Non stationarity of basin scale sediment delivery in response to climate change T. J. Coulthard1, J. Lewin2 and M. G. Macklin2 1 Geography Department, University of Hull, Hull, HU6 7RX, U.K. 2 Institute of Geography and Earth Sciences, University of Wales, Aberystwyth, SY23 4DB, U.K. T.Coulthard@hull.ac.uk http://www.coulthard.org.uk Abstract Results from a cellular river basin evolution model (CAESAR) that contains a detailed multi-grain size sediment transport model, indicate that over longer time scales (greater than 50 years) the relationship between sediment discharges (Qs) and water discharges (Qw) is not stable when routed through a prototype catchment. By modelling how the daily bedload yield from a medium sized river basin (383 km2) responds to changes in climate over the last 9000 years, high resolution and long term basin-scale bedload ratings curves can be simulated. These show that quasi linear relationships can be established between Qs and Qw during stable climatic periods, but an increase in flood magnitude and frequency over a sustained period (>10 years) can lead to a very large increase in the amount of sediment delivered for identical sized floods. Thus different ‘climatic periods’ can produce significantly different relationships between Qs and Qw. Furthermore, over the 9000 years simulated there is considerable variation in response, with daily sediment yields for a medium sized flows (16-32 m3s-1) varying over 8 orders of magnitude. The change in Qs/Qw relationship is believed to be conditioned by sediment supply and storage, and also by the ‘context’ of the climate period with respect to previous periods. These results have important implications for the application of sediment transport formulae, engineering calculations based on these (e.g. longer term reservoir siltation), and predictions of how river systems are likely to respond to climate change. 1 Introduction Understanding the relationship between catchment water and sediment discharges is important for comprehending how fluvial systems operate. For geomorphologists controls on the delivery of sediment have numerous implications, from understanding fluvial landform development to interpreting alluvial stratigraphies. And for engineers, knowledge of catchment sediment yields are vital, especially for assessing channel design, stability and sedimentation problems. Therefore, it is important for both scientists and practitioners to be able to model, and predict how the amounts of sediment transported from a catchment may respond to changes in water discharge over a range of time scales. To do this, researchers have measured catchment sediment yield - the amount of sediment that leaves a catchment over a measured period of time. This can be related to the size of the catchment and other factors such as climatic and altidudinal setting (e.g. Milliman and Syvitski, 1992). These records are variable, and changes in sediment yield over time have been related to changes in land use and land cover (Dearing, 1992) as well as to fluctuations in climate (Wilby et al., 1997). More detailed studies have directly related catchment sediment yields (Qs) to catchment water discharge (Qw), and these are termed sediment ratings curves. The relationship between Qs and Qw is not straightforward and rarely linear. Typically, there is a hysteresis effect over a flood, where there are reduced levels of sediment relative to water discharges in the latter parts of a flood, due to sediment exhaustion. Sediment ratings curves are usually used to describe suspended sediment to Qw relationships, but bedload ratings curves have also been developed (Whiting et al., 1999; Emmett and Wolman, 2001; Ryan et al., 2002; Hayes et al., 2002; Whiting and King, 2003; Ryan et al., 2005). These also show considerable variation, but do not show the same daily hysteresis effects, though Moog and Whiting (1998) described a seasonal hysteresis. Both suspended sediment and bedload ratings curves are characterised by large quantities of scatter, which like hysteresis effects can be attributed to limitations in sediment supply. However, empirical models have been developed that use these field data from ratings curves to predict both suspended sediment and bedload yields from catchments (Barry et al., 2004). A more sophisticated method is to use sediment transport functions. These typically measure sediment discharge per unit width, and can therefore incorporate 2 local channel characteristics such as slope, bed material etc. Therefore, they build in more of the local conditions and thus physics influencing sediment transport at that point. Sediment transport functions have been extensively developed, but have encountered difficulties as they tend to perform well on the data with which they were developed, but less so when applied to different environments. Gomez and Church (1989) clearly illustrated this point with their comparison of 12 formulae, applied to 8 different data sets. None performed accurately, the Bagnold equation was best, and only 3 others performed reasonably. The difficulties found in predicting bedload discharge are perhaps not surprising when the raw field and laboratory data shows significant variations (Reid et al., 1980; Hoey and Sutherland, 1991; Ashmore, 1988; Carling et al., 1998). For example, Cudden and Hoey (2003) measured bedload from two anabranches of a pro-glacial stream and noted substantial variations in bedload yields. These appear highly non-linear, and Gomez and Phillips (1999) calculated that most of the non-linearity on 10000 flume bedload measurements could be described as chaotic. Causes for these irregularities include the threshold based nature of sediment entrainment and other threshold dominated processes operating on the bed of heterogeneous gravel-bed rivers, such as bed armouring and equal mobility. Furthermore, the passage of bedforms (e.g. dunes and bars) and larger scale features such as sediment waves (Nicholas et al., 1995) can cause irregularity in at a point bedload yields ranging from 1 to 1000s m3 that may take seconds to years to pass. Over longer time scales, the impacts of climate or land cover change may also cause longer lasting perturbations in sediment delivery (Macklin and Lewin, 1989; Coulthard and Macklin, 2001). These problems are compounded by the physical difficulties associated in measuring bedload. Measurements are carried out using active methods (e.g. Helly Smith samplers, pressure pillows (e.g. Reid et al., 1980), magnetic devices (Tunnicliffe et al., 2000)) or passive techniques (sediment traps, repeat surveys of channel cut and fill or lake/reservoir sedimentation rates). Active methods can give temporally high resolution data (hourly or less) but are demanding in human resources and typically only span a few hours or days. Passive techniques can provide much longer records, but with a far coarser temporal resolution, as data points are only collected when traps 3 are emptied or cross sections re-surveyed. Therefore we are limited to either short high resolution, or longer low resolution records. Other researchers have looked to the geological record as a longer term recorder of changes in sediment yield. These demonstrate varying flood-frequency episodes, with ‘wet’ and ‘dry’ periods related to climatic fluctuations on a scale of tens to thousands of years (e.g. Knox, 1993; Macklin and Lewin, 2003; Macklin et al. 1992; ). Such timescales are compatible with the likely travel times of bed materials through catchment systems. However, the resolution of the dating methods used to establish fluvial chronologies is generally not precise enough to accurately link sediment response to individual floods. Similarly, the resolution and accuracy of proxies used to reconstruct palaeo climate records also hampers correlation with a stratigraphy. Furthermore, the sedimentary record is palimpsest, a record that may have been reworked and erased (Lewin and Macklin, 2003), and is not a straightforward ‘recorder’ being described as a ‘finicky’ by Paola, 2003. Therefore, we are faced with the reality of having temporally limited data on present day sediment yields (too short a length of record), and long term records (too coarse a temporal resolution). As a consequence, it is difficult to apply longer term records to shorter time scales (e.g. 10’s to 100’s of years) for instance to see how climate change may effect river systems. Similarly, it could be dangerous to apply present day process rates to interpret past longer term changes. Therefore, outside of our comparatively short length record we have little information on whether or how Qs/Qw relationships may change, and what might control it. Does the amount of sediment discharged for a given flood change over time, if so by how much, and what may influence this? In order to better understand this and how river systems may respond to both individual floods, as well as longer term shifts in average flood magnitude and frequency caused by climate changes, ideal data would be hourly water and sediment discharge data integrated across a river basin and spanning a period of several thousand years. This is clearly beyond the scope of monitoring studies and palaeoflood reconstruction techniques. But by using numerical modelling it is possible to explore the probable roles of changing flood frequency and sediment 4 supply and routing factors on a catchment wide basis. Indeed, previous studies have modelled non-linear sediment discharges from identical floods on small UK catchments (Coulthard et al., 1998) and larger abstract landscapes over longer time scales (De Boer, 2001). They also show how decadal outputs of sediment can vary significantly between catchments despite having identical drivers (Coulthard et al., 2005) as well as how thresholds are important for non-linear catchment response (Tucker, 2004). In this paper, we use a cellular landscape evolution model CAESAR (Coulthard et al., 2002) to reconstruct a series of bedload ratings curves based on 9000 year simulation of daily water and sediment discharges. The results presented here may have important implications for the management of catchments and some of the concepts that underpin our current understanding of geomorphology. Method The results presented in the paper are based on further analysis of data first presented in Coulthard et al. (2005), generated using the CAESAR landscape evolution model. A brief description of the model is provided below, but for a detailed account, readers are referred to Coulthard et al. (2002) and Van De-Wiel et al. (In Press). CAESAR is a cellular landscape evolution model (Coulthard, 2001; Willgoose, 2005) that represents a river catchment with a grid of equally sized square cells (as per a raster DEM). Each of these cells has properties, or values, for elevation, grain size, water discharge, flow depth, vegetation cover etc.. For every iteration or time step of the models operation these values are altered according to equations that describe four key groups of processes; hydrological, hydraulic, fluvial erosion and deposition, and slope processes. The hydrological model is an adaptation of TOPMODEL (Beven and Kirkby, 1979) and is driven by an hourly rainfall data set. Key parameters within TOPMODEL (m and K) can be altered to change the hydrograph peak and rate of flood decay, and these are altered within CAESAR to simulate the effects of land cover change on catchment hydrology. This provides distributed values of soil saturation levels, and when a grid cell is saturated, any excess discharge from the hydrological model is 5 treated as surface flow. This model is also dynamic, so wetted areas and thus the drainage network, can expand and contract during storm events. Surface discharge from the hydrological model is then fed into the hydraulic component, where depths, inundation areas and surface water routing are carried out. Flow depths are calculated for each cell using an adaptation of Mannings equation (using bed slope) and then routed according to a scanning multiple flow routing algorithm. This sweeps across the catchment four times (north to south, east to west, west to east and south to north) routing water to the three cells in front, as per Murray and Paola (1994), using the slope between the water surface height and the receiving cells bed elevation. The maximum depth calculated for a cell through each of the four scans is recorded and taken as the cell flow depth. Where a cell has a flow depth, fluvial erosion and deposition is carried out. Using water depth and bed slope between cells, bedload transport for 9 separate grain sizes (1-256 mm in Phi classes) is calculated using the Einstein-Brown (1950) formulae. Erosion and deposition of the separate size fractions is implemented through the use of an active layer system (Hoey and Ferguson, 1994) where a proportion of each fraction is held in a series of layers that during erosion can be replenished from layers below, or during deposition displace material to lower layers. Importantly, this system allows many of the processes associated with heterogeneous bed sediment transport to be modelled, such as bed armouring and size supply limited entrainment. Two sets of slope processes are modelled. First, mass movement occurs when a threshold slope angle is exceeded and secondly, soil creep is calculated according to local slopes. Importantly, both slope processes are fully integrated within the fluvial model, which allows the addition of sediment into the fluvial system from a landslide or eroding river bank, for example. In summary, CAESAR allows hydrological, hydraulic, fluvial and slope processes to be simulated over an entire catchment modelling series of individual flood events, driven by and hourly rainfall record and DEM. 6 The simulations described here were carried out on the upper part of the River Swale, U.K., the northern tributary of the Yorkshire Ouse system. The modelled catchment area covers 383 km2 and varies in relief from 514 to 20m A.O.D. The Swale has a mixed lithology of Carboniferous sandstone, gritstone and limestone, and was glaciated during the Quaternary, leaving wide steep walled valleys covered with periglacial and glacial deposits. The catchment was extensively forested (Flemming, 1998) but present day land cover is typically rough grazing on upland moorland, with pasture on the lower sections and valley floors. The catchment is sufficiently high to be unaffected by Holocene sea level rise. The model was applied to a 50 m resolution DEM of the Swale, and was driven by a proxy climate and land cover data set covering the last 9000 years. The topography of the Swale catchment 9000 years ago is impossible to reconstruct, so rather than make crude reconstructions based on limited data, the present day topography was used. This is not a wholly unreasonable assumption, as field evidence indicates that there has been less than 2-4 m of vertical channel and valley floor movement over the Holocene. For such a long simulation, a 9000 year hourly rainfall record was required. We used a combination of two climatic indices derived from peat bog surface wetness reconstructions from northern England for the period 6300 cal. BP to present (from Barber et al., 1994) and from 9000 to 6300 from Scotland (Anderson et al., 1998). A full description and discussion of this method can be found in Coulthard and Macklin (2001) and Coulthard et al., (2005). These indices provided climate data in 50 year steps and were then normalised between 0.75 and 2.25 to create a rainfall index. The model was then driven by a 10 year hourly rainfall record that is repeated 5 times to span 50 years, then multiplied by the rainfall index to create 9000 year proxy hourly rainfall data. For land cover changes, a basic reconstruction of catchment deforestation constructed from local palynological records were used to change the ‘m’ parameter in the hydrological model. Results The simulation generated a time series of 87x106 data points of hourly flow and sediment discharge. To simplify analysis of the data set, it was divided into 50 year sections, discharge data were averaged and bedload yields summed to create a daily 7 time series. In effect, we have used the model to generate daily water discharge and bedload totals for the River Swale over a 9000 year period. Figure 1 shows the relationship between Qw (water discharge) and Qs (sediment yield) for three 50 year sections (50-100, 100-150 and 150-200 cal. BP). Despite being plotted on log-log axis these data demonstrates a significant amount of scatter, with an interesting increase in variability to up to 8 orders of magnitude around the 15-50 m3/sec Qw. These periods were chosen as they contained very high (150-200 cal. BP), high (100-150 cal. BP) and medium (50--100 cal. BP) sediment discharges, though similar patterns of scatter are found in all 50 year sections. When the data is grouped into log classes according to Qw (1, 2, 4, 8, 16, 32, 64, 128, 256 and 512 respectively) the standard deviation of the Qs in each bin varied considerably, with a peak in the 1632 m3s-1 group. Using this group from the 150-200 cal. BP section and classing the Qs into linear bins (incrementing from 0 to 20000 in steps of 1000) the frequency of daily sediment yield (Qs) to water discharges in the 16-32 m3s-1 group can be plotted as a power law (Figure 2b). Data from 50 cal. BP period, with far lower sediment yields, also follows a similar relationship, though smaller bin sizes are required (Figure 2a). Examining Figure 1, there is also considerable difference in the angle of the regression lines plotted through the data sets. This suggests that there could be significant variation in the Qs/Qw relationship during different periods of the model operation. However, R2 values indicate that the relationship is clearly not significant due to the large amount of scatter. In order to reduce this scatter and to allow comparisons in the Qs/Qw relationship between each 50 year section of the simulation, these data were grouped into log classes as above. When the medians of Qs for each of these bins were plotted these revealed a good linear relationship as shown in Figure 3. This good level of fit was found to be consistent for each of the 50 year sections of data, though it must be remembered that using medians conceals a large amount of scatter and therefore should only be used as a comparator between individual 50 year sections. Linear regression was then carried out for the medians of all 50 year time sections and the slopes of the regression line are shown as a time series in Figure 4. This indicates how the angle of the modelled relationship between Qs/Qw changes quite dramatically 8 throughout the simulation as well as following sediment yield and climate input closely. The standard deviation and median for the 16-32 m3s-1 class were also calculated and are shown in temporal sequence for all data in Figure 5. Discussion (1) Scatter Figure 1 indicates that there is considerable scatter within the Qs/Qw relationship, and it is apparent that there is a flow size with substantially more scatter than others (1632 m3s-1). This is not the most frequent class of flow (8-16 m3s-1) and this area of scatter seems to persist throughout every 50 year section, though for some sections is less well defined. These results raise two questions: why is there so much scatter in the Qs and why does there appear to be a flow range that produces substantially more scatter than any other? At first the large amounts of scatter may seem to be inappropriate, as CAESAR is driven by a conventional sediment transport formulation (Einstein and Brown, 1950) that has no stochastic element. Thus it might be expected that sediment outputs would follow this relationship. However, within CAESAR this formulae is applied to individual cells, with sediment transport calculated locally, based upon the flow depths and bed slopes of each cell. The output presented here, is the product of erosion and deposition being carried out over thousands of grid cells within the model. Therefore, by calculating sediment transport on a cell by cell basis, CAESAR mimics the processes of internal storage and re-mobilisation of sediment that can operate within real catchments. Furthermore, as CAESAR contains a sophisticated sediment transport model using several grain sizes, when supply of a certain grain size is exhausted within a cell, no more of this fraction can be eroded. Thus sediment supply limitations, such as bed armouring and selective entrainment are incorporated in each cell. This can generate internal erosional thresholds, so sediment is not eroded until stream powers are great enough to move the particle itself or to remove the coarser armour layer above. Therefore, erosion is threshold dominated giving rise to high variations in sediment delivery. Furthermore, as the model incorporates slope processes (soil creep and mass movement) material can be added from landslides, 9 which are also threshold controlled, again introducing high variations in sediment delivery. Explanations for why one class of mean daily flows produces such high levels of variability are not straightforward. It is logical to assume that in threshold based supply limited catchments, small floods will largely generate little sediment yield and large floods will generally mobilise a substantial volume of material. Therefore the medium sized flows, that here show most variability, may be close to erosional thresholds, where there is high potential for variability. In order to evaluate what might control the variability shown in the model results, data from similar simulations on two nearby catchments were examined. Figure 6 plots Qs/Qw relationships from the River Swale, as well as the nearby River Nidd (281 km2) and River Wharfe (697 km2). Both exhibit similar patterns of scatter, with the similar sized Nidd being closest. However, the larger Wharfe catchment shows greatest variability in a larger flood size class. This suggests that the size of the catchment or possibly its morphology may be a controlling factor. Possibly it is not small scale erosional thresholds such as bed armour breaching that influence the scatter in this way, but larger controls on thresholds, such as valley floor shape and internal catchment storage. Further research is required into the precise causes of this behaviour, and it is possible that other factors, for example, grain size and land cover may strongly influence this. Figure 2b shows that the distribution of the scatter follows a power law. This indicates that Qs may vary from 0 to 20 000 m3day-1 for a mean daily flow between 16 and 32 m3s-1. Interestingly, Figure 2a indicates that different climatic periods produce different power law relationships, showing that whilst the magnitude of scatter may be different, they both follow similar distributions. Power laws can also be indicative of systems exhibiting self organised criticality (SOC) though this is not an area we wish to explore in this paper. Implications of scatter The large amounts of scatter (up to 8 orders of magnitude for flows between 16 and 32 m3s-1) suggest that bedload ratings curves are not ideal for predicting sediment 10 discharges. Sediment transport functions that take account of factors such as channel width, depth and local slopes where the bedload is measured (here the edge of the DEM) would improve these predictions. But, both will still feel the effects of changes in sediment supply outlined above and this study highlights some of the difficulties faced when trying to predict sediment transport. Possibly and alternative method is to smooth these data, and average it over monthly, annual or even decadal time steps, much of the scatter would be removed. By using such long measurement intervals we might then be able to derive generalised relationships between long term flow and sediment yield data. But such relationships, though statistically less variable would mask the extremes of response that might be crucial for management strategies. The clustering of most scatter around a certain flow size may be of great importance to engineers and practitioners. Akin to the concept of the effective flood (Wolman and Miller, 1960) is there a size of flood that may be most difficult to account for the extra levels of uncertainty associated with it? Certainly, these results suggest that this may be the case, and that such magnitude floods are very common. However, when reviewing these data presented here, it should be remembered that Figures 1 and 6 do not represent individual flood totals, they are daily sediment totals compared to daily discharge averages. Therefore, these data could easily conceal daily/event scale hysteresis effects. Furthermore, the data for 150-200 years cal. BP represents an extreme of wetness and the greatest level of variation. Under dryer conditions the maximum size of variation is far less (e.g. see Figure 2 b) but the amount of scatter remains very high. (2) Non stationarity of response Figures 4 and 5 show how 50 year climatic ‘periods’ produce a different relationship between Qs/Qw. There is an increase in the gradient of the relationship with increasing wetness, resulting in a greater Qs for the same Qw in a wetter climate (represented by an increase in the climate index). In a wetter period in a mid latitude temperate river basin, it would be anticipated that there would be a higher overall sediment yield, due to larger floods. But the slope shows how the ratings curve, the Qs/Qw relationship, changes significantly. In effect the relationship is non-stationary over time. 11 These changes can be explained firstly by morphological changes within the catchment. Increasing flood size leads to an expansion of the drainage network, that generates fresh material from expanding stream heads and incision within first order streams as demonstrated by previous applications of CAESAR (Coulthard et al., 2002). In a small catchment, it might be expected that all this new material would be flushed from the system. But within a medium to large sized catchment, such as the Swale, this material will be deposited and progressively moved downstream as a series of sediment waves or slugs. This in turn leads to the increase in the Qs/Qw relationship, as the larger floods have mobilised sediment that is then available for smaller floods to export. In effect, larger floods or wetter climatic periods loosen or release large amounts of sediment, increasing the volume of material that could be transported for smaller flood events. Similarly, changes in the general pattern of flood events may breach armour layers on the stream bed, which can release sediment as well as make the channel more vulnerable to incision (Coulthard and Van de Wiel, 2007). The sudden shift in Qs/Qw could also be explained by changes in the channel pattern. Significant increases in bedload yield would be consistent with a transformation from a single thread channel to a braided one. This can be seen in Figure 7, which shows the channel pattern for two identical size floods at 4500 and 200 years Cal. BP, corresponding with low and high Qs/Qw relationships. The upper 4500 cal. BP. image shows a predominantly single channel, with some overbank inundation cutting across meander bends. The lower 200 cal. BP image shows a widening of the channel in the upper parts (as an adjustment to increased flows) and a shift in channel pattern to a multi threaded braided planform. Such behaviour is consistent with field evidence from similar catchments in the UK (the river South Tyne; Macklin and Lewin, 1989) that showed dramatic changes in channel pattern and valley floor aggradation. Such a planform change could also explain increases in scatter and the standard deviation (Figure 6) during wetter climates, as braided channels can have a far greater fluctuating bedload yield. The changes in the Qs/Qw relationship can also be explained by the effects of sediment recharge and removal. For example, during several hundred years of relatively constant moderate rainfall events, the river channel armours and adjusts its 12 width/depth ratio’s to accommodate the flow. During this period, in upland areas soil creep will slowly move soil and sediment to the base of slopes and thus the margins of streams. Present sediment supply remains the same, but potential supply has increased. If there is then a wetter climatic period, with more frequent and larger floods, the relative equilibrium within the channel will be broken and erosion will start to remove this available material adjacent to channels. This increases the relative sediment delivery per flood size, as we see in Figure 1. Implications These results have significant implications for the application of bedload rating curves as they show that the Qs/Qw relationship can vary considerably over time. The modelling results presented here show that there is a double response, as not only will wetter climate periods increase sediment yields through larger floods, but will also increase the relative sediment yield for all floods (the Qs/Qw relationship). This is especially important for design predictions as if, for example, we predict the rates of sedimentation in a reservoir – or around a structure such as a bridge, based on existing sediment/bedload ratings curves, these may alter significantly in the future. Therefore future changes in climate may influence not only the size of future floods, but also the relative volumes of sediment released. This is especially pertinent given the rapid present day changes to our climate. For sedimentologists, when interpreting sequences, the double effect may lead to a large deposit being disproportionately thick, as there is clearly a far from linear relationship between climate and unit thickness. Relating both scatter and the Qs/Qw relationship, Figure 6 shows how an increase in the standard deviation of the 16-32 m3s-1 flood class follows the climate and sediment yield. This indicates that not only are Qs/Qw values going to change, but also the variability, which could make engineering decisions and sedimentological interpretations even harder. Finally, as changes in the Qs/Qw relationship are influenced by the release of stored sediment and the renewal of erosional activity in stream heads and adjacent parts of the channel, then catchment response will be heavily influenced by the previous patterns of erosion and deposition. Therefore, catchment sediment yields are heavily contingent on the system history and predictions, and assumptions based upon bedload ratings curves should take this into account. Preliminary analysis examining 13 how the Qs/Qw relationship varies during the 50 year blocks studied here indicates that there is significant adjustment over the first 20 years followed by more minor changes. This suggests that the reaction time to climatic changes of the simulated River Swale is fairly rapid, but how this timing scales between catchments of different shapes and morphologies is beyond the scope of this paper and warrants further study. However, this does give us some indication of how rapidly river systems may respond to climatic changes. Conclusions Catchment modelling of bedload sediment yield for a moderate sized river basin over the last 9000 years using the CAESAR model suggests that the spatial and temporal complexities of sediment transport and storage produce markedly non-linear relationships between water discharge and sediment yield. The yield of any particular discharge is conditioned by prior events, sediment availability and sediment supply. This is particularly striking for moderate sized flows (in the case of the modelled River Swale 16-32 m3s-1) for which daily yields vary over eight orders of magnitude. For a given flow magnitude, wetter periods also yield more sediment than drier ones, probably because sediment mobilisation then exceeds critical thresholds, as in the breaching of bed-armouring, landslide generation or storage activation. The spatial complexities of larger catchments, in terms of available storage and the temporal routing of sediment transfers would also appear to make predictions of Qs/Qw relationships much more variable than in the case of small catchments. Transitions between ‘wetter’ and ‘drier’ periods also seem liable to particularly uncertain predictions. All these uncertainties have serious management implications at times when rapid climate change and human impacts on sediment yields are much in evidence. There are no long term data sets available with which to attempt model validation, though there are no reasons to suppose that the use of a conventional sediment transport equation, along with their hydrological and process relationships, to model catchment wide transportation and delivery, is in any general sense invalid. Indeed, it suggests that ‘site’ transport equations do need to be aggregated with procedures which incorporate catchment behaviour, if sediment yield variability is to be simulated in an 14 apparently realistic manner. Modelling reveals some interesting characteristics of this variability which have management consequence at times of rapid environmental change. Acknowledgements We would like to thank Chris Paola, Jim Pizzuto and a third anonymous referee for their excellent comments and guidance in the preparation of this paper. TJC would also like to thank the organisers of GBR6 for the opportunity to present this research. 15 References Anderson, D. E., Binney, H. A., Smith, M. A., 1998. Evidence for abrupt climatic change in northern Scotland between 3900 and 3500 calendar years BP. The Holocene 8, 97-103. Ashmore, P. E. 1988. Channel morphology and bedload pulses in braided, gravel-bed streams, Geografiska Annaler, 73A(1), 37–52. Barber, K. E., Chambers. F. M., Maddy. D., Stoneman. R., Brew. J. S., 1994. A sensitive high resolution record of late Holocene climatic change from a raised bog in northern England. The Holocene, 4, 198-205. Barry, J. J., Buffington, J. M. and King, J. G. 2004. A general power equation for predicting bed load transport rates in gravel bed rivers. Water Resources Research. 10.1029/2004WR003190 Beven, K. J., Kirkby, M. J., 1979. A physically based variable contributing-area model of catchment hydrology. Hydrological Science Bulletin, 24, 43-69. Carling, P. A, Williams, J.J, Kelsey, A, Glaister, M. S, and Orr, H. G. 1998. Coarse bedload transport in a mountain river. Earth Surface Processes and Landforms 23: 141-157. Coulthard, T. J. 2001. Landscape evolution models: a software review. Hydrological Processes. 15, 165-173. Coulthard, T. J., Macklin, M. G., 2001. How sensitive are river systems to climate and land-use changes? A model-based evaluation. Journal of Quaternary Science 16, 347351. Coulthard, T. J., Kirkby, M. J., Macklin, M. G., 1998. Non-linearity and spatial resolution in a cellular automaton model of a small upland basin. Hydrology and Earth System Sciences 2, 257-264. 16 Coulthard, T. J., Macklin, M. G., Kirkby, M. J., 2002. A cellular model of Holocene upland river basin and alluvial fan evolution. Earth Surface Processes and Landforms 27, 269-288. Coulthard, T. J., Lewin, J. and Macklin, M. G. 2005. Modelling differential catchment response to environmental change. Geomorphology. 69, 222-241. Coulthard, T. J. and Van De Wiel, M. J. In Press. Quantifying Fluvial Non-linearity and Finding Self Organized Criticality? Insights from Simulations of River Basin Evolution. Geomorphology. Cudden, J. R. and Hoey, T. B. 2003. The causes of bedload pulses in a gravel channel: The implications of bedload grain-size distributions. Earth Surface Processes and Landforms 28: 1411-1428. De Boer, D. H. 2001. Self-organisation in fluvial landscapes: sediment dynamics as an emergent property. Computers and Geosciences. 27, 995-1003. Dearing, J. A. 1992. Sediment yields and sources in a Welsh upland lake-catchment during the last 800 years. Earth Surface Processes and Landforms. 17, 1-22. Einstein, H. A., 1950. The bed-load function for sediment transport on open channel flows. Tech. Bull. No. 1026, USDA, Soil Conservation Service, 71. Emmett, W. M. and Wolman, M. G. 2001. Effective discharge and gavel bed rivers. Earth Surface Processes and Landforms. 26. 1369-1380. Gomez, B. and Church. M. 1989. An assessment of bed load sediment transport formulae for gravel bed rivers. Water resources research, 25, 6, 1161-1186. Gomez, B. and Phillips, J. D. 1999. Deterministic uncertainty in bed load transport. Journal of Hydraulic Engineering, 125, 305-308. 17 Hayes, S. K., Montgomery, D. R. and Newhall, C.G. 2002. Fluvial sediment transport and deposition following the 1991 eruption of Mount Pinatubo. Geomorphology. 45, 211-224. Hoey, T. B., Ferguson, R., 1994. Numerical simulation of downstream fining by selective transport in gravel bed rivers: Model development and illustration. Water Resources Research. 30, 2251-2260. Hoey, T. B. and Sutherland, A. J. 1991. Channel morphology and bedload pulses in braided rivers: a laboratory study, Earth Surface Processes and Landforms, 16, 447– 462. Flemming, A. (1998) Swaledale: valley of the wild river. Edinburgh University Press, p138-140. Knox, J. C. 1993. Large increases in flood magnitude in response to modest changes in climate. Nature. 361, 430-2. Macklin, M. G, and Lewin J. 1989. Sediment transfer and transformation of an alluvial valley floor: The river south Tyne, Northumbria, UK. Earth Surface Processes and Landforms. 14, 233-246. Macklin, M. G., Lewin, J., 2003. River sediments, great floods and centennial scale Holocene climate change. Journal of Quaternary Science. 18, 101-105. Macklin, M. G, Rumsby B. T, and Heap T. 1992. Flood alluviation and entrenchment: Holocene valley floor development and transformation in the British uplands. Geological Society of America Bulletin, 104, 631-643. Milliman, J. D. and Syvitski, J. P. M. 1992. Geomorphic/tectonic control of sediment discharge to the ocean: the importance of small mountainous rivers. Journal of Geology. 100, 525-524. 18 Moog, D. B. and Whiting, P. J. 1998. Annual hysteresis in bed load rating curves. Water Resources Research. 34, No. 9, 2393-2399. Nicholas, A. P, Ashworth. P. J, Kirkby. M. J, Macklin M. G, and Murray. T. 1995. Sediment slugs: large scale fluctuations in fluvial sediment transport rates and storage volumes. Progress in physical geography, 19, 4, 500-519. Paola, C., 2003. Floods of record. Nature, 425, 459. Reid I, Layman JT, Frostick LE. 1980. The continuous measurement of bedload discharge. Journal of Hydraulic Research 18(3): 243-249. Ryan, S. E., Porth, L. S. and Troendle, C. A. 2002. Defining phases of bedload transport using piecewise regression. Earth Surface Processes and Landforms. 27, 971-990. Ryan, S. E., Porth, L. S. and Troendle, C. A. 2005. Coarse sediment transport in mountain streams in Colorado and Wyoming, USA. Earth Surface Processes and Landforms. 30, 269-288. Tucker, G. E. 2004. Drainage basin sensitivity to tectonic and climatic forcing: Implications of a stochastic model for the role of entrainment and erosion thresholds. Earth Surface Processes and Landforms. 29. 185-205. Tunnicliffe, J,., Gottesfeld, A. S. and Mohamed, M. 2000. High resolution measurement of bedload transport. Hydrological Processes. 14, 2631-2643 Van De Wiel, M. J., Coulthard, T. J., Macklin, M. G., and Lewin, J. In Press. Numerical modelling of alluvial landscape evolution. Geomorphology Whiting, P. J. and King, J. G. 2003. Surface particle sizes on armoured gravel streambeds: Effects of supply and hydraulics. Earth Surface Processes and Landforms. 28, 1459-1471. 19 Whiting, P. J., Stamm, J. F., Moog. D. B. and Orndorff, R. L. 1999. Sediment transporting flows in headwater streams. GSA Bulletin. 111. no. 3, 450-466. Wilby, R. L., Dalgleish, H. L. and Foster, I. D. L. 1997. The impact of weather patterns on historic and contemporary catchment sediment yields. Earth Surface Processes and Landforms. 22, 353-363. Willgoose, G. 2005. Mathematical modelling of whole landscape evolution. Annual review of Earth and Planetary Science. 33. 433-459. Wolman, M. G. and Miller, J. P. 1960. Magnitude and frequency of forces in geomorphic processes. Journal of Geology. 68, 54-74. 20 0.01 1000000 1000000 1000000 100000 100000 100000 10000 10000 10000 1000 1000 1000 100 100 100 10 10 10 1 1 0.1 0.1 1 10 100 10000.01 0.1 0.1 1 1 10 100 0.01 1000 0.1 0.1 0.01 0.01 0.01 0.001 0.001 0.001 0.0001 0.0001 0.0001 0.00001 0.00001 0.00001 0.000001 0.000001 0.000001 1 10 100 1000 1000000 y = 37.833x + 323.58 R2 = 0.0422 150 100000 100 10000 50 Linear (50) Daily sediment yield (m3) 1000 100 Linear (150) Linear (100) y = 6.4567x - 16.271 R2 = 0.5217 10 1 0.1 y = 2.9363x - 17.33 R2 = 0.306 0.01 0.001 0.0001 0.00001 0.000001 0.01 0.1 1 10 Mean Daily Flow (m 3 sec -1) 100 1000 Figure 1. This illustrates Qw and Qs plotted for 3 50 year sections of the data set, 50100, 100-150 and 150-200 cal. BP. Note, both axis are log scaled. The top three plots show each data set (50, 100 and 150) plotted separately. 21 10000 150 50 Pow er (150) 1000 Frequency Pow er (50) 100 y = 235347x -1.9608 R2 = 0.9345 10 50-100 cal. BP y = 5E+08x -1.9822 R2 = 0.9154 150-200 cal. BP 1 10 100 1000 10000 100000 Daily Bedload Yield (m3) Figure 2. Power law distribution of bedload yield for (a) time period 150-200 cal. BP and (b) 50-100 cal. BP. 1400 1800 Median daily sediment yield (m 3) 1200 y = 8.7701x - 71.444 R2 = 0.9406 1850 1900 1000 Linear (1900) Linear (1850) 800 Linear (1800) y = 4.7874x - 45.638 R2 = 0.9319 600 400 y = 1.2572x - 2.9453 R2 = 0.9709 200 0 0 20 40 60 80 100 3 120 140 -1 Mean Daily Flow (m s ) Figure 3. Medians of simulated sediment yields for the periods 1800, 1850 and 1900 Cal. BP. 22 2.5 100 100 4.3724x y = 0.0008e R2 = 0.4535 Climate 10 Qs/Qw slope Slope of Qs Qw 2 10 1 Climate proxy 0.01 1.5 1 0.001 0.5 1 climate index 1.5 2 2.5 1 0.1 0.5 Slope of Qs Qw relationship 0.1 0.01 0 0 1000 2000 3000 4000 5000 Years Cal. BP 6000 7000 0.001 9000 8000 100000000 100 100 0.7415 y = 0.0002x R2 = 0.6475 Sediment yield 10 10000000 Qs/Qw slope Slope of Qs Qw 10 0.1 0.01 100000 0.001 100 1000 10000 100000 1000000 10000000 1E+08 1 Sediment Yield 10000 0.1 1000 Slope of Qs Qw relationship Sediment Yield (m3 per 50 yrs) 1000000 1 100 0.01 10 1 0 1000 2000 3000 4000 5000 Years Cal. BP 6000 7000 8000 0.001 9000 Figure 4. Slope of Qs/Qw relationship plotted with the climate driver for the simulations (top) and the modelled sediment yield (bottom). 23 100000000 10000 Sediment yield Standard deviation 10000000 1000 1000000 Sediment Yield (m3 per 50 years) 100 100000 10 10000 1 1000 0.1 Standard Deviation and Median Median of 16-32 flow class 100 0.01 10 1 0 1000 2000 3000 4000 5000 Years Cal. BP 6000 7000 8000 0.001 9000 Figure 5. Standard deviation and medians of the 16-32 m3s-1 flood class plotted with the modelled sediment yield 100000 10000 nidd wharfe 1000 swale Daily Sediment Yield (m 3) 100 10 1 0.1 0.01 0.001 0.0001 0.00001 0.000001 0.01 0.1 1 10 3 100 1000 -1 Mean Daily Flow (m s ) Figure 6. Qs/Qw plots for the rivers Swale, Wharfe and Nidd. 24 Figure 7. Areas inundated by a 50 m3s-1 flow at 4500 (top) and 200 (bottom) Cal B.P. 25