Entangled external and internal controls on submarine fan evolution : 1 an experimental perspective 2

9 Submarine fans are formed by sediment-laden flows shed from continental margins into ocean basins. 10 Their morphology represents the interplay of external controls such as tectonics, climate, and sea11 level with internal processes including channel migration and lobe compensation. However, the 12 nature of this interaction is poorly understood. We used physical modelling to represent the evolution 13 of a natural-scale submarine fan deposited during an externally forced waxing-to-waning sediment 14 supply cycle. This was achieved by running five successive experimental turbidity currents with 15 incrementally increasing then decreasing sediment supply rates. Deposits built upon the deposits of 16 earlier flows and the distribution of erosion and deposition after each flow was recorded using digital 17 elevation models. Initially, increasing sediment supply rate (waxing phase) led to widening and 18 deepening of the slope channel, with basin-floor deposits compensationally stepping forwards into 19 the basin, favouring topographic lows. When sediment supply rate was decreased (waning phase), the 20 slope-channel filled as the bulk of the deposit abruptly back-stepped due to interaction with 21 depositional topography. Therefore, despite flows in the waxing and waning phases of sediment 22 supply having nominally identical input conditions (i.e. sediment concentration, supply rate, grain size 23 etc.), depositional relief led to development of markedly different deposits. This demonstrates how 24 external controls can be preserved in the depositional record through progradation of the basin floor 25 deposits but that internal processes such as compensational stacking progressively obscure this signal 26 through time. This evolution serves as an additional potential mechanism to explain commonly 27 observed coarseningand thickening-upwards lobe deposits, with abrupt transition to thin fine28 grained deposits. Meanwhile within the slope channel, external forcing was more readily detectable 29 through time, with less internally driven reorganisation. This validates many existing conceptual 30 models and outcrop observations that channels are more influenced by external forcing whilst internal 31 processes dominate basin floor lobe deposits in submarine fans. 32

Submarine fans are formed by sediment-laden flows shed from continental margins into ocean basins. 10 Their morphology represents the interplay of external controls such as tectonics, climate, and sea-11 level with internal processes including channel migration and lobe compensation. However, the 12 nature of this interaction is poorly understood. We used physical modelling to represent the evolution 13 of a natural-scale submarine fan deposited during an externally forced waxing-to-waning sediment 14 supply cycle. This was achieved by running five successive experimental turbidity currents with 15 incrementally increasing then decreasing sediment supply rates. Deposits built upon the deposits of 16 earlier flows and the distribution of erosion and deposition after each flow was recorded using digital 17 elevation models. Initially, increasing sediment supply rate (waxing phase) led to widening and 18 deepening of the slope channel, with basin-floor deposits compensationally stepping forwards into 19 the basin, favouring topographic lows. When sediment supply rate was decreased (waning phase), the 20 slope-channel filled as the bulk of the deposit abruptly back-stepped due to interaction with 21 depositional topography. Therefore, despite flows in the waxing and waning phases of sediment 22 supply having nominally identical input conditions (i.e. sediment concentration, supply rate, grain size 23 etc.), depositional relief led to development of markedly different deposits. This demonstrates how 24 external controls can be preserved in the depositional record through progradation of the basin floor 25 deposits but that internal processes such as compensational stacking progressively obscure this signal 26 through time. This evolution serves as an additional potential mechanism to explain commonly 27 observed coarsening-and thickening-upwards lobe deposits, with abrupt transition to thin fine-28 grained deposits. Meanwhile within the slope channel, external forcing was more readily detectable 29 through time, with less internally driven reorganisation. This validates many existing conceptual 30 models and outcrop observations that channels are more influenced by external forcing whilst internal 31 processes dominate basin floor lobe deposits in submarine fans. 32 Sediment gravity flow, allogenic, autogenic, submarine fan architecture, experimental modelling 34 1 INTRODUCTION 35 Submarine fans, the terminal portion of sedimentary source-to-sink systems, are amongst the largest 36 sedimentary accumulations on the planet (Normark, 1970;Posamentier and Kolla, 2003;Talling et al., 37 2007). Shaped by sediment gravity flows which deliver a range of natural and (more recently) 38 anthropogenic materials to deep-water environments, they provide an invaluable record of Earth's 39 climatic and tectonic history, and the dispersal of sediment, organic carbon and pollutants in the deep 40 ocean (Emmel and Curray, 1983  signals are only partially preserved, it will be necessary to acquire more robust datasets (e.g. multiple 83 core locations) in natural systems in order to confidently reconstruct turbidity current volume and 84 recurrence across sediment routing systems . 85 Here, we ask the question: how is an externally forced sediment supply cycle recorded in the 86 morphology and stratigraphy of a submarine fan? We investigate this question using a series of 87 experimental turbidity currents with incrementally increasing then decreasing sediment supply rates 88 (suspension discharge from a mixing tank) ( Figure 1B The experiments were conducted at Utrecht University in the Eurotank Flume Laboratory (Figure 2). 99 The experimental basin was 11 x 6 m in planform and filled to a water level of 1.2 m above the 100 horizontal floor. The initial tank bathymetry consisted of an 11° slope of 3 m in length (the "slope"), 101 followed by a 4° slope of 4 m in length (the "proximal basin floor"), ending in a 4 m long horizontal 102 "distal basin floor". This slope gradient, high for natural settings, promoted flow velocities high enough 103 to erode sediment and bypass sediment to the basin floor (de Leeuw et al. 2016). The tank floor was  104  covered by approximately 20 cm of loose sand of the same grain-size distribution as the turbidity  105 current mixture ( Figure 3F) enabling turbidity currents to erode into the substrate. A straight, 0.8 m 106 wide, 0.05 m deep, symmetrical channel form was sculpted into the initial 11° slope from the inlet box 107 to the break of slope (Figure 2A). The dimensions of this initial channel form were selected based on 108 the dimensions of a self-formed channel produced by de Leeuw et al. (2016). The turbidity currents 109 entered the basin via an inlet box with an un-erodible base of 0.7 m in length and gradually expanding 110 sides before continuing down the sediment covered slope. All boundary conditions were consistent 111 across all runs except for suspension discharge (see section 2.3 for details; Table 1; Figure 3).  2.2 Turbidity current suspension parameters 123 Prior to each experiment, the sediment mixture was prepared in a separate mixing tank with two 124 impellers that homogenised the mixture (Figure 2). The volume of the suspension (sediment and water 125 mixture) was 900 litres (L) in each event; sediment contributed 17% of this. Quartz sand (Sibelco BR-126 37) with a specific density of 2650 kg m -3 constituted 75% (300 kg) of the total sediment suspension 127 volume with the remaining fraction being 100 kg of silt-sized ground glass. The median grain size (D50) 128 of the mixture was 131 µm, with a D10 of 25 µm and a D90 of 223 µm ( Figure 3F). Grain size was analysed 129 using a Malvern Mastersizer particle sizer (Malvern Instruments Limited, Malvern, UK). 130

131
Five successive sediment-laden turbidity currents entered the basin from the inlet at the top of the 132 slope ( Figure 2). These currents were created by pumping the suspension from the mixing tank to the 133 basin via a supply pipe. Suspension discharge (i.e. volume per hour of flow into the tank) was 134 monitored using a discharge meter (Khrohne Optiflux 2300) mounted on the supply pipe and 135 regulated using a Labview control system (National Instruments Corporation (UK) Limited, Newbury, 136 UK). To simulate an external control on the system, in this case a waxing-to-waning sediment supply 137 cycle, the suspension discharge rate was increased between runs 1 to 3 from 20 m 3 h -1 , to 30 m 3 h -1 , 138 then 40 m 3 h -1 , before being decreased back to 30 m 3 h -1 , and then 20 m 3 h -1 in runs 4 and 5 respectively 139 ( Figure 1B  they are considered to each represent protracted phases of sediment delivery to the system. In each 151 phase, a similar volume of sediment was supplied, the effect of the higher discharge being that 152 turbidity currents were larger, and more powerful. Our scenario should thus be seen as an analogue 153 for increasing then decreasing turbidity current strength during an externally forced cycle (e.g. sea-154 level, climate, or tectonic variability). With this specification in mind, the suspension discharge rate 155 shall henceforth be referred to as 'sediment supply rate' for simplicity. A base case equivalent where 156 sediment supply rate was kept constant was not included in this study as earlier works serve to fill this 157 role and are referred to where appropriate (e.g. grains along a vector aligned with the probe axis. Bed-parallel velocity was calculated from the 172 measured data using trigonometry under the assumption that bed-normal velocity was zero. This was 173 plotted against time for each run and used to infer bed-base deposition and erosion through time as 174 the bed base increased or decreased in height (Supporting Figures 4-8). Time-averaging the velocity 175 data created profiles that enabled analysis of the downslope velocity evolution ( Figure 5). These 176 profiles were compared between runs to examine how velocities changed as the experiment 177 progressed. Velocity averages were taken for the entire run durations, minus the head and tail of each 178 flow (first and last five seconds). 179 Run deposits accumulated sequentially, illustrating how the turbidity currents responded to the 180 evolving topography in the basin. After each experimental run, the basin was drained, and the deposit 181 was scanned using a high-resolution laser scanner. This allowed production of digital elevation models 182 (DEMs) with a horizontal grid spacing of 2 x 2 mm, and a vertical resolution of < 0.5 mm. By comparing 183 DEMs from before and after each experimental run, deposition/erosion maps were generated ( Figure  184 4 and Supporting Figure 1). Due to high amounts of erosion directly after the inlet box where flows 185 passed over the boundary from un-erodible to erodible substrate, the upper 1 m of the slope channel 186 was restored to its original 0.8 x 0.05 m geometry to maintain the incoming flow properties between 187 experimental runs. 188

189
To realistically represent a natural system that can erode and transport sediment in suspension 190 downslope, the experimental turbidity currents of this study utilised Shields scaling . The fine-grained sand used for the flow and substrate (D50 =131) ensures transitionally rough flow in 203 the slope channel, promoting erosion through turbulent interaction with the bed. 204 The Shields parameter and the particle Reynolds number are calculated with: where is the sediment density (2650 kg / m 3 ), is the current density, g is the gravitational 208 acceleration (9.81 m s -1 ), D50 is the median grain size (131 µm), is the kinematic viscosity of fresh 209 water at 20°C (1x10 -6 ), and U* is the shear velocity, estimated using (Middleton and Southard, 1984;210 van Rijn, 1993): where Umax is the time-averaged velocity maximum, hmax is the height of the velocity maximum, k is 213 the von Kármán's constant (0.40), and the D90 of the grain size was 223 µm. See Supporting Table 1  214 for breakdown of dynamic and sedimentary experimental flow properties. 215 With this scaling approach we ensure the mobility of particles in the flow, generating turbidity currents 216 that can erode, suspend, or deposit sediment. The depositional pattern formed by these flows allows 217 identification of the general response of the system to external and internal controls. Section 4.2 218 places the experimental deposits into a hierarchical framework to assist comparison with natural 219 settings. However, it should be noted that the purpose of these experiments is not to directly replicate 220 the exact depositional architecture and hierarchy of natural settings, but to provide a practical 221 reference for their development. The spatial evolution of the flow field for runs 1 to 5 is presented in time-averaged velocity profiles to 305 show how flows developed between runs ( Figure 5). The maximum velocity (Umax) on the slope 306 increased from approximately 0.83 m s -1 (UVP 2) in run 1, to 1.09 m s -1 in run 3 (UVP 1) as sediment 307 supply rate was increased between runs. Umax then decreased in line with the sediment supply rate 308 to approximately 0.97 m s -1 in run 5 (UVPs 1 and 2). This trend of increasing then decreasing flow 309 velocity with sediment supply rate was also documented on the basin floor (UVPs 5-8). Uncertainty is 310 attached to the later (e.g. run 5) basin floor readings as the highly variable flow pathways created by 311 the depositional topography (see Figure 4A) hindered the probes' ability to accurately record the 312 dominant flow direction. Despite this uncertainty, the broad trends of increasing velocities with 313 increasing sediment supply rate were consistent across the slope and basin floor ( Figure 5). This flow-314 field evolution correlates with the depositional trend of a forward then back-stepping depositional 315 system, demonstrating a clear link between process and product. 316 scenarios plausible, but also that both processes may act upon the system concurrently. Rather than 353 progressively less sediment being overspilled with each run through the experiment, we observed 354 consistently high amounts of overspill in the waxing phase which abruptly declined in the waning 355 phase ( Figure 6). This newly documented evolution is driven by the interplay of sediment supply rate 356 (external) and constructional/erosional channel confinement mechanisms (internal). 357 358 the channel width was dropped from 1.2 to 0.53 m. This is approximately 9 times less than the velocity 389 increase we document between runs 1-3 in our study (0.27 m s -1 at UVP 2), indicating that the external 390 signal of sediment supply rate was the dominant control on the flow field evolution. 391 It is possible that by altering the pattern of sediment supply to the experimental system from flows 392 with quasi-steady sediment supply rate that incrementally increased between runs, to flows that also 393 had internal sediment supply rate variability (i.e. 2 nd order supply cycles) that sediment distribution in 394 the basin would be affected. However, physical and numerical experiments by Li et al. (2016) and 395 Foreman and Straub (2017) on deltaic and alluvial systems suggest that external controls (they use 396 relative sea-level and climate oscillation respectively) had to be of a greater spatial and temporal scale 397 than that of the internal dynamics of the system. This suggests that smaller-scale variation than that 398 applied to this experimental system may be undetectable in the depositional record, particularly in 399 increasingly distal settings. Supporting this, recorded discharge rates in our experiments show varying 400 amounts of deviation from the reference discharge values (sediment supply rate) but there is no 401 evidence of this small-scale variability in the resultant deposits ( Figure 3). 402 that the channel-fill deposits that contain the record of the external signal are not preserved in the 463 rock record. If we take the channel-fill deposits of our experiments for example, we record only the 464 deposits associated with back-filling and nothing of the erosive runs 1-3 that came before ( Figure 4B; 465 cross-section A-A'). Only with our high-resolution data set are we able to identify the complex 466 relationship between the channel axis and levees through time and attribute this to external and 467 internal factors (Section 4.1.1). In natural modern and ancient datasets, extracting explicit information 468 to differentiate between external and internal mechanisms within slope channels will continue to be 469 challenge due to resolution issues. By investigating modern systems with repeat monitoring over short 470 time-scales the degree of preservation within the channel axis may be more confidently resolved. In 471 contrast, basin floor deposits in natural settings do not record smaller turbidity currents that fail to 472 reach them, but their preservation potential is substantially higher than channels due to the 473 predominantly depositional nature of basin floor environments. 474

External versus internal controls on basin floor deposition
Our results suggest that whilst slope channel-levees may provide the best record of external signals, 475 they have low preservation potential in the channel axis. Meanwhile basin floor lobes feature a lower 476 resolution record of external signals, but a better-preserved depositional record. Section 4.3 provides 477 a possible mechanism whereby we may still be able to use this limited rock record in tandem with the 478 observations of this study to interpret stacking patterns in outcrop and core. flow events and so were technically beds by the above definition, the key aim was to represent a 499 protracted phase of waxing-to-waning sediment supply to a submarine fan over geological time. This 500 would be very difficult to resolve by considering five flow events in isolation. Each run of this study is 501 consequently considered to represent a lobe element, with the whole series of runs representing a 502 lobe (sensu Spychala et al., 2019). This approach is further supported by evidence that beds stack more 503 aggradationally relative to lobe elements which show more pronounced compensation (Straub and  504 Pyles compensation in modern intraslope settings, however, the compensation recorded in these 506 experiments is substantially more pronounced than that of the beds recorded in the western Niger 507 Delta slope. 508 Despite the usefulness of comparing our data to hierarchical schemes of natural systems, doing so 509 highlights some of the difficulty in applying strict organisational structure to nature. In the transition 510 between the channel and lobe in our experiment, the deposition is clustered or 'anti-compensational' 511 across all five runs ( Figure 4B, cross-section B-B'), with the deposits stacking on top of each other 512 (Straub et al., 2009). This aggradational character is likely due to the channel position effectively 513 controlling the depositional location. Therefore, compensation does not appear able to develop until 514 a distance down-dip from the channel-lobe transition ( Figure 4B, cross-section C-C', and Figure 8). 515 If the simplified view is taken that discrete 'hierarchical components' (i.e. bed-sets, lobe elements, 516 lobes, and lobe complexes) are internally composed of clustered units, at the break of slope in our 517 study there was only a single hierarchical component. There was no deposit compensation at this 518 location ( Figure 4B, cross-section B-B'), implying that multiple lobe elements did not exist there. If we 519 take this to be true, the hierarchical component becomes more of a local geometric definition rather 520 than a hierarchically delineated correlatable unit. This raises fundamental questions about 521 depositional hierarchy and its spatial applicability. For example, how do hierarchical components vary 522 in their geometry from proximal to distal and what are the implications for their practical application? 523 Out results suggest that lobe element-scale strata may be more challenging to distinguish near the 524 channel to lobe transition where deposits behave more aggradationally, versus the lobe fringe where 525 compensation is common. 526

527
The evolution from forward-stepping and compensational stacking, to abrupt back-stepping recorded 528 in this experimental fan can be used as a possible explanation for bed stacking patterns commonly 529 observed in outcrop and subsurface-cores from examples in the rock record. A thickening-and 530 coarsening-upwards trend in submarine lobe deposits has been described from several outcrops and 531 this is often followed by an abrupt transition to thin-bedded fine grained sediments, usually 532 interpreted as hemipelagic abandonment or distal fringe facies (Pickering, 1983 is typically attributed to the local depositional environment becoming progressively higher in energy, 535 transitioning from marginal to more axial fan localities (Kane and Pontén, 2012). However, the forcing 536 mechanism for the abrupt transition from thick sandstones to packages of fine-grained sediments is 537 less clearly understood. We argue that the evolution of the 'experimental lobe' in this study provides 538 an elegant way to explain this stacking pattern. Figure 9 shows the temporal evolution of the 539 experimental lobe in both 2D and 3D space. The 2D diagram ( Figure 9A zone on these logs indicates 'Lobe 5', a typical example of this coarsening and thickening trend that 561 abruptly reverts to siltstone. Conventionally, the siltstone at the top of the sandstone has been 562 interpreted to represent one of two models: 1) A condensed section of hemipelagic deposition during 563 an externally driven reduction in sediment supply (Johnson et al., 2001;Hodgson et al., 2006); 2) 564 Lateral fringes of additional lobes, representing system-internal lobe-scale compensation (Prélat et 565 al., 2009). Recent studies are beginning to challenge the notion that mud deposition within active 566 submarine fan systems is purely hemipelagic in nature, more likely representing the distal fringe of 567 active systems (Boulesteix et al., 2019). This suggests it is unlikely that the siltstones above the lobe 5 568 sandstones are reflecting a complete 'shutdown' of sediment supply. The model of lateral fringe 569 aggradation of later lobes is more likely due to widely recognised compensational stacking and 570 associated grain-size distributions in lobe deposits (Deptuck et al., 2008;Prélat et al., 2009;Straub and 571 Pyles, 2012). However, this does not explain the abruptness at which deposits transition to fine-572 grained sediments ( Figure 9C). We propose an adapted version of this model whereby this transition 573 can be more readily explained by a combination of compensational stacking and rapid back-stepping 574 of 'lobe elements' (Figures 8 and 9). It is suggested that depositional relief and waning sediment supply 575 as is observed in our experiments drives this evolution, leading to the stratigraphic patterns we 576 observe in nature. 577 Identification of back-stepping deposits from compensationally lateral-stepping deposits will always 578 be challenging in outcrop and core due to the likelihood of similar facies being present in distal along-579 axis and off-axis trends. Differentiating criteria for back-stepping deposits would include: a) abrupt 580 vertical transition from sand-dominated to mud dominated facies; b) beds that thin across-strike in 581 both directions, rather than thickening laterally into an adjacent lobe axis; and c) a preference for 582 deposition of hybrid event beds relative to ripple-laminated deposits. Hybrid event beds have been 583 documented to characterise deposition in frontal fringe environments where we might expect to 584 observe back-stepping, whilst ripple-laminated deposits show preference for the lateral fringe 585 . Identifying any or even all of these criteria would not mean unequivocal proof 586 for back-stepping due additional basinal complexity such as complex regional topography, however, 587 they would provide the basis for assessment of submarine fan evolution when considered within the 588 context of the regional picture. Identification of abruptly back-stepping strata in the rock record of 589 any given system would have implications for our understanding of the distribution of sediment within 590 that basin. If strata are identified as abruptly back-stepping, this suggests that the system may have 591 built depositional relief to the point of forcing the system backwards irrespective of external sediment 592 supply, perhaps due to a degree of (scale-dependent) basin confinement. If no evidence for abrupt 593 back-stepping is observed, this may imply that incoming flows have had space to continue to stack 594 compensationally until sediment supply has waned, allowing for a more 'classic' gradational upwards 595 transition to fine grained deposits.

608
Using physical models with a signature of waxing-to-waning sediment supply, the interplay of external 609 signals with internal processes within submarine fans has been evaluated. On the channelised slope, 610 increasing sediment supply rate resulted in increased channel erosion and overbank deposition. The 611 evolved channel dimensions improved flow efficiency, enhancing the external signal on the slope. 612 Concurrently on the basin floor, increasing sediment supply rate led to forward-stepping of lobe 613 elements, however this was partially obscured by internal reorganisation through compensational 614 deposit stacking. When sediment supply rate was subsequently reduced, basin floor deposits back-615 stepped abruptly due to depositional relief to onlap the slope and infill the slope channel. Flows were 616 then underfit with respect to the evolved channel dimensions and confined within the widened and 617 deepened channel. Consequently, limited overbank deposition took place in the waning phase of 618 sediment supply. This complex overall evolution resulted in deposits that were distinctly different in 619 the waxing and waning phases of sediment supply, despite similar external input conditions. 620 A comparison of the slope and basin floor environments revealed that external factors have a 621 stronger influence upon slope channels whilst internal processes dominate basin floor lobe deposits. 622 These finding validate many conceptual models of submarine fans, including sediment supply driven 623 progressive channel confinement, and how internal reorganisation can shred external signals in the 624 deepest parts of the sedimentary sink. Despite this internal 'dilution' of the external signal and the 625 poorer preservation potential of deposits in the slope channel axis, the external signal could still be 626 observed on the basin floor, with deposits from higher sediment supply rates extending farther into 627 the basin before depositional relief dominated. 628 The recorded evolution of forward-stepping and compensation followed by abrupt back-stepping 629 represents the signature of an entangled external-internal cycle of sedimentation in a submarine 630 fan. This evolution is a possible new mechanism to explain common vertical stacking patterns of 631 coarsening and thickening upwards sandstone successions followed abruptly by thin-bedded fine-632 grained sediment in outcrop and core. These findings should encourage continued analysis of 633 submarine fan architecture from a perspective that integrates both external and internal controlling 634 mechanisms and provide a new evolutionary model to search for in natural systems. Future work 635 may aim to test a range of different external signals such as variable sediment concentration or grain 636 size to assess whether these have a different impact on the organisation of submarine fans. 637

638
Equinor ASA is acknowledged for funding this research. Thony van der Gon Netscher is thanked for 639 technical assistance with the experiments. Michael Clare and an anonymous reviewer are thanked for 640 their insightful comments that broadened the scope of this work. Zane Jobe, Brian Romans, Peter 641 Burgess, and an anonymous reviewer are thanked for their helpful comments on an earlier version of 642 this manuscript. Euan Soutter is acknowledged for digitising Supporting Figure 9. 643

644
The authors have no conflict of interest to declare. 645

646
The data that support the findings of this study are available as supporting information.