How useful are catchment sediment budgets?

I Introduction

Catchment sediment budgeting is an attempt to identify the sources, sinks and pathways of eroded material within catchments (Slaymaker, 2003) or, as expressed succinctly by Beach (1994), to determine ‘the fate of eroded soil’. Typically, the outcome of such a study can be expressed in graphical terms as shown in Figure 1. It has been claimed (e.g. Ramos-Scharrron and MacDonald, 2007; Walling and Collins, 2008; Wilkinson et al., 2005) that sediment budgets are useful as the basis of catchment management, and that, through the use of sediment ‘fingerprinting’, sources of eroded material can be identified (e.g. Walling et al., 2002). This paper takes a critical look at this methodology and its claims. It examines the empirical base on which sediment budgeting rests and its conceptual framework, considers issues of budget closure and temporal scaling, assesses the validity of sediment fingerprinting and evaluates the role of sediment budgets in catchment management.

OPEN IN VIEWER

Figure 1.

Characteristic representation of a catchment sediment budget

II The empirical base

At the most elementary conceptual level, catchment sediment budgets consist of estimates of input (eroded material), output (stream sediment load) and storage (the difference between the two). Input of eroded material may comprise that which is sourced from interrill areas, rills and gullies, stream banks, and mass movements. Not all sediment budgets take account of all forms of input or output (see budget closure below). At its simplest, inputs to sediment budgets have consisted of estimates of interrill and rill erosion (so-called ‘upland erosion’, e.g. Banasik et al., 2005). These estimates have usually been obtained using models of soil erosion (most widely the Universal Soil Loss Equation, e.g. Banasik et al., 2005; Trimble, 1983) or from the extrapolation of plot measurements (e.g Wang et al., 2007). Inasmuch as the USLE is, itself, based upon measurements of erosion at the plot scale (see Wischmeier and Smith, 1978), both methods ultimately depend on a similar conceptual approach. Parsons et al. (2006) have argued this conceptual approach is flawed because of scale dependency of erosion rates (Evans 1992, 1995; Parsons et al., 2004). Fundamentally, plot-based measurements of erosion have assumed that the sediment recorded at the plot outlet derives from the whole area of the plot (i.e. flux has been interpreted as being convertible to specific yield). However, there is increasing evidence that this is not the case (e.g. Parsons et al., 1996; Rejman et al., 1999; Wilcox et al., 1997), and that during individual storm events, or over short periods of records (one or other of which applies to measurements made at the plot scale), the travel distance of sediment is actually quite small. Consequently, the flux of sediment recorded at the plot outlet may, largely, come from only a small, but unquantified, lower portion of the plot, but be entered into the sediment budget as a rate that is a function of the plot area.

An alternative method to measure erosion, that has been applied especially at the scale of interrill erosion (but which has also been applied to other forms of erosion) is the use of radionuclides, particularly ¹³⁷Cs. The argument in favour of this method is that it gives medium-term rates of erosion integrated over a period of decades (e.g. Longmore et al., 1983). However, this approach is not without its problems. In many studies the loss of ¹³⁷Cs is assumed to be proportional to the loss of soil (e.g. Brown et al., 1981). This assumption raises two issues. First, in order to convert the loss of ¹³⁷Cs to an amount of eroded soil, it has to be calibrated to a measured or modelled soil loss. Ritchie et al. (1974) correlated their loss of ¹³⁷Cs with USLE predictions, and used this correlation to produce the estimate of the amount of eroded soil. Thus, these medium-term estimates of erosion are, in fact, calibrated on the same plot data as the USLE. Rogowski and Tamura (1970), for example, specifically use the USLE to estimate erosion from their plots. Second, the assumption of simple proportionality with soil erosion ignores the known bias in adsorption of ¹³⁷Cs by some minerals (Francis and Brinkley, 1976). In addition, all studies that have used ¹³⁷Cs to measure erosion have assumed that all the fallout is deposited in the catchment and that the difference from a reference site that has undergone no erosion is a measure of soil erosion. However, in a laboratory study, Dalgleish and Foster (1996) found evidence to suggest that the fallout may be preferentially adsorbed onto eroding particles, so that in environments with significant erosion by Hortonian overland flow, erosion will be overestimated. For a fuller discussion of the difficulties of using ¹³⁷Cs to estimate rates of soil erosion, see Parsons and Foster (2011).

Inputs derived from other sources are even more difficult to assess. Gully erosion has been found to comprise between 10% and 94% of catchment soil erosion (Poesen et al., 2003). The significance of gully erosion for any particular catchment can be derived from mapping of the gully network. This method provides a quantitative measurement of total sediment input from the gully network to the catchment since gully initiation. Similarly, models to predict the threshold conditions for gully initiation are reasonably well developed (e.g. Begin and Schumm, 1979; Vandaele et al., 1996). However, neither gully mapping nor modelling of gully initiation estimates sediment flux from gullies. It has been argued that gully erosion is at its highest when the gully network is growing and that it declines as the network reaches the limits of growth defined by the threshold conditions (Nyssen et al., 2006; Rutherford et al., 1997). Sidorchuk (1999, 2006) has argued that gully evolution can be divided into two stages: active and stable. Whereas the former constitutes only about 5% of gully lifetime, it is responsible for 35% of gully volume. In attempting to account for the temporal change in contribution of gully erosion to a catchment sediment budget, Rustomji et al. (2008) used a scaling factor to account for this change in gully sediment through time. However, the use of such a scaling factor depends on knowing the age of the gullies (see also Wilkinson et al., 2009).

Because of their episodic nature, the contributions of mass movements to catchment sediment budgets are particularly difficult to determine, and the flux of sediment over any particular time period may show very large variations. The example shown in Figure 2 (taken from Matthews, 1999) illustrates this problem. As with gullies, mapping the distribution of mass movements, particularly using sequential aerial photography obtained over a long time period, can be effective in providing quantitative estimates of inputs of sediment to the catchment. Incorporating mass movements into sediment budgets presents further difficulties because of their units of measurement (see Nyssen et al., 2008). Mass movements move a relatively large volume of material a short distance at discrete locations. Any attempt to produce either an areal or temporal average over an entire catchment of sediment movement by this process is problematic.

OPEN IN VIEWER

Figure 2.

An assessment of the contribution of landslides to the sediment budget of the Noyo River

Prosser et al. (2001) suggested that channel banks are a more important source of sediment than hillslopes in many Australian catchments. However, far less attention has been given to methods to measure bank erosion than has been afforded to hillslope erosion. A common method to measure bank erosion is through the use of erosion pins (e.g. Bartley et al., 2007). Typically measurements of bank erosion are short term, and spatially limited, so that extrapolation both spatially and temporally is subject to significant uncertainty.

Output from catchments (sediment yield) is obtained from stream sediment-discharge data. Such data may be obtained from measurements of reservoir sedimentation, or from data on sediment concentration in stream discharge. The most common method for deriving the latter is to develop a sediment-rating curve based upon occasional sampling (Lu et al., 2005). Alternatively, continuous turbidity data are calibrated to suspended sediment load (Davies-Colley and Smith, 2001). However, Yuill and Nichols (2011) show that different grain sizes may have different transport patterns, so that attempts to define sediment yield using total sediment concentration may be erroneous. Data derived from different methods are not directly comparable (Lu et al., 2005), not least because of differences of timescale (see below) and the fact that reservoir-sedimentation data include coarse material, which is excluded from estimates of sediment yield derived by the other two methods. However, though inclusive of total clastic output, reservoir-sedimentation data are compromised by uncertainties about trap efficiency (for further discussion, see Verstraeten and Poesen, 2000).

The ratio of these catchment inputs to output has given rise to the concept of the sediment delivery ratio. In many studies of sediment budgets (e.g. Maner, 1958), it has been assumed that this ratio is an estimate of catchment storage, though seldom is this quantity actually measured, being simply inferred from the difference between the other two values. Trimble (1983) used soil profiles to locate buried pre-settlement soil horizons from which to estimate floodplain sedimentation, but acknowledged that spatial variability played a significant role in the uncertainty surrounding estimates based on such measurements. A significant problem of the sediment delivery ratio is the units in which the two components are measured. According to Glymph (1954) both sediment yield and erosion are measured in mass (or more strictly mass/unit time), but in reality both have been usually (but incorrectly) expressed in mass/unit area/unit time in order for their ratio to be calculated (e.g. Ebisemiju, 1990; Livesey, 1975). One reason for this use of units is to allow comparison of sediment yield with estimates of erosion derived from the USLE, or plot data, which have been expressed in mass/unit area/unit time. The issue is crucial because of the conceptual underpinning of the choice of units of measurement. Mass/unit time measures flux, and flux tells us nothing about the contributing area. It is quite possible for the flux at one point in the landscape (near the catchment head) to be the same as that at another (near the catchment outlet) without there being any storage of sediment occurring. Such a situation would apply when the upper part of the catchment was the source of all eroded material that was then transmitted by streamflow to the catchment outlet. However, if both measurements of flux are divided by their respective upstream areas (as is typically the case), the latter will be smaller than the former. Much of the inferred storage of sediment within catchments may be due to this misuse of units of measurement (see Parsons et al., 2006, for a more extensive discussion of the problems of the concept of the sediment delivery ratio).

III The conceptual framework

In developing sediment budgets, crucial questions concern the timescale that the budget is valid over, and whether the constituent components are measured over the same timescale. Clearly, any budget that has input and output unequal is untenable in the long term. As Graf has commented, in the long term the delivery ratio concept is untenable because all long-term delivery ratios must approximate 1. Unless the flux of sediment through the river channel to the catchment outlet matches that from the hillslopes, catchments would be progressively filling with sediment (Graf, 1988). Consequently, any sediment budget that includes a storage term is, by definition, timebound. Furthermore, and probably more crucially, it assumes that the catchment is in process equilibrium over that timescale, i.e. the inequality of input and output is steady over the timescale. Lu et al. (2005) point out that Wischmeier and Smith (1978) have argued that a minimum of 22 years of data are required to obtain long-term average rainfall erosivity. Consequently, ‘at the plot scale, a few decades of erosion measurements are needed to get statistically meaningful long-term average erosion rates’ (Lu et al., 2005: 5), and ‘characteristic timescales of sediment yield at the catchment scale can be expected to be much longer’ (p. 5). In reality, few studies show such careful attention to appropriate timescales of measurement: data on catchment input and output may be measured over different timescales, and/or at different time periods. Trimble (1983) attempted to produce a sediment budget for Coon Creek for the period 1853–1977, i.e. commencing at the time of European settlement. The study illustrates the problems of compatibility of temporal scaling of data. Suspended sediment data are available for the period 1934–1938 and yield a long-term estimate of sediment output (including bedload) of 160 Mg km^-2 a^-1. However, Trimble argued this is a poor estimate of average sediment yield since the time of European settlement, because of likely changes over that time. Instead he used data from a nearby, similar catchment in which a reservoir had been constructed in 1867 that was surveyed in 1939 and for which repeat surveys were made in 1976 and 1977. Taking into account trap efficiency of the reservoir, Trimble estimated a sediment yield of 105 Mg km^-2 a^-1 over the period 1867–1939. To obtain upland interrill and rill erosion, Trimble used the USLE for conditions at the time of European settlement (1853), in the 1930s (considered to be the period of peak erosion) and in the 1970s. He constrained these three estimates by an independent estimate of erosion in the 1930s based on an upland debris basin, and then integrated the area under the curve drawn through these data points to obtain total interrill and rill erosion over the periods of interest. No data are given on the rainfall data used to calculate rainfall erosivity in this use of the USLE. Floodplain storage was obtained from auger borings and mapping of historical accumulations, and Trimble assumes that this rate can be applied throughout the catchment. Because sedimentation rate (on this assumed timescale) exceeds the calculated interrill and rill erosion, Trimble attributed the difference to channel and gully erosion. As the former had been measured, the residual was ascribed to gully erosion (thereby achieving budget closure – see below). Although Trimble’s study gives closer attention to relevant timescales of measurement than many, it is evident from his honest analysis of his data that there are significant assumptions about the temporal scales at which processes operate and the relevant measurement periods that underpin the sediment budget that is produced.

The concept of equilibrium provides a more general underpinning to attempts to produce sediment budgets. Bracken and Wainwright (2006), in an extensive analysis of the concept of equilibrium in geomorphology, demonstrate that the term has been widely interpreted and used in different ways. Here, in referring to process equilibrium, the term is being used to imply that the processes are steady, at the very least over the period and spatial scale for which the sediment budget is being developed. However, it does not need to imply that catchment form is also in steady state. What is essential, on the other hand, is that all processes that are providing input to or output from the sediment budget are steady over the same timescale. Bracken and Wainwright argue that because landscape is a complex system the notion of equilibrium may be invalid and, even if it is not, specific instances of equilibrium may be untestable. Current understanding of gully evolution (Sidorchuk, 1999, 2006), for example, would imply that the notion of equilibrium is invalid for this form of sediment input. To the author’s knowledge, no instance exists of a sediment budget for which process stability has been successfully tested. The study by Tunnicliffe and Church (2011) emphasizes the argument made by Bracken and Wainwright by showing that as spatial scale is changed different disequilibria are identified.

IV Issues of budget closure

Budget closure is usually achieved by lumping errors in the measured terms into one or more unmeasured terms (Kondolf and Matthews, 1991). These authors demonstrate that the apparent contributions of unmeasured elements may be substantial. They cite Kelsey (1980) in which fluvial hillslope erosion was computed as a residual term, and that this residual term accounted for 66.1% of the sediment budget. They further demonstrate (Kondolf and Matthews, 1991: Table 1) that unmeasured large residuals are by no means atypical. They advocate that unmeasured components of sediment budgets that are obtained by subtraction should be identified as such and their values treated with caution.

Not only does the practice of closing budgets via unmeasured terms hide the uncertainty associated with the measured terms, but it also assumes that the budget being measured is a closed system. This can seldom be taken for granted. Catchment sediment budgets typically concern themselves with clastic sediment, and sometimes not even all of that (e.g. Fan and Cai, 2005). Where output is derived from measurement of suspended-sediment concentration or turbidity data, no account is taken of bedload transport. While suspended load constitutes the dominant clastic river load for lowland rivers in humid environments, this is not the case elsewhere. In semi-arid and arid environments, in particular, bedload comprises a large proportion of stream clastic discharge. Although this is generally less than 50% (see Graf, 1988: 139; Powell et al., 1996), Rovira et al. (2005) estimated more than 75% for a catchment in NE Spain. Such high bedload amounts are attributed to the ephemeral flow and lack of channel armouring (Laronne et al., 1994; Reid and Laronne, 1995). Likewise, bedload is likely to constitute a high proportion of sediment load of streams in mountain catchments (e.g. Johnson and Warburton, 2002). Furthermore, in order to derive suspended-sediment load from measurements of suspended-sediment concentration, or turbidity, and water discharge, it is implicitly assumed in such a calculation that the velocity of the suspended sediment is equal to the velocity of the water (Bennett, 1974), though no evidence in support of that assumption exists (see Wainwright et al., 2008, 2010). In many cases, no consideration is given to the dissolved load of catchment (other than such loads are subject to a separate budgetary analysis). Yet it cannot be assumed that sediment mobilized as clastic load will leave the catchment as such. In humid landscapes, dissolved load may account for upwards of 75% of all catchment output (Meybeck, 1976). Though it is certainly true that much of this may be both mobilized and leave the catchment in a dissolved form, given estimates of storage time for sediment on floodplains (c.1000 years, Leopold et al., 1964) it is unlikely that dissolution of some of sediment mobilized as clastic load will not take place, so that it leaves the catchment as dissolved load. Even if all forms of sediment transport were taken into account in a sediment budget, the issue of budget closure would not necessarily be resolved because of problems of temporal scaling associated with each type of measurement. Returning to the issue of process stability, if the dissolved load is measured at the catchment outlet and some of that dissolved load derives from clastic sediment stored on floodplains, then such a measurement assumes that process stability exists for c.1000 years. Differential rates of adsorption of particles of different sizes further complicate efforts to derive a budget from measurements at the catchment outlet.

V Sediment sources

As well as quantifying the process mechanisms (rills, mass movements, etc.) that contribute to the eroded sediment within catchments, sediment budgeting has also attempted to identify the spatial distribution of their sources. An approach that has been widely used to achieve such source recognition has been through techniques of sediment ‘fingerprinting’ (e.g. Peart and Walling, 1988). Two principal techniques exist: one relying on the exposure of different locations to incoming radionuclides, the other based on differences in chemistry of the weathered products in catchments underlain by diverse lithologies. Both techniques rely on mixture modelling. This technique uses the composition of the potential sources and that of the catchment output to derive the proportion of the output that derives from each of the sources. However, a fundamental problem of the approach is that it provides the ultimate (i.e point of origin) source of the sediment yield from that catchment, rather than the proximate source (i.e the point from which it was last moved), and takes no account of the timescale over which sediment moves through a catchment. Different components of catchment sediment yield will have travelled different distances from their ultimate source and, almost certainly, at a variety of rates that will vary with grain size. Therefore, unless it can be assumed that conditions of sediment mobilization within the catchment have remained unchanged and that the rates of movement of sediment of different sizes and from different sources are known, mixture modelling of the sediment yield may not be very informative of the sediment flux through a catchment over any clearly defined timescale. Such problems become particularly acute when using sediment fingerprinting in large catchments (e.g. Garzanti et al., 2006). Here, travel distances, and relevant timescales of erosion phases become crucial in sound interpretation sediment sources.

VI Management

The extent to which any methodology can be used for management depends on the reliability of the science that underpins it. Despite the claims that have been made for the utility of sediment budgets, this review suggests there are some serious weaknesses in the methodology that suggest the actual management value of sediment budgets is limited. At the very least, as Kondolf and Matthews (1991) propose, the errors associated with both measured and unmeasured components of sediment budgets need to be identified so that catchment managers are not misled by the apparent precision of sediment budgets. In a discussion of soil erosion rates in the United States, Trimble and Crosson (2000) point to the very large disparity between presumed catchment erosion and measured sediment yield, and the contradictions between this disparity and the observational record of alluviation. These authors conclude that ‘the uncritical use of models is unacceptable as science and unacceptable as a basis for national policy’ and that ‘the limitations of the USLE … are such that we do not seem to have a truly informed idea of how much soil erosion is occurring’. Consequently, it is difficult to use sediment budgeting to determine the fate of eroded soil when that very quantity is, itself, subject to significant uncertainty. Several fundamental weaknesses of the sediment budget methodology can be identified. First, the timescales are often poorly defined and not consistent across the elements that are included in the budget. Walling et al. (2002) recognized this problem in their study, and argued that, since land use had remained constant, the problem was unlikely to be of great significance. However, since it is land use that is the most easily modified variable in catchment management, or past management, this may not always be the case. Indeed, the practical utility of sediment budgeting is almost predicated upon the idea of land-use change (see, for example, Reid and Dunne, 1996:1). Second, the methodology takes no account of the time taken for sediment to move through a catchment. This issue has already been alluded to with reference to sediment fingerprinting. For river habitats, however, this issue may have far greater significance. Of particular importance is the passage of large-scale bed waves, or sediment slugs (James, 2006). As Nicholas et al. (1995) note, anthropogenic influences that lead to significant volumes of sediment in rivers (e.g. mining) have been related to sediment slugs that may persist for hundreds to thousands of years. Similarly, Church and Slaymaker (1989) have argued that catchments in British Columbia are still recovering from the effects of sediment inputs from the last glaciations. The timescales of catchment response, particularly to significant input of river bedload, limits the management potential of sediment budgets. Related to this weakness is the emphasis that has been placed on sediment budgets on suspended sediment. As Bartley et al. (2007) comment:

The majority of sediment budgets that have been developed using field measurements deal primarily with suspended load only … Bedload fractions may never reach coastal areas, yet they can still have considerable impact on the habitat structure and function of freshwater reaches. (Bartley et al., 2007: 304)

VII Conclusion

This review of catchment sediment budgeting has shown that, despite a long history of research, budgeting remains problematic. There are practical difficulties in determining the quantities involved in both sediment inputs and outputs, and most of all in sediment storage. These practical difficulties are exacerbated by conceptual issues surrounding notions of process equilibrium, units of measurement, and unquestioning reliance on the use of modelling to determine some of the quantities involved. These uncertainties, coupled with problems of multiple timescales of sediment movement severely limit the practical utility of sediment budgets for catchment management. Three guiding principles should be adhered to whenever a sediment budget is produced with the aim of guiding land managers. First, no sediment budget should be produced without an explicit statement of the timescale over which it is purported to be valid, or without a demonstration of process stability over that timescale. Second, no sediment budget should include unmeasured elements, the values for which are determined simply by subtraction on the assumption of budget closure. Third, any sediment budget must provide estimates of uncertainty associated with any reported value. Figure 1 is useless unless the reported values show some indication of the error margin associated with each of them. Sophisticated models that are used to provide sediment budgets (e.g. Wilkinson et al., 2009) must include some indication of parameter uncertainty and its implication for model uncertainty (see Beven, 2006; Beven and Freer, 2001). It may well be the case that the uncertainty with which a catchment sediment budget can be specified exceeds the magnitude of the effect of a land-use change.

Aside from its potential practical uses, Kondolf and Matthews (1991) conclude the exercise of sediment budgeting is useful because it requires that sediment sources and pathways be identified and, to some extent, quantified. Sediment budgeting can act as a framework for a research agenda on catchment processes. Specifically, it allows researchers to identify the salient gaps in current understanding of sediment movement through catchments. What has been lacking from this agenda has been an adequate consideration of the time taken for sediment to travel via the various pathways to the catchment outlet. The storage term in such budgets has been used as a poor substitute for a thorough understanding of sediment velocity through catchments, and there is an urgent need for more emphasis on this topic.