consistent with other studies (e.g. Brayshaw and Hassan, 2009;Marchi et al., 2002, Bennett et al., 2012.

Possible causes of seasonal fluctuations in debris supply
The Manival headwater experienced low geomorphic activity through the summer, and consequently low sediment recharge of the torrent, even though rainstorms were of sufficiently high intensity to trigger debris flows of significant magnitude in torrent. Considerations of the temporal pattern Table 6. Rock slope debris production rate estimated from the inventory analysis using power law distribution of volume for potential rockfall (Fig. 10).
Class of 10 −3 -10 −2 10 −2 -10 −1 10 −1 -1 1-10 10 1 -10 2 10 2 -10 3 10 3 -10 4 10 4 -10 5 10 5 -10 6 10 6 -10 7 volume (m 3 ) of sediment transfer and the analysis of erosion features, like alternating areas of scouring and infilling in gullies, suggest that runoff still has an important role in the headwater sediment dynamics. A clear relation between sediment transfer magnitude and precipitation remains complex, however (Fig. 3), as is often the case in mountainous catchments (Van Steijn, 1996;Bovis and Jakob, 1999;Pelfini and Santilli, 2008). The enhanced geomorphic activity observed in the hillside of several headwater subsystems, for instance during the autumn period, induced a simultaneous yet highly heterogeneous response in their channel reaches. A significant increase in bed incision and reworking similar to debris flow was observed in the upper reaches of the Manival subcatchment, implying an important sediment transfer. In contrast, the activity of other channel reaches was reduced by half, e.g. in Roche Ravine, or even remained geomorphically much less active, with only little sediment recharge. Considering that meteorological conditions were similar, this opposite behaviour may only be explained by a certain depletion of debris availability. This reduction in sediment yield can come not only within a supply-limited regime of the contributing area (Jakob et al., 2005;Glade, 2005) but also from the fact that check dams, like bedrock-dominated reaches, inhibit channel bed incision. Hence, the sediment storage has to be refilled either from the contributing hillside or from the upstream mass movement. A similar observation can be drawn from the Grosse Pierre Ravine sediment budget, whose gully downslope remained completely disconnected from the head of the subcatchment over the entire study period. Although this ravine is very steep and incises the large old rock deposits, no geomorphic work was observed, resulting most likely from the absence of debris supply from upstream. Hillside sediment delivery seems therefore to be clearly a limiting factor to sediment yield from low-to high-order channels and thus to the sediment recharge rate of the debris flow torrent downstream. As the occurrence of bedload transport and micro debris flows is controlled predominantly by the availability of sediment, even very intense rainstorm-derived runoff does not automatically lead to a significant transfer of sediment from the hillside to low-order channels in the case of material depletion.
Nevertheless, this behaviour is somehow equivocal, considering the fact that the transport capacity of ephemeral stream runoff and sheetwash related to high-intensity rainstorms is larger than the one generated by low-intensity long-duration rainfall, above all, when gully material (like in Manival) can be characterized as coarse and poorly sorted rockfall-fragment-derived debris. Lenzi et al. (2003) interpreted the annual fluctuation in sediment yield as the effect of sediment source destabilization or reactivation following a high-magnitude flow event, which facilitates material entrainment by subsequent runoff. Johnson and Warburton (2006) refer to the influence of sediment source characteristics in the control of hillslope sediment discharge. The explanation may be that the 25 August rainstorm dramatically altered the debris sources in a way that the autumn rainfalls -which, although they were of lower intensity, had a longer flood time -were able to transfer sediment downslope. Excess pore-fluid pressure in debris deposits can persist for days to weeks after sediment emplacement (Major A. Loye et al.: Headwater sediment dynamics and Iverson, 1999;Major, 2000), making debris deposits geotechnically less stable.
Although they depend on the local geomorphological setting, such as slope gradient, local topographic hollow, and degree of convergence (Reneau et al., 1990;Stock and Dietrich, 2006;Mao et al., 2009), these observations tend to show that long-lasting rainfall reduces the stability of the coarse surface layer that armours the gullies and scree slopes. This in turn affects the amount of debris supply from the hillside, despite the flow capacity and sediment availability.

Conclusions
This investigation of a yearly pattern of sediment dynamics underlines the fact that the seasonal cycle of sediment discharge from the headwater supplying the Manival torrent with debris consisted of two phases of recharge: one phase in early spring, linked to enhanced debris production and runoff conditions, and a second phase in autumn, during long periods of rainfall. Furthermore, the occurrence of the debris flow events was conditional on a net sediment delivery toward the torrent.
Overall, the torrent effectiveness seems to be controlled early in the year, from winter to spring, by sediment production and later in the year by the ability of hydrological effects to weaken the remnant debris sources, with debris availability being only one of the limiting factors at the Manival torrent. The rate of sediment delivery, directly recharging both hillside and low-order channels, is controlled by high-magnitude slope failure of moderate frequency which occurred mostly during winter time. Consequently, material re-entrainment concentrates locally in specific tributary gullies. The delivery of sediment to the torrent may be related to the hydrometeorological conditions since the last rainstorm rather than to flow capacity directly. Low-order reaches contribute significantly to the sediment delivery mechanism of the catchment headwater by controlling storage and routing processes. Hence, the recharge threshold required for a new debris flow to occur at the Manival depends primarily on the short-term debris supply, partly derived from the rate of rock slope sediment production and partly derived from mobilizing debris on the hillside. The rate of sediment recharge in the torrent is, however, greatly intermittent, since production and entrainment are both highly stochastic processes. This regime of headwater sediment delivery may have been identified in other nearby mountain environments, but very little literature exists (Alvarez and Garcia Ruiz, 2000;Veyrat-Charvillon, 2005;Berger et al., 2011) that has explored the timescale of sediment discharge in sufficient detail, e.g. on a seasonal basis.
Debris flow magnitudes have so far been mostly determined based on volume estimates derived from past events, reducing the susceptibility analysis to the known history. Monitoring of the in-storage changes within the torrent linked to the debris supply can help to improve knowledge on the recharge threshold leading to debris flow activity and therefore on their prediction. According to the rock slope production observed in this study, 10 000 m 3 yr −1 of debris supplying the headwater channels can be expected in Manival over a century. Despite the multiplicity of sediment sources and the mode of transfer operating on different spatial and temporal scales, the pattern of processes governing the sediment dynamics can be considered precisely on a seasonal basis using TLS techniques. Therefore, maximum sediment discharges from the torrent system can be specified. Without direct measurement of the rate of sediment flux and of the coupling between hillslope and channel processes, this cannot be rigorously determined. The timing of sediment budget monitoring is, however, a crucial aspect for their later interpretation.