Received 1 May 2004; accepted 6 April 2005

Transcription

1 Sedimentary Geology 179 (2005) Comparison of spatio temporal evolution of experimental particulate gravity flows at two different initial concentrations, based on velocity, grain size and density data C.M.A. Choux a, *, J.H. Baas a, W.D. McCaffrey a, P.D.W. Haughton b a Earth Sciences, School of Earth and Environment, University of Leeds, Leeds, LS2 9JT, UK b Department of Geology, University College Dublin, Belfield, Dublin 4, Ireland Received 1 May 2004; accepted 6 April 2005 Abstract Flume experiments were conducted to investigate the spatio temporal structure of subaqueous particulate gravity flows with an initial concentration of 14% by volume. Time series of downstream flow velocity and its calculated degree of turbulence, median grain size and sediment concentration at different positions along the path of nominally identical flows are analysed and combined to constrain the spatio temporal evolution of a single idealised flow. Comparison of the 14% flow with a flow of 5% initial concentration reveals similarities in the basic spatio temporal structure of velocity, turbulence, grain size and concentration. Both flow types exhibit a velocity maximum at about 1/3 of the flow height above the flume floor. At that level, velocity decreases slowly in the flows body and more rapidly in their tails. Moreover, turbulence intensity is highest in the head and at the base of the flows, whereas the level of maximum velocity and the tail of the flows typically are weakly turbulent. The zones of high turbulence are associated with shear at the front and base of the gravity flows. The flow of 5% and 14% initial concentration also agree in stratification patterns of median grain size and concentration. Grain populations are relatively well mixed in the head, show normal grading in the main part of the body and normal to inverse grading in the rear of the body and tail. The inverse grading is thought to originate from particles transported from the head upward and backward into the body of the flows, where they subsequently settle. The main difference between the flow of 5% and 14% initial concentration is that the higher-density flows appear to develop from a jet into a turbidity current closer to the inception point than the lower-density flow. This difference is interpreted from dimensionless vertical profiles of the flow parameters: horizontal velocity, concentration and grain size distribution. In the turbidity current phase of both flows, the dimensionless variables collapse well. This indicates that the flows behave in a dynamically similar manner and inspires confidence that the dimensionless variables can be used to predict the dynamic behaviour of particulate gravity flows across the measured concentration range in the flume, which due to dilution/sedimentation effects, was from ~7 to b1 vol.% concentration. D 2005 Elsevier B.V. All rights reserved. Keywords: Turbidity current; Flume experiments; Horizontal velocity; Root-Mean-Square velocity; Concentration; Grain size * Corresponding author. address: choux@earth.leeds.ac.uk (C.M.A. Choux) /$ - see front matter D 2005 Elsevier B.V. All rights reserved. doi: /j.sedgeo

2 50 C.M.A. Choux et al. / Sedimentary Geology 179 (2005) Introduction Particulate gravity currents, both subaerial and subaqueous, exhibit a wide range of density. Turbidity currents are an intrinsic part of the spectrum of subaqueous sediment gravity flows. Although they too encompass a wide range of concentration, for several decades now, turbidity currents have been consistently described as sediment-laden gravity-driven flows in which the sediment is supported principally by fluid turbulence (Sanders, 1965; Middleton and Hampton, 1973; Middleton, 1993; Simpson, 1997; Shanmugam, 1997). Nevertheless, grain support mechanisms other than fluid turbulence may co-occur in turbidity currents, such as hindered settling, particle particle interactions and buoyancy enhancement (Hiscott, 1994; Mulder and Alexander, 2001; also see reviews by Stow et al., 1996 and Kneller and Buckee, 2000). The contribution of these mechanisms to grain support is highly dependent on the local concentration of suspended sediment in the flow, and thus the relative importance of these mechanisms may change if flows change their concentration structure as they develop. In most cases, turbidity currents do indeed evolve in concentration as they flow, either through sediment erosion and entrainment (e.g., Pantin, 1979), or through deposition and entrainment or detrainment of ambient water (Simpson, 1997 and references therein). This triggers the question of how flows of differing initial concentration compare in terms of internal grain size distribution, concentration and velocity structure as they develop. Do initially dense, depositional flows, propagating for sufficient time to become dilute, show the same basic dynamical behaviour as initially dilute flows? Are high concentration turbulent flows viable as a long range transport mechanism, or are high concentrations only developed transiently, during sediment entrainment and/or deposition? A large volume of experimental work has been undertaken in order to analyse the role of particle concentration in turbidity current behaviour and structure, as well as the geometry and internal structure of their deposits (e.g., Kuenen, 1966; Middleton, 1967; Britter and Simpson, 1978; Lüthi, 1980; Laval et al., 1988; Middleton and Neal, 1989; Altinakar et al., 1990; Bonnecaze et al., 1993; Garcia and Parsons, 1996; Gladstone et al., 1998; Hallworth and Huppert, 1998; Kneller et al., 1999; Stix, 2001; Choux and Druitt, 2002; McCaffrey et al., 2003; Baas et al., 2004; Al-Ja Aidi et al., 2004; see also reviews by Edwards, 1993; Middleton, 1993; Kneller and Buckee, 2000, and Shanmugam, 2000). Despite the value of these experimental works, the results cannot be used directly to answer questions regarding the spatio temporal evolution of natural particulate gravity currents. This requires a detailed characterisation of both the spatial and temporal evolution of the properties of experimental particle-driven flows in terms of velocity, granulometric and concentration structure, across a range of concentrations. Until recently, however, only temporal (time series) data were collected (see discussion in Peakall et al., 2001), with experiments therefore focussing on flow unsteadiness rather than flow non-uniformity. Best et al. (2001) used 4-MHz ultrasonic Doppler velocity profiling (UDVP) to quantify, for the first time, the spatial and temporal evolution of mean flow and turbulence structure of sediment-laden particulate flows. However, concentration and grain size data were not collected, and the length over which flow evolution was characterised (up to 85 mm) was relatively small compared to the scale of the flows. By coupling the UDVP method with a siphoning technique and sampling several identical depositing flows at different locations, McCaffrey et al. (2003) were the first to produce a detailed description of the spatio temporal evolution of a particulate current of 5% initial concentration in terms of instantaneous velocity, grain size and concentration. Their study was limited to flows of a single initial starting concentration. Computer modellers also endeavour to shed new lights on describing turbidity current structure (e.g., Stacey and Bowen, 1988; Zeng and Lowe, 1997; Mulder et al., 1998). With the help of a non-depth averaged model, using a multiphase flow approach, and particle particle interaction, incorporating the turbulence model of Mellor and Yamada (1982), Felix (2001, 2002) produced a 2-D, vertical plane, numerical model that simulates unsteady flow behaviour in terms of velocity, turbulence, grain size and concentration distribution. However, to date his results have been tested only against a small number of incomplete historical flow data, because of the lack of suitable experimental data (Felix, 2002).

3 C.M.A. Choux et al. / Sedimentary Geology 179 (2005) The experimental data of McCaffrey et al. (2003) can only be used to validate numerical models of flows in which particle particle interactions can be ignored, as the initial particle concentration (5% by volume) was below the 9 vol.% threshold proposed by Bagnold (1966) above which these effects become significant. In the present paper, a new set of experiments is presented, using the same experimental setting as McCaffrey et al. (2003), but for a higher initial concentration (14%), at which particle particle interactions should influence flow behaviour, via moderate turbulence suppression (Middleton and Hampton, 1976; Lowe, 1982). Thus the aim of this work is to enable the spatio temporal evolution of the internal structure of relatively high concentration turbidity currents (14%, hereafter referred to as high-density flow) to be compared with that of more dilute flows (5%, hereafter referred to as low-density flow) in terms of the vertical gradients in instantaneous horizontal velocity, grain size and concentration distribution, as well as turbulence structure. Accordingly, the experimental set up of McCaffrey et al. (2003) are described, and the key results are summarised. The flow structure of a turbidity current with 14% initial sediment concentration is then detailed and compared with the 5% flows results of McCaffrey et al. (2003). Subsequently, original analysis of the turbulence structure of flows of both 5% and 14% initial concentration (as expressed by root-meansquare RMS velocities) is presented. Dimensionless parameters are established with the aim of comparing the low- and high-density flows in more detail and expanding the results to a wider range of initial sediment concentrations. 2. Previous related work Based on experimental data, the approach of McCaffrey et al. (2003) allowed for the first time, the structure of a turbidity current to be reconstructed at any position and time, for the parameters streamwise velocity, grain size and suspended sediment concentration. In a series of flume experiments, McCaffrey et al. (2003) generated subaqueous particulate gravity flows through release of a 30 l suspension of non-cohesive material (silica flour) at an initial concentration of 5% by volume. They measured simultaneously the temporal evolution of the vertical stratification in streamwise velocity, flow concentration and grain size distribution as the entire flow passed a measurement location. A series of five nominally identical flows were run, with measurements repeated at five different locations along the flow path. The results were then combined to constrain the spatio temporal evolution of a single idealised flow. The inbound jet transformed into a gravitydriven current at a distance between 1.32 and 2.64 m from the reservoir, and thereafter developed under its own internal action. The experimental depositional particulate gravity currents of McCaffrey et al. (2003) were non-uniform, i.e., their structure varied spatially (see Allen, 1985 and Kneller and Branney, 1995), indicating that it would be erroneous to interpret time series data of such flows in terms of longitudinal flow structure, as commonly done in the existing literature. For example, the velocity data showed that the flow duration increased downstream, as the flow stretched out. The concentration data showed that in proximal locations, the rate of decrease of concentration was high, indicating rapid sedimentation, whereas in distal locations the rate of decrease was more gradual. The transition between the head and the body of each nominally identical turbidity current was described by a sharp decrease in the maximum velocity and median grain size, whereas the transition between the body and the tail was well defined by a decrease in the concentration. The velocity maximum was located at approximately one third of the flow s height from its base. An interesting normal to inverse vertical pattern in grading observed in the grain size distribution of material suspended in the flow s body was linked to the presence of coarse sediment inferred to have been swept upwards and backwards over the head then falling passively into the upper part of the flow. 3. Experimental set up The experimental set up in this study was the same as that used by McCaffrey et al. (2003), with the sole difference that the concentration of the initial suspension was increased from 5% to 14% by volume (i.e., with an initial suspension density of 1231 kg m 3

4 52 C.M.A. Choux et al. / Sedimentary Geology 179 (2005) compared to 1082 kg m 3 ). The flume in which the five nominally identical flows were run was 10 m long, 0.3 m wide and 0.3 m deep (Best et al., 2001), with an overhead reservoir containing 30 l of suspension (Fig. 1a). A homogeneous suspension in water of silica flour particles (density: 2650 kg m 3 ) was created in the reservoir, and kept well mixed by a mechanical stirrer. The particle size ranged from b0.01 to c 60 microns, with a median grain size (D50) of about 8 Am (Fig. 1b). At the start of each experiment, the stirrer was stopped and a sealing stopper at the bottom of the reservoir was removed swiftly. Time series of the low-density experiments of McCaffrey et al. (2003) taken 0.04 m downstream from the reservoir outlet (their Flows 1 and 2, illustrated in their Fig. 4) showed that the inbound flows were steady in terms of velocity, grain size and concentration for 21.5 s before swiftly decelerating. From this it may be inferred that the suspension in the reservoir remained essentially uniform as it drained. This is probably due to the inherited turbulence from the mixer, plus any turbulence generated by shear against the reservoir walls as the suspension flowed out and into the flume. Although similar outlet time series were not collected for the high-density experiments reported here, it is inferred that the input to the flume was essentially steady in this case too. The suspension drained into the water-filled flume through a circular pipe of m diameter, emptying the reservoir in about 21.5 s at constant discharge (cf., McCaffrey et al., 2003) and forming a particulate gravity current that propagated along the length of the flume. The flow was sampled by an array of instruments all positioned at the same location (Figs. 1 and 2). Due to the intrusiveness of the data acquisition method, each flow could be measured at one location only. Thus the reservoir was shifted upstream by an interval distance of 1.32 m between each of the five nominally identical flows, increasing the distance between the entry point of the flow and the measurement point (Fig. 1a). The height of the flow s head was fairly constant at about m, when reaching the sampling devices, at all locations. Upward flow motion was observed in the head of the flow as the flow propagated; this upward moving fluid was then forced back to horizontal by the ambient water swept over the front of the head (Fig. 2). At Location 1, the Reynolds number, calculated using average values within the head, was 210 4, and so was well within the turbulent regime. Data were acquired from the time of flow inception until upstreampropagating, solitary waves (e.g., Pantin and Leeder, 1987; Edwards, 1993), generated by the reflection of the inbound flow from a distal overflow weir (Fig. 1) passed the array of instruments. Two sets of instruments were used at each measurement location. A vertical array of six 4-MHz UDVP (Ultrasonic Doppler Velocity Profiler) probes, positioned at 6, 16, 26, 36, 46 and 76 mm above the bottom of the flume, recorded the streamwise component of flow velocity upstream of the probes, following the technique described by Best et al. (2001). The sampling rate of each UDVP probe was 4.5 Hz. Located at the same height (except for 76 mm) and adjacent to the UDVP probes were 5 siphoning tubes of 6 mm diameter, which continuously sampled the flow as it passed by. The suspension samples were collected continuously in 5 aligned rows of 20 sample containers, one row for each of the 5 siphoning tubes. The 100 container array was set up on a sliding trolley, which was rapidly advanced below the outflow ends of the siphon tubes at 4 s intervals, allowing each successive column of 5 beakers to fill synchronously over ~4 s. The tubes were 1.2 m long, and Fig. 1. a) Experimental set up and b) particle size distribution.

5 C.M.A. Choux et al. / Sedimentary Geology 179 (2005) Fig. 2. Frame captured from the video recording of the turbidity current of 14% initial concentration at Location 4, i.e., at 5.28 m from the inlet. 1, 2, 3, 4, 5 refers to siphon and UDVP probe positions, located at respectively 6, 16, 26, 36, 46 mm heights. 6 refers to UDVP probe only, located at 76 mm height. The arrow points to the turbulent eddy, which generates a rapid deceleration event in the velocity time series. The field of view is approximately 0.55 m long and 0.35 m high. allowance was made for the measured transit times when registering the timing of sample collection with that of UDVP data collection The content of each container was later analysed using a Malvern Mastersizer Plus laser diffraction grain sizer, yielding grain size distribution and suspended sediment concentration information at a rate of 0.25 Hz. We will show below (Section 5.1) that the duration of the head, delimitated by the horizontal velocity data, is 4 s for the low-density experiments, i.e. the sampling duration, and about 2.5 s for the high-density experiments. Care was taken that as little ambient fluid as possible was collected prior to arrival of sediment-bearing fluid at the siphon outlet. This implies that for the highdensity flows, the sampling beaker of the head will have incorporated some body material for up to 1.5 s. However, we will show, in Section 6.3, that the dimensionless concentration and grain size data in the head and body are roughly similar; hence we assume that any effect of dilution from the body material into the head were insignificant. A detailed study of the spatio temporal evolution of the downstream velocity, grain size distribution and concentration of the flow was feasible. Moreover, additional information on the spatio temporal evolution of the downstream component of turbulence of the flow was obtained from the calculation of root-meansquare (RMS) values of downstream velocity. RMS velocity is equal to the standard deviation of velocity averaged over a certain time period (Kneller et al., 1997; Buckee et al., 2001; Baas and Best, 2002). Time series of RMS velocity were calculated by averaging the instantaneous velocity data over a 2-s long period along the whole duration of the velocity time series and for each measurement height. Tests carried out to verify the effect of other lengths of averaging periods on the time series showed no significant differences. In order to remove the unwanted effect on RMS velocity of long-term flow deceleration, the velocity profile in each time window was de-trended using standard linear regression analysis prior to calculating the RMS velocities. In analysing velocity signals from UDVP probes in still water, the long-term average deviation from the mean was found to be between 2 and 3 mm s 1. These values are therefore considered a threshold value for the UDVP instrument noise. Areas of the experimental flows with RMS values below 3 mm s 1 need not be caused by turbulence.

6 54 C.M.A. Choux et al. / Sedimentary Geology 179 (2005) Description of 14% experimental data The time series of downstream velocity, grain size, and concentration are presented for each sampling location in Fig. 3. This figure also illustrates the calculated root-mean-square (RMS) values of the downstream velocity data (Fig. 3b). The reference time is taken as the time from the removal of the reservoir stopper. The flow arrival time ranges from 7 s at Location 1 to 35 s at Location Downstream horizontal velocity data The six UDVP probes acquired time series of the downstream horizontal velocity as the flow travelled by. The temporal evolution of the streamwise velocity field for each location is shown in Fig. 3a. A coherent structure that evolves slightly between the different locations is observed. A zone in which the velocity values are consistently high (up to 265 mm s 1 ) through the flow depth is recorded for 2 3 s after the arrival of the flow front (Fig. 3a). Thereafter the velocity drops at each measurement height. The flow deceleration is very rapid for the upper two probes (at heights 46 and 76 mm), forming a zone with velocities as low as 40 mm s 1. The velocity decrease is reduced for the lowermost probe (6 mm high), with velocities reduced by a third of the maximum value at that level. The probes at heights of 16, 26 and 36 mm reveal a zone of high velocity, with the maximum velocity occurring close to 20 mm. The height of this interpolated maximum does not vary temporally, and varies only slightly spatially (+/ 2 mm;fig. 3a). However, the internal structure of this zone does evolve with time and distance. At Location 1, the time series exhibits rapidly fluctuating flow velocity for about 30 s after passage of the head. These fluctuations become less conspicuous at intermediate locations and disappear distally (e.g., Location 5 in Fig. 3a). The flow duration, measured as the time taken for the downstream velocity to fall below a predefined reference value (cf., McCaffrey et al., 2003), progressively increases as the flow propagates downstream. For example, using a reference velocity of 100 mm s 1, duration increases from 28 s at Location 1 to 32 s at Location RMS downstream velocity data The time series of the RMS of downstream velocity is shown in Fig. 3b. All graphs exhibit short-term changes in RMS velocity, which take the form of small concentric structures in the contoured plots presented. The highest RMS values were found immediately after flow arrival and for a couple of seconds only, for the highest UDVP probes. Rapid fluctuations in RMS values are also found close to the bed at 6 mm. The periodicity of these fluctuations is ~3 s. The time span between the passage of successive zones of high RMS velocity remains quasi-constant from Location 1 to Location 5 whilst the RMS values decrease. Above the velocity maximum, between 25 and 50 mm, and after the zone of highest RMS values, a zone of intermediate RMS velocity exists (Fig. 3b). RMS values are lowest at the level of the maximum velocity and particularly during the last s of the time series. The most variable RMS time series is observed at Location Grain size data At all locations along the flow path, median grain size evolves in a temporally and spatially consistent way (Fig. 3c). At each location, and during the first 5 s after the flow arrival, the grain size data exhibit enrichment in coarse grains relative to the initial particle distribution. Thereafter, the vertical grain size profile is characterised by an upward decrease of the median grain size for ~25 s at Location 1, ~20 s at Location 2 and ~15 s at Location 3. Subsequently, the normal grading changes into a characteristic vertical pattern of normal to inverse grading for these locations. The normal to inverse grading is particularly well-developed at Locations 4 and 5. At each location, the zone of grain size reversal moves closer to the base of the flow with time. A period during which the flow carries relatively coarse grains is seen near the base of the flow at about 10 s after flow arrival at Location 1 and at s after flow arrival at the other locations (Fig. 3c). During these periods, the flow contains the coarsest sediment measured, with D 50 -values of up to 7.7 Am (Location 1).

7 Fig. 3. Time series of a) downstream velocity (millimeter per second), b) calculated root-mean-square (RMS) values of downstream velocity (millimeter per second), c) median grain size (micron), and d) concentration (volume percent), at six different flow heights 6, 16, 26, 36, 46 and 76 mm, for five different measurement locations, for the flows with 14% initial concentration. Note that the 76 mm time series is only collected for downstream velocity. The time, in seconds, is expressed from the removal of the stopper from the bottom of the reservoir, i.e. the time of inception of the flows. For each graph, the two dashed lines mark the position of the uppermost and lowermost probes, above which and below which no more data are acquired. The scales and grey shades are the same as in Fig. 4, for easier comparison. C.M.A. Choux et al. / Sedimentary Geology 179 (2005)

8 56 C.M.A. Choux et al. / Sedimentary Geology 179 (2005) Fig. 4. Time series of a) downstream velocity (millimeter per second), b) calculated root-mean-square (RMS) values of downstream velocity (millimeter per second), c) median grain size (micron), and d) concentration (volume percent), at six different flow heights 6, 16, 26, 36, 46 and 76 mm, for five different measurement locations, and for the flows with 5% initial concentration. All data, except RMS velocities, were presented in McCaffrey et al. (2003), but have been redrawn at the same scale and grey shades as in Fig. 3. The time, in seconds, is expressed from the removal of the stopper from the bottom of the reservoir, i.e. the time of inception of the flows. For each graph, the two dashed lines mark the position of the upper and lower probes, above which and below which data extrapolation is likely to be biased.

9 C.M.A. Choux et al. / Sedimentary Geology 179 (2005) Concentration data The time series of sediment concentration as a function of flow height are given in Fig. 3d. Concentration consistently decreases upwards. At each measurement location, the concentration values are initially very low (down to 2 3 vol.%) compared with the initial concentration of 14 vol.% within the reservoir. The maximum concentration, located at the base of the flows, occurs at each location some s after passage of the flow front. The maximum concentration at Location 1, which is closest to the inlet, equals 6.4%. At other locations, the maximum measured concentration ranges from 5.9% at Location 2 to 7% at Location 5, i.e., furthest away from the inlet. The interpolated near-bed concentrations values are 6.4% at Location 2 and 7.9% at Location 5. The maximum heights reached by concentration contours V 4.5% gradually decrease in a downstream direction. 5. Interpretation of high-density flow data and comparison with low-density flow data In this section, the data of the high-density experiments are interpreted and compared with the lowdensity experiments of McCaffrey et al. (2003). The spatio temporal graphs of the low-density experiments are reproduced in Fig. 4 to facilitate the comparison. Fig. 4 also includes time series for RMS values of downstream velocity, which have not been published before. Due to the fact that the low-density flows were slower than the high-density flows, the low-density flows arrived later at each measurement location than the high-density flows, explaining the different initial times on the graphs in Figs. 3 and 4. Particulate gravity currents are commonly divided into three flow regions: head, body and tail (see review by Kneller and Buckee, 2000). The head, with its overhanging nose due to no-slip condition at the lower boundary and frictional resistance at the upper boundary, is the area where mixing of the current with ambient fluid occurs, essentially by detraining of dense fluid out of the back of the head in a series of transverse vortices (Allen, 1971; Britter and Simpson, 1978; Simpson and Britter, 1979). The body is the area which has a thin, relatively dense layer of fluid near the base of the current, and which is overlain by a mixing zone at its upper boundary displaying succession of large eddies (Ellison and Turner, 1959; Middleton, 1966). The tail is the terminal part of the flow, where velocity is low and gradually decreases to zero; here slow settling from suspension is the dominant depositional process. The same subdivision is applied below, because it was possible to confidently delimit head, body and tail by trends in the experimental data (see also McCaffrey et al., 2003) Downstream velocity The head of the flow with 14% initial concentration is delimited by the rapid increase in velocity at the flow front (Fig. 3a) and the midpoint of a slightly longer period of relatively strong flow deceleration present at heights of 46 and 76 mm in all locations. The head passes the measurement locations in about 2 3 s. The strong deceleration may relate to the presence of an eddy at the back of the head; analysis of video recordings of the experiments confirms the existence of such a structure, located at the back of, and defining the extension of the head (Fig. 4). A more gradual flow deceleration event at s, prominent in particular at proximal locations, defines the transition between the body and tail of the flow. It corresponds to the beginning of the waning tail of the flow, probably because the reservoir empties and thus no longer supplies the flow (the reservoir was seen to empty at ~21.5 s after the start of the experiments). The velocity structure of the high-density flow evolves in time and with distance along the flume. As a result of the absence of a bed slope and progressive deposition of sediment, flow velocity gradually decreases at all levels in the flow and at all locations along the flume. Also, the short-term fluctuations in flow velocity, which were particularly clear at Location 1, gradually disappear. The increasing duration of the flows away from the inlet is primarily caused by extension of the tail away from the inlet, because the duration of the passage of the head and body is almost constant at the studied locations. In general terms, the high- and low-density flows have similar time-dependent velocity structure. In detail, however, there are important differences. The dense flow is thinner and travels faster than the dilute flow, with a shorter head duration (about 2.5 s instead

10 58 C.M.A. Choux et al. / Sedimentary Geology 179 (2005) of 4 s for the low-density flows). The height of maximum velocity is lower for the high-density flow (c20 mm) than for the low-density flow (c25 mm). However, the velocity maximum is located roughly at about 0.3 of the height of the head in each case, which is in agreement with previous experiments (Altinakar et al., 1996; Kneller et al., 1997, 1999; Best et al., 2001) RMS downstream velocity The time series of RMS downstream velocity, shown in Figs. 3b and 4b, indicates that, overall, the flows of 14% initial concentration have higher RMS values than the flows of 5% initial concentration, although the distribution of RMS values is similar. Below, the RMS velocity structure of the low-density flows is described first, then the RMS velocities of the low- and high-density flows are compared. All the graphs of RMS velocity for the low-density flows exhibit numerous rapid fluctuations in the head of the flows as well as close to their base (Fig. 4b). The highest RMS values are obtained from the upper front of the head at Location 2. Locations 3 5 have their maximum RMS velocities at the same position, but absolute values decrease downstream. Location 1 is characterised by strong fluctuations in RMS velocity in the entire head and body. The rear part of the flows and the upper part of their bodies are characterised by low RMS velocities (Fig. 4b). Particularly striking are the regular fluctuations in RMS values near the base of the flow. As for the high-density flows, a zone of relatively low RMS values exists at the level of the velocity maximum, above which intermediate fluctuating RMS values are observed. In both the low- and high-density flows, the zone of maximum RMS velocity within the upper part of the head is interpreted to result from high turbulence levels linked to the friction between the propagating flow and the ambient fluid, leading to the formation of Kelvin Helmholtz waves and mixing at the back of the head (Best et al., 2001). The concentric RMS structures near the base of the flows, whose periodicity slightly increases with time, are interpreted as coherent flow structures (Baas et al., in press), corresponding to turbulent eddies generated by friction at the lower flow boundary. The relatively low RMS velocities at the level of maximum velocity, supporting previous measurement by Kneller et al. (1999), Best et al. (2001) and Buckee et al. (2001) as well as in the tail of the flows indicate that these areas are less turbulent than other areas. It thus appears that the propagation distance of turbulent eddies generated by shear at the lower, upper and frontal flow boundaries is relatively small (cf. Felix et al., 2005). The degree of turbulence decreases downstream along the flume in both flows, which correlates with decreasing downstream velocity and thus decreasing shear Median grain size The median grain size structures of the high-density flows (Fig. 3c) and the low-density flows (Fig. 4c) evolve in a similar way. Yet, the basal zone of maximum grain size is less well developed in the lowdensity flows. At each location, the grain size helps to define the transition between the head and the body of the flow. A drop in median grain size by up to 1 Am marks this transition. The inverse grading in the body of the flows is interpreted to result from the movement of coarse sediment from the upper part of the head, enriched in coarse particles (cf., Section 4.3), upwards and backwards by turbulent motion (McCaffrey et al., 2003). During the backward motion, the coarse sediment is probably located above the measurement area, i.e. above the highest sampling tube located at 46 mm. They then fall passively back into the body and tail of the flow. Video data (Fig. 2) and velocity (Figs. 3a and 4a) support the interpretation that an eddy is present at the back of the head, which may be responsible for this redistribution of coarse sediment towards the rear of the flows. A major difference between the low- and highdensity flows is that a basal zone enriched in coarse grains is present at s after the passage of the head at Locations 2 5 in the high-density experiments (Fig. 3c). This phenomenon was observed only at Locations 3 and 5 in the low-density experiments. Their position below the velocity maximum and far behind the head classifies these coarse-grained zones as coarse tail lags sensu Hand (1997). The fastest settling grains are thus concentrated towards the base of the flow, while slower settling grains are distributed more evenly throughout the flow depth

11 Fig. 5. Spatio temporal evolution of a single idealised flow, created by using the data acquired at the five measurement locations, for a) downstream horizontal velocity (millimeter per second), b) RMS velocity (millimeter per second), c) median grain size (micron), and d) concentration (volume percent), at 21.5, 29, 36.5, 44 and 53.5 s after inception of the flows. The time, in seconds, is expressed from the removal of the stopper from the bottom of the reservoir, i.e. the time of inception of the flows. The grey dashed lines show the position of the uppermost and lowermost probes respectively above which and below which no more data are acquired. A small contouring artefact is noticeable before the flow arrival, mainly visible for the downstream horizontal velocity and RMS velocity graphs, at 21.5 and 29 s. C.M.A. Choux et al. / Sedimentary Geology 179 (2005)

12 60 C.M.A. Choux et al. / Sedimentary Geology 179 (2005) (Middleton and Southard, 1984; Hand, 1997 and references therein) Concentration The initial sediment concentration in the reservoir was 14%. The fact that a maximum concentration of 6.4% was observed at Location 1 (Fig. 3d) indicates that strong flow dilution occurred due to flow expansion and entrainment of ambient water and/or sedimentation. In the flows of 5% initial concentration, McCaffrey et al. (2003) also noticed an abrupt change in concentration between Locations 1 and 2 (Fig. 4d). They explained the reduction of nearly 50% by invoking high rates of sedimentation from suspension between those locations. In the high-density experiments (Fig. 3d), no such drastic change is observed. This point will be discussed in more detail in the section on dimensionless analysis below. The progressive spatio temporal decrease in the height of concentration contours, from Locations 1 to 5, attests to ongoing sedimentation and dilution as the flow propagates, although these processes are less marked than for the low-density flows. At Location 5, the maximum measured concentration of 7% by volume, seen at the base of the flow at around 55 s, is higher than the corresponding maximum seen in Location 4, immediately upstream, at around 45 s, which is equal to 6% by volume, interrupting the overall pattern of downstream-decreasing concentration. A possible explanation is that the flow undergoes a slight increase in the rate of fallout of sediment from suspension between Locations 4 and 5 due to decreasing velocity and turbulence intensity Spatial flow evolution A series of instantaneous snapshots of the highdensity flow was constructed at five selected times (i.e., 21.5, 29, 36.5, 44 and 53.5 s), for downstream velocity (Fig. 5a), RMS velocity (Fig. 5b), median grain size (Fig. 5c) and concentration (Fig. 5d). This was done by extracting the respective data for each point in time and for each measurement location from the time series, and then plotting the data as a function of distance along the flume for each point in time. These spatial plots permit the visualisation of the internal structure of a single flow and its temporal evolution. They would also allow for a direct comparison with numerical modelling results, as suggested by Felix (2002). The snapshots of downstream velocity (Fig. 5a) show maximum values at the front of the head and progressively slower flow with increasing distance behind it. The decrease in velocity in the tail part is particularly evident around the height of maximum velocity. The corresponding snapshots of RMS velocity data (Fig. 5b) do not show a steady evolution. Generally, the head of the flow has the highest RMS values, and therefore is the most turbulent part of the flow. In the body, RMS velocities are clearly less than in the head, except for the basal part of the flow, where turbulence remains strong even at large distances behind the flow front. The spatial plots of median grain size (Fig. 5c) show that the zone of minimum grain size becomes more pronounced as the flow evolves temporally and that coarse tail lagging occurs behind the head of the flow (e.g., at 44s). The spatial plots of concentration reveal a higher rate of sedimentation as time goes by. Indeed at 29 s, the height between the 2% and 6% concentration contours is about 40 mm whereas it is only 20 mm at 53.5 s. A zone of maximum concentration is visible close to the base of the flow (Fig. 5d). It moves progressively down the flume as the flow evolves. The zone is centred at ~1 m after 29 s, at 2.5 m after 36.5 s, at 6.5 m after 44 s and beyond 7 m after 53.5 s. The temporal evolution of the internal structure of the high-density flow seen in Fig. 5 for all measured flow and sediment parameters, indicates that the flow is non-uniform. Decreasing velocities and overall concentrations indicate that this is probably caused by the flow s depositional character. This interpretation reinforces the conclusions drawn by McCaffrey et al. (2003) and extends them to flows of higher concentration. The fact that the low- and high-density flows are non-uniform implies that it is impossible to deduce the structure of flows from studies of time series data obtained at only one location, at least over the concentration range encompassed by the flows as they evolve. 6. Dimensionless analysis Dimensionless analysis was carried out in order to compare the flow structure of the low- and high-

13 C.M.A. Choux et al. / Sedimentary Geology 179 (2005) density flows in detail, and to investigate if data collapse could be achieved. Here, the flows are compared by means of dimensionless vertical profiles of normalised downstream velocity (Fig. 6), RMS velocity (Fig. 7), median grain size (Fig. 8) and concentration (Fig. 9). A careful choice of normalisation parameters is essential in order to ensure that flow and sediment parameters are compared in analogous zones of the flows (cf., Felix, 2004). Therefore, the dimensionless parameters used in this study are defined first. Thereafter, the location of vertical profiles in the head, body and tail of the turbidity currents are selected. The dimensionless profiles of flow and sediment parameters are presented in the last part of this section Normalisation parameters Selection of reference heights, velocities, median grain sizes and concentrations was necessary in order to normalise and compare the data from the two sets of experiments. The height of the maximum downstream velocity was used as reference for the calculation of normalised height (cf., Altinakar et al., 1996; Kneller et al., 1999). This reference height, which was shown to be independent of measurement location, is ~25 mm for the low-density flows (Fig. 4a) and ~20 mm for the high-density flows (Fig. 3a). Other reference heights, such as the height at which the downstream velocity in the upper part of the flow is half the maximum velocity (e.g., Kneller and Buckee, 2000), could not be reliably applied because there were too few sampling heights above the velocity maximum to accurately determine them. In the vertical profiles for the head (Fig. 6a) and body (Fig. 6b), dimensionless downstream flow velocity was defined as the ratio between average velocity over a time span of 1.25 s (flow of 14% initial concentration) or 2 s (flow of 5% initial concentration) and average head velocity for all measurement Fig. 6. Dimensionless flow velocity as a function of dimensionless height and measurement location for the head (a), body (b) and tail (c) of the low- and high-density (5% and 14% initial concentration, respectively) turbidity currents.

14 62 C.M.A. Choux et al. / Sedimentary Geology 179 (2005) Fig. 7. Dimensionless RMS velocity as a function of dimensionless height and measurement location for the head (a), body (b) and tail (c) of the low- and high-density (5% and 14% initial concentration, respectively) turbidity currents. locations. The time spans of 1.25 and 2 s were chosen to obtain velocity values over approximately equal lengths in both flows. Average downstream head velocities, at the height of the velocity maximum, were and 189 mm s 1 for the low- and high-density flows, respectively. The use of a single head velocity for each flow concentration is warranted, because head velocity changes between the most proximal and most distal measurement locations were insignificant. The calculation of the dimensionless flow velocity for the tail zone (Fig. 6c) was carried out using the maximum velocity found at the body tail transition because the head and tail are sufficiently far apart that processes in the head may not be relevant to tail dynamics. The dimensionless RMS velocity for the head (Fig. 7a) and body (Fig. 7b) was calculated by normalising the RMS velocity values to the average head velocity. For the tail (Fig. 7c), the same method was used as for the normalised downstream velocity, i.e., the maximum downstream value measured at the body tail transition was selected. Median grain size (Fig. 8) was normalised to the initial median grain size (8 Am) in the overhead reservoir for the low- and high-density flows. Dimensionless concentrations (Fig. 9) were calculated by dividing the local concentration values by the initial concentration in the overhead reservoir Division of flows into head, body and tail segments Vertical profiles of flow parameters were outlined through predefined segments in the experimental flows. The underlying methodology relies on the determination of equivalent zones in the flows in each of the two sets of experiments, and subsequent selection of equivalent locations within these zones where vertical profiles are to be compared. Below, the zones are defined in terms of the flows head, body and tail.

15 C.M.A. Choux et al. / Sedimentary Geology 179 (2005) Fig. 8. Dimensionless median grain size as a function of dimensionless height and measurement location for the head (a), body (b) and tail (c) of the low- and high-density (5% and 14% initial concentration, respectively) turbidity currents Head length In the time series, the passage of the head is defined as the period immediately following the arrival of the flow, during which the velocities are high along the entire vertical profile (Fig. 3a). The upstream boundary of the head, i.e. the transition between head and body, is defined by the rapid decelerating event in the velocity time series, recorded by the probes at 46 and 76 mm height. The average period in which the head passed a measurement location was 4 s in the low-density flow and 2.5 s in the high-density flow. In length, this corresponds to about 0.38F0.005 m for both flows, using the average head velocity as reference. It was then decided arbitrarily to select a relative distance of 25% of the head length behind the front of the head to locate the vertical profiles. Thus, at m from the front of the head, the original data located along a vertical profile were chosen for normalisation. Because of the large fluctuations of the downstream and RMS velocity, instead of presenting an isolated profile from this location, average velocity values were calculated for all the values between 0 and 0.19 m (giving the average velocity value at m) Body length The body stretches from the upstream limit of the head to the point marked by a sudden change from high to low velocity (at proximal locations, Fig. 3a) and a change from relatively high to low RMS velocity (predominantly at distal locations, Fig. 3b). This is interpreted to represent the time when the overhead reservoir emptied. The reservoir emptied in 22 s for the low-density flow and 21.5 s for the high-density flow. Allowing for the head duration, the duration of the passage of the flow bodies in the low- and highdensity flows was thus 18 and 19 s, respectively, with corresponding respective body lengths of 1.73 and

16 64 C.M.A. Choux et al. / Sedimentary Geology 179 (2005) Fig. 9. Dimensionless concentration as a function of dimensionless height and measurement location for the head (a), body (b) and tail (c) of the low- and high-density (5% and 14% initial concentration, respectively) turbidity currents m (once again using average head velocity as the reference velocity). The vertical profiles for dimensionless downstream velocity and its RMS values, median grain size and concentration were drawn at an arbitrary relative distance of 30% of the length of the body from its front, hence at 9.4 s and 0.90 m for the low-density flows, and 8.2 s and 1.23 m for the high-density flows. Here, 9.4 and 8.2 s refer to the time since the arrival of the flow Tail length The tail is the most distal part of the flows bounded by the body at one end. In the experiments, the tail was disrupted by the arrival of flow reflections before it had come to rest. Because the tail is the section where most of the flow stretching takes place, a different method is required to select an equivalent location for the vertical profiles. First, a power function was fitted to the velocity time series at a height of 2.6 cm (i.e., close to the level of maximum velocity) in the tail of each flow. Subsequently, the best fit power function was used to calculate the time period from the time of first arrival of the tail to the time at which velocity reached 10 mm s 1. Finally, the time for the tail velocity to decrease by an arbitrary 40% of the range between its value at the body tail boundary and the 10 mm s 1 boundary was calculated for each measurement location. At these times vertical profiles for downstream flow velocity and RMS velocity, concentration and median grain size were determined. As mentioned above, dimensionless velocities were calculated by dividing the values at the 40% boundary by the velocity at the body tail boundary rather than by the average head velocity, because the head and tail are so far apart that head processes should not affect the tail Dimensionless vertical profiles The normalised downstream velocity (Fig. 6), RMS velocity (Fig. 7), median grain size (Fig. 8) and concentration (Fig. 9) are plotted versus the dimensionless height for each of the three zones of the flow (i.e., body, head and tail).

17 C.M.A. Choux et al. / Sedimentary Geology 179 (2005) Downstream velocity profiles For the head and body (Fig. 6a and b), normalised downstream flow velocities greater than 1 indicate flow towards the front of the turbidity current, while normalised velocities smaller than 1 signify flow that moves away from the head for an observer moving with the flow (i.e., within a Lagrangian reference frame). For the tail (Fig. 6c), velocities are relative to that of the body tail transition. The normalised downstream velocities for the low- and high-density flows at Locations 2 5 are similar for the head (Fig. 6a), body (Fig. 6b) and tail (Fig. 6c), hence the data collapse in a satisfactory manner. At Location 1, however, the normalised velocity of the high-density flow is significantly higher than that of the low-density flow (Fig. 6a), particularly within the head. Between Locations 1 and 2 the dimensionless values for the low-density experiment increase drastically, thus supporting the inferred occurrence of an episode of high sedimentation rate and change from jet to turbidity current (McCaffrey et al., 2003) between these locations. In contrast, the dimensionless velocity remains quasi-constant between Locations 1 and 5 in the high-density flow. This suggests that no regime change occurred along this transect. The flow may therefore have developed into a turbidity current by the time that it reached Location 1. In turn, this implies that any episode of high sedimentation rate must have occurred upstream of Location 1, and thus within 1.32 m of the inlet RMS downstream velocity The vertical profiles of dimensionless RMS velocity versus height (Fig. 7) in the head region of the flows (Fig. 7a) are irregular, with little similarity between the low- and high-density experiments. However, a broad trend with higher RMS velocity values at the top of the flow compared with the rest of the profile and a slight increase of RMS close to the base of the flow, exists. In the body (Fig. 7b), normalised RMS values are more uniform than in the head, and maximum RMS velocities are almost exclusively found at the base of the flows. A weak zone of relatively low RMS velocities is discernable at or around the height of the velocity maximum, particularly at distal locations. The RMS velocities in the tail of the low- and high-density flows collapse well (Fig. 7c), displaying a quasi-uniform pattern of RMS velocities along the entire flow depth, except for a slight increase at the lowest data point. At Location 1, the vertical profile of the flow of 14 initial concentration is broadly concave to the right (as are the profiles at all other Locations), whereas the profile of the flow of 55 initial concentration is concave to the left. This difference is interpreted to arise because the high-density flow has undergone the jet to turbidity current transition at this Location, whereas the low-density flow still has elements of jet structure. The higher dimensionless RMS values observed near the base of the body and tail profiles represent the turbulent eddies generated by friction with the base of the flume. This pattern is also seen close to the base of the flow for the head, yet the basal non-dimensional RMS values are less than those in the upper part of the flow, confirming that the head is more turbulent at its top than at its base. The profiles also show the general loss of turbulence as the flow propagates. The profiles become quasi-vertical straight lines with RMS values close to zero in the tail areas, which indicate that the flow approaches a laminar regime over its entire height Median grain size profiles The vertical profiles of median grain size (Fig. 8) essentially redisplay the data in Figs. 3c and 4c, but now allow a more direct comparison between the lowand high-density flows. The vertical profiles of the low- and high-density flows generally exhibit a similar trend for the head (Fig. 8a), body (Fig. 8b) and tail (Fig. 8c). At most locations, the sediment is relatively coarse near the base of the flow. In the central part of the flows, the sediment is relatively fine, while its size increases slightly in the uppermost part of most flows. The height of minimum grain size decreases from body to tail and from Locations 2 to 5, the grain size minimum being more prominent distally than proximally. Location 1 once again differs from the other locations in that the normal-to-inverse grading in the low- and high-density flows cover different dimensionless size ranges, and normally graded profiles (high-density flow) versus weakly graded to nongraded profiles (low-density flow) prevail in the body and tail. Once again, these discrepancies suggest that the flows may not have developed to the same state at Location 1.