You are on page 1of 37

A.I.J.M. van Dijk (2002) Water and Sediment Dynamics in Bench-terraced Agricultural Steeplands in West Java, Indonesia.

PhD Thesis, Vrije Universiteit Amsterdam

Chapter 13 Measurement and modelling of erosion and sediment transport on bench terraces

Abstract Despite widespread bench-terracing of rain-fed hillsides soil loss rates in upland West Java remain high. This chapter reports on measurements of sediment yield from terrace risers and beds and from terrace units made during three consecutive seasons. The results demonstrate that soil loss from the terraces takes place in two stages: rainfall-driven transport by splash and shallow overland flow (wash) from the terrace riser and bed to the toe drain is followed by a combination of onwards wash transport of fine sediment and runoff entrainment of coarser sediment deposited in the toe drain. A model (TEST: Terrace Erosion and Sediment Transport) was conceptualised, describing these processes as a function of vegetation and soil surface cover and the presence of a layer of deposited sediment. It uses as few parameters as possible, while at the same time retaining physical process description. The model was calibrated with some of the sediment yield data sets and hence tested by simulating the other data sets. The results were generally satisfactory: modelled fractions of sediment transported by splash, wash and flowdriven runoff transport compared well to observed fractions of fine and coarse particles in eroded sediment from the various plots. Parts of this chapter are submitted to Journal of Hydrology with L.A. Bruijnzeel as co-author

13.1. Introduction The combination of young steep geology, seasonal humid tropical climate and high population pressure and poverty that is found on the Indonesian island of Java has resulted in a situation in which the upland agricultural hillsides are eroding at alarming rates. Whereas this problem was acknowledged as early as the 1930s, many past attempts to bring land degradation to a halt have failed, either because the proposed changes of land use management were not adopted or because they did not bring about the expected effects (Nibbering, 1991). For example, Purwanto (1996) estimated that currently over 80% of Javas cultivated uplands have been terraced, yet river sediment loads continue to be high. This has prompted attention to sources of sediment other than the unirrigated hillsides such as landslides, bank erosion, expansion of paddy rice fields and contributions from roads and villages (Carson, 1989; Diemont et al., 1991). It has also been suggested that the (temporary) storage of sediment in and near rivers may give rise

259

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES to considerable delay between the introduction of soil conservation and the first signs of reduction in downstream sediment discharge. However, it has also been shown that (back-sloping) bench terracing was not always instrumental in reducing soil loss to acceptable levels. For example, Critchley and Bruijnzeel (1996) studied erosion on back-sloping bench terraces on volcanic Hapludalf soils in south-central Java, Indonesia, and observed terrace soil losses of 9-24 t ha-1 in four months. Similarly, recent research in the volcanic Cikumutuk catchment in West Java suggested that about two-thirds of the annual stream sediment load (49-63 t ha-1 in 1994/95 and 1995/96) was generated on the terraced rainfed hillsides (Purwanto, 1999). In both cases, important reasons for the lack of effectiveness of terracing as a means of soil conservation were thought to be related to the sparse vegetation cover maintained on the bench terraces, and particularly on the terrace risers. In addition, sediment transport from terraces was influenced by the amounts of runoff generated and the gradient and management of the toe drain (Critchley and Bruijnzeel, 1996; Purwanto, 1999). The economically vulnerable position of Javas rural population warrants a thorough and careful reassessment of causes and consequences in the area of soil erosion and land degradation in the uplands before adopting new panacea. Thus, there is a distinct need for a better understanding of the processes involved in erosion from terraced fields. Recent research in the Cikumutuk catchment has shown that soil loss from back-sloping bench terraces occurs in three phases, viz.: (i) impacting rain drops detach soil particles on the terrace riser and bed and cause net splash transport to the central toe drain; (ii) unconcentrated runoff carries some of the fine particles detached by rainfall in suspension and washes them to the toe drain and, ultimately, away from the terrace; and (iii) sediment deposited in the toe drain may again be detached by rainfall or entrained by concentrated runoff, part of which is transported away from the terrace (Chapter 9 and 10, Purwanto, 1999). Clearly, the black-box approach represented by the conventionally used (Revised) Universal Soil Loss Equation (Renard et al., 1997) does not include an explicit representation of these processes. In particular, it fails to describe the lateral flow (i.e. through the central drain, parallel to the slope) of runoff and sediment that is involved. Although a supporting practice (P) factor has been introduced to simulate the effects of terracing on soil loss (Renard et al., 1997), this was not intended to be used for the relatively narrow, back-sloping bench terraces that are found in Javas uplands. In addition, and perhaps not surprisingly in view of its empirical nature, the factor has been shown to be rather unreliable and ill-defined (Toy et al., 1999; Critchley, 2000). Nonetheless, a wide range of P or CP factors has been used in this environment, adding to general confusion (Agus et al., 1998). Additional uncertainty surrounding the (R)USLE in the bench-terraced Javanese environment can be ascribed to the obstruction of runoff created when Wischmeier-type plots are superimposed on such terraces in order to derive equation parameters from measurements of sediment yield (Bruijnzeel and Critchley, 1996). More in general, the (R)USLE technique has been tailored to predict erosion rates on agricultural fields in the U.S.A. and the reduction in these through soil conservation practises such as changing the direction and timing of tillage. Applying this technique in Java may be considered a questionable extrapolation to a situation with a high rainfall climate, different soils, steeper slopes, smaller field sizes (and therefore shorter slope lengths), and different agricultural practices, both in terms of crop varieties and tillage practises. Although some progress has been made in calibrating the (R)USLE for rainfall regimes and soils found in the humid tropics (e.g. El-Swaify and

260

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES Dangler, 1976; Ambar and Wiersum, 1980; Undang Kurnia et al., 1986), these adaptations have not been tested as rigorously as in the U.S.A. In the current study a more rigorous, process-based approach was taken to describe, explain and predict the sediment sources and pathways on back-sloping bench terraces. To this end a process model called TEST (for Terrace Erosion and Sediment Transport) was developed in which the various processes and their interactions are explicitly expressed. Because runoff entrainment and transport represents the last essential step in soil loss from the bench terraces, it is imperative for such a model that the generation of runoff and the patterns and rates at which it occurs are described accurately. Therefore considerable attention was given to modelling runoff generation and the results of this are presented in Chapter 12. In the present Chapter, the parameterised runoff model discussed in Chapter 12 is used to model sediment yield from sections of terrace beds and risers and from terrace units. The characteristics of the studied environment are presented in Chapter 3, whereas the overall methodology to measure runoff and sediment yield and the dimensions and treatment of the various erosion plots are described in Chapter 12. 13.2. Model theory 13.2.1. Rainfall-driven erosion A distribution theory for the interpretation and prediction of rain splash transport on a horizontal surface was advanced in Chapter 8, based on the assumption that splash of particles from the point of raindrop impact results in an exponential radial distribution of particles. Based on this distribution theory, for which there is experimental support, expressions were derived for splash transport across a boundary. The resulting equation relates the amount of splash detachment ( in g m-2) to transport across the boundary, expressed as mass per unit boundary length (Tsp in g m-1). The connecting variable is the average splash length (% in m) corresponding with the exponential distribution. The resulting relationship is (cf. Chapter 8):
Tsp =

[13.1]

Using the same exponential distribution theory, equations were developed that can be applied to interpret results obtained with various devices designed to measure rain splash detachment and transport. The theory was successfully used to interpret measurements of rain splash on the studied bench terraces using splash cups on the terrace bed and a combination of a splash box and a Gerlach-type trough on the riser (Chapter 9). Similarly, measurements of splash in a soil tray equipped with splash guards could be interpreted with the developed theory (Chapter 10). While the use of essentially horizontal theory inevitably introduced some errors, it proved to be a valuable tool for describing splash detachment and transport. It led to a number of conclusions with regard to the studied bench terraces: (i) rain splash was well correlated to the kinetic energy of rain falling at intensities higher than a threshold intensity of 20 mm h-1 (E20 in J m-2; Chapter 10; cf. Hudson, 1965);

261

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES soil detachability (D20 in g J-1), defined as the amount of sediment detached per unit effective kinetic energy (E20), was high and appeared to be independent of slope gradient; and (iii) the average distance and relative importance of splash in the downslope and upslope directions increased and decreased, respectively, with increasing slope. These findings were incorporated in a framework for modelling rain splash on the terrace risers and beds as a function of slope gradient. The following equation describes this relationship (cf. Chapter 10): (ii)
Tsp = 1 1 S = 1 + S p D20 E20

[13.2]

where %S (in m) represents the apparent splash length in the downslope direction, which can be expressed as a function of the horizontal splash length (%) and slope gradient (S) via the dimensionless coefficients & and p, for which respective values of 0.35 and 0.44 were derived in Chapter 10. This results in an increase of apparent downslope splash length from about 1 cm on a horizontal surface, to ca. 12 cm on a 64% (40-) slope (cf. Chapter 9). As detachment appeared independent of slope gradient, this implies a similar increase in splash transport with slope. In principle, net downslope transport will be smaller than downslope transport per se, since it is to be expected that some sediment will also be splashed upslope. However, the latter decreased very rapidly with slope (cf. Chapter 10). Moreover, the design of the terrace risers largely prevents this upslope movement, as is discussed in Section 13.6.3. Based on field observations it was hypothesised that runoff created on the terrace riser or bed and flowing towards the central toe drain in an unconcentrated manner did not have enough erosive power to actively entrain particles. The validity of this hypothesis is discussed in Section 13.6.3. In such a case, sediment is detached only by impacting raindrops. Some detached sediment particles may be transported a distance before settling, while others will be so small that they will not settle again before leaving the terrace unit. The latter will be indicated here by the term wash. A quick survey of literature on the subject suggests that there has been significant recent progress in the understanding and physical description of rainfall-driven runoff transport in laboratory conditions, but the translation of these findings to the field situation meets with serious problems (Chapter 10). An alternative, more pragmatic approach was therefore proposed in Chapter 10. It was demonstrated that the absolute upper limit to the concentration of sediment in water from rain drops that have just hit the soil can be approximated as the ratio of the total rate of detachment () associated with an event over event rainfall depth (P in mm). It is furthermore assumed that a fraction j of this detached material is small enough to be transported from a given area before settling, and that this washed fraction is independent of storm characteristics. In that case, event soil loss by wash erosion (Mwa in g m-2) is the product of the average concentration of washed sediment (cwa in kg m-3) and storm runoff depth (Qtot in mm):
M wa = c wa Qtot = j

Qtot = jD20 E 20 rc P

[13.3]

where rc is the runoff coefficient (rc=Qtot/P). Despite obvious simplifications inherent to Eq. [13.3] it provides a clear and consistent relationship between rainfall detachment and

262

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES wash transport and was successfully used to interpret measurements of sediment transported by wash from 0.3x0.6 m soil trays subject to natural rainfall (Chapter 10). Values for the washed fraction (j) ranged between 2.3 and 4.4% of total detachment for trays with bare, non-tilled soil (Chapter 10). Eqs. [13.2] and [13.3] can be combined to yield the mass of sediment transport per unit area of terrace riser or bed. To this end, splash transport rate given by Eq. [13.2] needs to be divided by the projected width of the terrace riser or bed (W in m) to derive splash erosion per unit area (Msp in g m-2):
M sp + M wa = Q 1 S 1 + cwa Qtot = 1 + S p D20 E20 + jD20 E20 tot W W P

[13.4]

On a very small area the use of Eq. [13.4] will introduce an error because material splashed out of the area is not available for wash transport. However, given the contrast between apparent splash length (0.01-0.12 m) and terrace riser or bed width (typically 0.5-2.5 m; Table 12.2) this error was neglected in the current setting. An exponential rainfall depth-intensity distribution was assumed in Chapter 11 to show that effective storm kinetic energy (E20) can be estimated with considerable accuracy from the depth-averaged rainfall intensity ( in mm h-1) and total rainfall depth (P) of a storm. The equation reads (cf. Chapter 11):
20 a 20 E 20 = Pemax exp   exp 20b  R R bR + 1

[13.5]

where emax (in J m-2 mm-1), a and b (in h m-1) are parameters that define the exponential relationship between rainfall intensity (R in mm h-1) and kinetic energy content (eK in J m-2 mm-1 ; cf. Kinnell, 1981; Rosewell, 1986): eK = emax [1 a exp( bR )] [13.6]

Existing studies of eK-R relationships were reviewed in Chapter 7, where it was found that using values of emax= 28.3 J m-2 mm-1, a=0.52 and b=0.042 h mm-1, produces good estimates in most environments, provided a correction is made for the effect of altitude. 13.2.2. Runoff-driven erosion Unlike rainfall-driven processes, runoff entrainment and transport have received considerable attention from physical modellers. Examples of developed models include WEPP (Laflen et al., 1997), EUROSEM (Morgan et al., 1992), LISEM (De Roo et al., 1996) and GUEST (Yu et al., 1997a). Of these, GUEST arguably is the most intensively validated and tested in tropical environments (Soil Technology, 1997; Yu and Rose, 1999). The GUEST model theory has received extensions and adaptations in the recent past, but the basic concept has remained the same. Derivation of the model theory will not be repeated here, but may be found in e.g. Hairsine and Rose (1992a, 1992b) and Rose (1993). The model is essentially based on the assumption that the maximum concentration of sediment that can be maintained in flowing water under equilibrium

263

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES conditions results from a balance between the rates at which particles settle out of suspension and new particles are entrained. Entrainment is considered to be a function of stream power per unit area (S in W m-2) defined as (Bagnold, 1977):
= e gSR H V

[13.7]

where e (in kg cm-3) is the suspension density, g the gravitational constant (9.81 m s-2), RH (in m) the hydraulic radius and V (in m s-1) flow velocity. RH and V are related and can be estimated from runoff rates using Mannings equation. With the condition that -2 rhrrprrqhhyyuruyqhyr 0 (ca. 0.008 W m , Proffitt et al., 1993a), the balance between deposition and entrainment can be expressed as (Yu et al., 1997):
ct

gD = F

[13.8]

where ct (in kg m-3) is sediment concentration at the transport limit (i.e. the maximum sustainable concentration), and (in kg cm-3) sediment and water density, respectively, D (in m) the depth of flow, (in m s-1) the depositability (i.e. the average sediment settling velocity) and F the fraction of stream power effective in erosive processes. Extensions and adaptations to Eq. [13.8] relate to, inter alia, the possibility that: (i) runoff may be concentrated in rills (Hairsine and Rose, 1992b); (ii) the soil surface may be only partially immersed (Lisle et al., 1995); (iii) some of the erosive stream power is used to entrain new, cohesive soil and therefore not effective in sustaining sediment concentration, introducing an erodibility coefficient ( <1; Rose, 1993); and (iv) a fraction H of the coherent soil is covered with more easily re-entrainable sediment (Rose, 1993). For modelling purposes, a method has also been developed based on Mannings equation to calculate an effective runoff rate (Qe in mm h-1) that can be used as a substitute for instantaneous runoff rates (cf. Chapter 12). This eventually resulted in an equation that can be used to estimate soil loss (Men in g m-2) by flow-driven runoff erosion (entrainment) as the product of the concentration of entrained sediment (cen in kg m-3) and storm runoff depth:

M en = c en Qtot = c t Qtot = kQe

0.4

Q tot

[13.9]

where c t (in kg m-3) is the flow-weighed average concentration of entrained sediment at the transport limit and k a coefficient that is assumed to be constant during the event or during a series of events. Including the fractional cover of the deposition layer (H=1 at the transport limit by definition; Rose, 1993), the parameter k is given by: HFSL0.4 k= ( 1) e S n
0.6

[13.10]

where n (in s m1/3) is Mannings roughness coefficient, relating flow velocity to discharge (see Section 13.3.3). Despite the well-tested physical basis of these equations, a number of simplifications and assumptions needs to be made in their derivation, which are discussed in e.g. Rose (1993), Rose et al. (1997a) and Yu (in prep.). Nonetheless, Eq.

264

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES [13.10] has been used with reasonable success to simulate measured sediment concentrations from agricultural plot studies in Australia, Thailand, Malaysia, The Philippines and China (Soil Technology, 1997; Yu et al., 1999). The already-mentioned exponential rainfall depth-intensity distribution was used to derive expressions linking effective runoff rate (Qe) and total runoff depth (Qtot) to , requiring knowledge of maximum average infiltration capacity (Im in mm h-1) and initial additional infiltration (SI in mm). The expressions were tested successfully in Chapter 12, though it appeared that the runoff response from an entire terrace unit was not simply the ~ sum of its parts. Instead, spatially-averaged effective values for m and S I needed to be ~ ~ used to derive total runoff depth ( Q tot) and maximum effective runoff rate ( Q e) on a ~ terrace unit basis. Comparison of calculated Q e values with those observed from bench terrace units suggested that the two did not differ systematically at the temporal resolution of the measurements (cf. Chapter 12). For modelling runoff entrainment on the bench terraces, a provisions was to be made for the fact that runoff is accumulated in the central toe drain:

~ W ~ Qe terrace Qe Wdrain

[13.11]

13.2.3. The effect of vegetation and soil surface cover and model integration Cover provided by the vegetation canopy or other forms of soil surface cover (e.g. mulching or stones) are well-known to reduce soil loss. The ratio between erosion from soil with and without a fraction C covered by mulch may be called the mulch factor (MF) and has often been described by exponential decay functions (Laflen and Colvin, 1981):

MF = e

[13.12]

where is a coefficient between 3 and 7 if mulch cover is expressed as a fraction. A similar relationship may be used to describe the effect of other types of cover that can be expected to have different values of . The value will also depend on the erosion processes that are dominant. Here, it was assumed as a working hypothesis that for a particular combination of vegetation cover (Cv) and soil surface cover (Cs) on a terrace bed or riser, and given the corresponding dominant erosion mechanisms, the resulting decrease in erosion could be described by:

M* =e

V CV

S C S

[13.13]

where M* and M denote cumulative soil loss through the various processes with and without cover, respectively, and v and s are coefficients associated with the vegetation and soil surface cover at hand. Finally, a number of simplifications needed to be made in order to integrate the various model components presented in the preceding paragraphs. The most important simplifying assumptions were: 1. all sediment transported from the terrace riser and bed by rain splash was first deposited in the terrace toe drain;

265

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES 2. all sediment transported from the terrace bed and riser by wash travelled onwards and left the terrace unit; 3. during each event entrainment was assumed to occur only after soil was transported from the terrace riser and bed to the toe drain; 4. the fraction of the toe drain covered by deposited sediment (H) was assumed to have an exponential relationship with the amount of sediment per unit area (m in kg m-2); and 5. mulch cover affected all erosion processes in the toe drain equally. During an event new sediment is deposited in the drain while sediment is simultaneously carried by the flow. The process was treated as a running mass balance, but because a single value for H was needed for each storm, assumption (3) had to be made. The sediment amount per unit area of toe drain that was available during the i-th storm in a series of storms, denoted by mi, was calculated as:

W W mi = mi 1 M drain ,i 1 + M sp ,i + M sp ,i W W bed riser drain drain

[13.14]

where Mdrain,i-1 (in kg m-2) denotes the loss of sediment from the toe drain during the preceding event and Msp,i (in kg m-2) the loss of soil from the terrace riser or bed by splash during the considered event (again, all washed sediment was assumed to travel onwards without being deposited, conform assumption (2)). The resulting value of mi was used to estimate H for that event. This was done by assuming an exponential relationship between m and H, having the form:

H = 1 e

H m

[13.15]

where H (in m2 kg-1) is the coefficient relating the two. The resulting value of H was used to model sediment entrainment with the theory outlined in Section 13.2.4. Finally, transport of toe drain sediment detached by rainfall through wash and splash was also estimated using the theory explained in Section 13.2.3, with appropriate values for toe drain gradient, length and width.
13.3. Materials and methods

13.3.1. Introduction
This section deals with measurements of sediment yield from terrace risers and beds and from terrace units, as well as with measurements and experiments designed to develop, calibrate and test the model outlined in the preceding sections. Data on rainfall intensity and runoff are essential for this purpose and are described in Chapter 12. Experiments on splash and wash erosion using field devices and soil trays under natural rainfall yielded most of the parameters needed for the model components presented in Section 13.2.1; these are described in Chapters 9 and 10. Characteristics of the various plots from which runoff and sediment yield were measured are described in Chapter 12 (Tables 12.2 and 12.3). The manner in which sediment yield was measured is discussed

266

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES


below, as well as the flume experiments designed to calibrate the runoff-driven model component.

13.3.2. Plot sediment yield from terrace riser, bed and unit erosion plots
Runoff in the collecting drum of terrace riser and bed plots (Artificial Boundary Erosion Plots or ABEPs) and in the collecting basin of the terrace unit plots (Nonimposed Boundary Erosion Plots or NBEPs), as well as in the drums of the divider system when present (cf. Chapter 12), was sampled after each storm to determine the amount of sediment that was still in suspension at the time of sampling. A depthintegrated sample of runoff was taken using a 0.6 l bottle and in the laboratory some coagulating agent (Al2(SO4)3 (aq)) was added. After the sediment had settled a few days later, most of the water was carefully siphoned out of the bottle; the remaining suspension was transferred to glass beakers and oven-dried at 80- C. The residue was allowed to cool and attract air moisture for more than five hours before the weight of the sediment was determined. Tests showed that desiccated sediment gained 1-4% weight in this time depending on air humidity. Comparing the weight of dry residue from water with and without suspended sediment, as well as with and without coagulating agent, suggested that both slight under-estimations (because sediment was lost in the process) and over-estimations (because of the inclusion of dissolved salt in the dry residue) occurred, but differences were within an equivalent 0.08 g l-1 either way and no corrections were made. The determined concentration was used to calculate the amount of suspended sediment yield corresponding to the measured runoff volume. The bed load sediment left on the bottom of the ABEP collecting drums was measured in different ways depending on the volume. If the volume of sediment was less than about one litre, it was all brought to the laboratory and oven-dried at 80- C. Greater volumes were measured using a graded cylinder; on several occasions, the whole volume of sediment was oven-dried to obtain a factor for converting wet sediment volume to dry mass. Even greater sediment volumes were sometimes found in the NBEP collecting basins and were measured by moulding the material into a rectangular shape and measuring its size. On several occasions a sub-sample of the sediment was taken using a 100 cm3 soil core cylinder and oven-dried to enable conversion of volume to dry mass. In all cases, sediment was allowed to attract air moisture after oven-drying. Sediment weight was determined to the nearest milligram for masses less than 400 g, and to the nearest 0.1 g for greater amounts.

13.3.3. Sediment characteristics and flume experiments


Use of the flow-driven erosion model component presented in Section 13.2.2 requires knowledge of the wet density () and the effective settling velocity or depositability () of the original sediment. The density of wet sediment aggregates and fine-ground particles was measured in the laboratory using a 25 ml-volume pycnometer (Klute, 1986). The aggregates were slowly wetted on a cloth and sieved wet to exclude particles finer than 0.15 mm before being introduced into the pycnometer. Measurements of depositability () were made for two different materials: sediment collected in the toe drain of terraces E and F also used in later flume experiments, and material from the upper 10 cm of soil collected on the beds of these same terraces. Two
267

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES


approaches were followed to determine settling velocity: using a modified bottom withdrawal tube (Lovell and Rose, 1988a, 1988b) and directly for single aggregates of different size classes. This dual approach was taken because the samples contained many coarse, fast-settling aggregates. This created problems when using the modified bottom withdrawal tube, as suspension samples could not be withdrawn fast enough to prevent clogging of the tube exit. The samples were therefore sieved to exclude aggregates coarser than 0.84 mm. The finer fraction was introduced into the modified bottom withdrawal tube and measurements were interpreted using GUDPRO software (Lisle et al., 1995). Material coarser than 0.84 mm was wet-sieved into seven size classes. For each sample and size class ten to twenty aggregates were introduced, one at the time, at the top of the tube. The time that it took each aggregate to travel 80 cm was measured and from this, the average settling velocity was calculated for each aggregate size class. Combining the results from the two experiments, a relationship was established between aggregate size, wet density and settling velocity, and this relationship was subsequently used to estimate the weight-averaged settling velocity from the aggregate size distribution of the original samples using GUDPRO software. A series of flume experiments was conducted to find values for Mannings roughness coefficient (n) and effective stream power fraction (F; cf. Eq. [13.10]). In many aspects the experiments were similar to those described by Proffitt (1988). However, the dimensions of the flume were adapted to better represent the dimensions of the terrace toe drain: a 7.0 m long and 0.3 m wide flume was positioned at a 2- (3.5%) slope. No rainfall was applied and therefore all sediment transported was originally entrained by the flow. An adjustable pump delivered water to a reservoir at the upper end of the flume and the water passed through stacked 5 mm diameter tubes to distribute the flow more evenly (cf. Proffitt, 1988). Water was applied at five different rates considered to represent conditions likely to occur under field conditions: 0.030(0.002), 0.076(0.003), 0.17(0.02), 0.35(0.01) and 0.66(0.04) m3 h-1. Runoff from the flume was collected in a drum equipped with a pressure transducer-logger system recording water level each minute. The system was calibrated to allow calculation of average runoff rate per unit flume width (q in m2 s-1) for each 1-minute interval. The experiments were conducted for 40 to 60 minutes, depending on discharge rate. During this period, every 5 minutes a 0.15 mm mesh sieve was held in the flow for one minute, while simultaneously a time-integrated 0.6-litre sample of runoff passing through the sieve was taken. The sediment and runoff samples were processed in the manner described for field samples (Section 13.3.2) after which the coarse sediment was sieved using six sieves of 0.15 to 8 mm mesh size. During the various experiments flow velocity (V in m s-1) was also measured a few times by adding a small volume of mixed dye and NaCl(aq) electrolyte solution to surface runoff at 4 m from the lower end of the flume. The electrical conductivity (EC in S cm-1) of the water exiting the flume was measured continuously, and the time lag between introducing the solution and the occurrence of the highest EC was recorded. Using the corresponding discharge rates, Mannings n was calculated from the relationship:

n=

S 0.5 q 0.667 V 1.67

[13.16]

268

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES


Eq. [13.16] implicitly assumes fully turbulent flow and its use may have introduced some errors. The formation, width and tortuosity of proto-channels or rills were observed in some experiments but not routinely measured; the implications of this are discussed in Section 13.6.2. Using the measured sediment concentrations and flow velocities, the value of F was calculated as (Rose et al., 1997a, cf. Eq. [13.10]):
1 i ci Fi = e S Vi

[13.17]

Both and e were assumed to equal 1.00)103 kg m-3, which will not introduce a significant error as long as sediment concentrations remain below 50-100 kg m-3 (Rose et al., 1997a). The parameters H and (cf. Eq. [13.10]) do not occur in Eq. [13.17] because sediment collected from a terrace toe drain was used in the flume experiments. This sediment was already detached and transported at least once and showed little cohesion. According to entrainment theory it should not offer any resistance to re-entrainment and, by consequence, both H and were assumed to equal unity (cf. Rose, 1993).
13.4. Results

13.4.1. Sediment yields


Amounts of sediment in suspension at the time of sampling, settled sediment (bed load) and total sediment yield from the various terrace riser and bed erosion plots used over the three measurement seasons are listed in Tables 13.1 and 13.2 for bed and riser plots, respectively. The resulting erosion rates are not readily compared because the three periods varied in length, timing and rainfall depth. Nevertheless some general observations can be made. Bare soil plots generally produced the highest amounts of sediment, at 71 t ha-1 in the full year 1999/00 (2871 mm rainfall) and 88-91 t ha-1 for the 2000/01 season (5 months; 1754 mm), respectively. Producing 62-107% of bare plot soil loss, clean-weeded cassava did least to reduce erosion (44-53 t ha-1 in 1999/2000, 70-95 t ha-1 in 2000/01), while ginger planted on ridges still produced sediment amounts that were 42-49% of those from bare plots (30-34 t ha-1; Table 13.1). Terrace beds with unweeded cassava crop in 1999/00 experienced 26-63% of bare soil erosion rates (23-56 t ha-1) while fallow plots invaded by weeds showed 9-21% of these (9-18 t ha-1). The mulched and nonmulched plots with inter-cropped cassava, maize and rice or peanut had 13-34% of the soil loss from bare plots (9-24 t ha-1 in 1998/99). Erosion rates from bare risers were overall considerably higher than from bare beds, at 31-89 t h-1 in 1998/99, 142-289 t ha-1 in 1999/2000 and 157-300 t ha-1 in 2000/01. Grass or shrub cover on the riser proved efficient in decreasing erosion: it reduced sediment transport to the terrace toe drain to 25-79% of that from bare risers in 1999/2000 (4-8 kg m-1) and to 7-44% (1-6 kg m-1) in 2000/01. Notable exceptions to this pattern were observed in cases when shrub cuttings were planted during, or just before the measurement season (R-I-9 and R-II-5 in Table 13.2). Because the riser plots generally had much less width than the bed plots (measured along the slope), amounts of sediment transport per unit length of adjoining toe drain

269

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES

Treatment a

Plot code

Area W m2 m

S %

rc %

css g l-1
b

ctot frac. Mtot Ttot susp. g l-1 % t ha-1 kg m-1 16 7.1 29 15.3 29 4.2 20 0.3 3.3 2.5 0.8 0.0

15 Dec 1998 (4 Jan 1999) - 11 May 1999: P=1775 mm (1418 mm) Mixed crops (peanut) B-I-1 6.1 4.7 7 4.8 Mixed crops (rice) B-I-8 1.6 1.7 17 23.1 Mixed crops B-I-5 1.7 1.8 19 4.2 Mixed crops (peanut) b B-I-12 4.3 0.9 2 0.1 1 Nov 1999 - 31 Oct 2000: P=2871 mm Bare soil B-II-3 Bare soil B-II-6 Ginger in ridges B-II-4 Ginger in ridges B-II-5 Cassava (1st season, weeded) B-II-9 Cassava (1st season, weeded) B-II-10 Mixed crops (rice) B-II-1 Mixed crops (rice), mulch B-II-2 Mixed crops (peanut) B-II-8 Mixed crops (peanut), mulch B-II-7 1 Nov 2000 - 6 Apr 2001: P=1754 mm Bare soil B-III-4 Bare soil B-III-5 Cassava (2nd year) c B-III-7 Cassava (2nd year) c B-III-8 Cassava (2nd year, weeded) c B-III-1 Cassava (2nd year, weeded) c B-III-2 Cassava (2nd year, weeded) c B-III-9 Cassava (2nd year, weeded) c B-III-10 Fallow B-III-3 Fallow B-III-6
a b

1.3 8.3 1.1 3.7 1.6 5.6 5.1 24.9

3.8 3.5 3.8 3.5 3.8 3.6 3.2 3.5 3.6 3.5

2.5 2.3 2.5 2.3 2.5 2.4 2.1 2.3 2.4 2.3

16 15 16 21 11 14 11 17 8 9

18.9 16.4 10.4 10.6 12.6 16.5 13.6 9.2 5.7 9.2

2.9 3.3 2.9 2.5 2.1 2.5 2.1 1.8 2.1 3.5

13.0 15.0 10.0 11.1 12.1 11.1 6.2 4.7 5.7 7.3

22 22 29 23 17 23 34 37 36 48

70.7 70.6 29.8 34.3 44.0 52.5 24.2 12.8 9.2 19.3

17.7 16.2 7.5 7.9 11.0 12.6 5.1 2.9 2.2 4.4

3.8 3.5 3.5 3.6 3.2 3.5 3.8 3.6 3.8 3.5

2.5 2.3 2.3 2.4 2.1 2.3 2.5 2.4 2.5 2.3

16 21 9 8 11 17 11 14 16 15

26.1 24.0 26.7 15.3 34.2 28.4 28.9 30.6 18.7 9.5

3.8 3.9 2.0 1.9 2.8 4.0 2.8 3.1 2.1 2.0

19.8 21.0 11.9 8.7 13.4 19.0 13.7 17.5 5.5 5.1

19 19 17 22 21 21 21 18 37 39

90.7 88.3 55.5 23.3 80.4 94.5 69.5 94.1 18.2 8.5

22.7 20.3 12.8 5.6 16.9 21.7 17.4 22.6 4.5 1.9

third crop in addition to cassava and maize placed between brackets where appropriate. numbers between brackets refer to indicated plots, which were measured for a shorter period. c left fallow after harvest.

Table 13.1. Characteristics of terrace bed plots erosion plots and summary of measurements during three seasons. Listed for each plot are area, width (W, measured along the gradient), slope gradient (S), runoff coefficient (rc), concentrations of suspended (css) and total sediment (ctot), percentage of sediment in suspension at the time of sampling (frac. susp.), overall sediment yield (Mtot) and sediment transport (per unit plot length, measured along the toe drain). Details about treatment such as vegetation and mulch cover can be found in Table 12.2.

were comparable or even somewhat lower, at 10-16 kg m-1 versus 16-18 kg m-1 in 1999/2000 and 13-17 kg m-1 versus 20-23 kg m-1 in 2000/01, in all cases for bare soil (Tables 13.1 and 13.2). Sediment yields are also listed as overall average sediment concentration (Tables 13.1 and 13.2). The terrace bed plots showed average concentrations of 3-25 g l-1, of which 1-5 g l-1 (16-48 %) was in suspension. There was a clear relationship with plot treatment: sediment concentrations were highest in runoff
270

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES

Treatment a

m2 m % % g l-1 g l-1 % t ha-1 kg m-1 15 Dec 1998 (4 Jan 1999) - 11 May 1999: P=1775 mm (1418 mm) b No cover R-I-7 2.1 1.2 62 12 2.9 41.9 7 89.2 11.1 No cover R-I-4 2.2 0.9 61 20 1.9 14.9 13 53.0 4.6 No cover b R-I-13 1.6 1.5 61 20 1.1 25.5 4 70.9 10.6 No cover R-I-3 2.0 0.9 68 8 2.0 22.9 9 30.6 2.9 Shrub planted (44%) c R-I-9 2.5 1.3 73 13 2.9 45.8 6 109.4 13.8 Grass (10 %) b R-I-11 2.0 1.2 70 4 3.7 93.2 4 52.1 6.0 b Grass (96 %) R-I-10 1.0 0.8 59 1 1.9 7.9 24 0.7 0.1 1 Nov 1999 - 31 Oct 2000: P=2871 mm No cover R-II-2 2.0 No cover R-II-3 1.9 No cover R-II-6 3.8 Grass planted (54 %) R-II-1 1.3 Shrub planted (15 %) R-II-5 3.8 Grass (15 %) R-II-8 1.4 Shrub (22 %) R-II-7 2.4 Shrub (85 %) R-II-4 1.5 1 Nov 2000 - 6 Apr 2001: P=1754 mm No cover R-III-2 No cover R-III-3 No cover R-III-6 Grass (ca. 54 %) R-III-1 Grass (ca. 15%) R-III-8 Shrub R-III-5 Shrub (ca. 22 %) R-III-7 Shrub (ca. 85 %) R-III-4
a b

Plot code

Area

rc

css

ctot

frac susp

Mtot

Ttot

0.6 0.5 1.1 0.3 1.0 0.8 1.2 0.5 0.6 0.5 1.1 0.3 0.8 1.0 1.2 0.5

57 52 55 58 55 74 68 83 57 52 55 58 74 55 68 83

23 25 12 11 10 5 10 16 21 28 13 8 11 7 5 12

2.4 2.3 1.8 0.8 2.9 3.9 3.3 0.6 2.2 2.2 3.2 0.8 2.6 1.1 1.8 0.5

44.6 5 28.5 8 41.9 4 38.2 2 41.4 7 54.8 7 23.0 14 18.4 3 62.4 62.1 69.2 43.6 36.6 35.7 25.6 10.8

289.0 200.4 142.2 118.1 118.7 86.2 66.2 82.9

16.4 9.9 15.3 4.1 12.2 6.9 7.8 3.7 13.3 14.9 16.9 2.0 5.8 4.2 2.4 1.1

2.0 1.9 3.8 1.3 1.4 3.8 2.4 1.5

3 234.0 4 300.2 5 157.3 2 59.3 7 72.3 3 41.4 7 20.5 5 23.6

percentage cover given between brackets. numbers between brackets relate to plots that were measured for shorter periods. c cuttings planted during season (9 February 1999).

Table 13.2. Characteristics of terrace riser plots erosion plots and summary of measurements during three seasons. Listed for each plot are area, width (W, measured along the gradient), slope gradient (S), runoff coefficient (rc), concentrations of suspended (css) and total sediment (ctot), percentage of sediment in suspension at the time of sampling (frac. susp.), overall sediment yield (Mtot) and sediment transport (per unit plot length, measured along the toe drain).

from bare plots at 13-21 g l-1 (Table 13.1). Clean-cultivated cassava and ginger resulted in sediment concentrations that were not much lower than those for bare soil (10-19 g l-1), while low concentrations were found for plots that were invaded by weeds or planted with mixed crops (4-12 g l-1). An exception to the general pattern was terrace bed plot BI-12, which showed a high sediment concentration of 25 g l-1, but this was associated with a very low runoff coefficient of 0.1% (Table 13.1). At 15-69 g l-1 and 13-21 g l-1, respectively, sediment concentrations for bare or unweeded risers were higher than those for bare bed plots. This was caused mainly by a higher concentration of coarser sediment;

271

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES

Code Treatment: riser / bed 2 Dec 1998 -11 Apr 1999 (130 days): P= 1697 mm A Grass (30 %) / cassava-maize-peanut a B Grass (98 %) / cassava-maize-peanut a C Grass (25 %) / cassava-maize-rice a D Grass (75 %) / cassava-maize-rice a E No cover / cassava-maize a F No cover / cassava-maize a

rc % 7.0 7.1 16.7 26.2 6.5 7.8

css g l-1 1.8 2.2 2.2 1.4 1.0 1.3 9.1 3.1 7.9 3.4 3.3 3.8 4.6 5.6 4.1 3.5

ctot g l-1 11.2 5.9 16.0 5.0 1.1 1.4 38.6 3.8 37.2 3.8 6.3 3.8 29.0 35.2 14.1 11.9

frac. susp. % 16 37 14 29 88 95 24 81 21 87 52 100 16 16 29 30

Mtot t ha-1 13 7 43 22 1 2 55 6 148 13 12 9 90 112 29 7

25 Oct 1999 - 24 Oct 2000 (365 days): P = 2567 mm A No cover / cassava-maize-rice 5.5 5.9 B Grass (98 %) / cassava-maize-rice b C No cover / cassava-maize-rice 15.1 12.6 D Grass (75 %) / cassava-maize-rice b E No cover / cassava-maize-rice 7.3 9.1 F Shrub planted / cassava-maize-rice b 25 Oct 1999 - 6 Feb 2001 (104 days): P = 1367 mm A No cover / cassava-maize-rice C No cover / cassava-maize-rice E No cover / cassava-maize-rice F Shrub / cassava-maize-rice
a b

22.8 24.7 16.1 3.9

harvest residue left on bed and in toe drain. silt pit and mulching at end of toe drain.

Table 13.3. Treatment of bench terrace units and summary of measurements during three seasons. Listed for each plot are runoff coefficient (rc), concentrations of suspended (css) and total sediment (ctot), percentage of sediment in suspension at the time of sampling (frac. susp.) and overall sediment yield (Mtot).

concentrations of material still in suspension at the time of sampling were comparable at 1-3 g l-1 (4-13% of total sediment concentration) and 3-4 g l-1 (19-22%), respectively. Terrace unit sediment yields were measured during three consecutive rainy seasons, but only during the second season (1999/00) were data collected for a full year. Measurements made on terraces B and D in the 2000/01 season suffered technical problems and are therefore not included. The results are listed in Table 13.3. Again, values are somewhat difficult to compare. However, the effect of riser cover on sediment concentration is readily observed: sediment concentration increased significantly after the partial grass cover on the risers of terraces A and C was removed before the 1999/00 season, from 11 to 39 g l-1 on terrace A and from 16 to 37 g l-1 on terrace C (Table 13.3). Also noteworthy are the initially low sediment concentrations measured on terraces E and F steadily increasing over the three seasons; from 1 g l-1 in 1998/99 via 6 g l-1 in 1999/2000 to 14 g l-1 in 2000/01 on terrace E, and from 1 g l-1 via 4 g l-1 to 12 g l-1 for the respective periods on terrace F. The toe drains of these two terraces had a gradient that was less than that of the other terraces (2% versus 4% for terraces A and B and 5-6% for terraces C and D in 1999/00) and in fact were even horizontal or slightly backsloping towards the end. This is probably the main reason for the low overall sediment concentrations and the high percentage of suspended material in the total load (Table
272

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES


13.3). The gradual increase of sediment concentrations prior to the 1998/99 wet season may be explained by the fact that the two terraces were reconstructed on a field that had been left fallow for two years.

13.4.2. Sediment characteristics and flume experiments


The wet density () of aggregates from the original sediment used in the flume experiments was 1.64-1.77)103 kg m-3 (N=5) with an average value of 1.68)103 kg m-3. The density of fine-ground sediment was 2.53)103 kg m-3 (2.50-2.57)103 kg m-3, N=4). Results of the modified bottom withdrawal tube and direct settling velocity measurements are listed in Table 13.4 and shown in Figs. 13.1a and b. For the fraction smaller than 0.84 mm an average settling velocity (vi) of 0.0472 m s-1 and 0.0446 m s-1 was found, for toe drain sediment and terrace bed topsoil, respectively. Using the GUDPRO programme the settling velocity for the fraction of material smaller than 0.15 mm was determined to be 3.76)10-3 m s-1 and 3.86)10-3 m s-1 for the respective samples. There was no significant difference between settling velocities for aggregates from toe drain sediment and from the soil; the measured values are shown together with values predicted by GUDPRO using the measured value of =1.68)103 kg m-3 in Fig. 13.1a. In order to match the predicted and average measured settling velocities for the individual aggregate size classes a wet density between 1.56)103 and 1.90)103 kg m-3 needed to be assumed, with the value tending to decrease with increasing aggregate size. This may be explained by an increase in porosity with aggregate size. Table 13.4 lists the average settling velocity for each aggregate size class; these values were used to calculate the depositability () of samples with known wet-sieved aggregate size distributions. For sediment in the terrace toe drains this yielded a depositability of 0.106 m s-1 (st. dev. 3.7 %, N=3). The measurements of discharge, flow velocity and sediment concentration made during the flume experiments are listed in Table 13.5, together with derived values of flow depth, Mannings roughness coefficient, stream power, depositability (determined from the aggregate size distribution for individual samples) and the calculated F values. The relationship between discharge and flow velocity is shown in Fig. 13.2. The roughness coefficients (n) determined for individual experiments varied between 0.012 and 0.123 s m-1/3 with a tendency to decrease with increasing discharge (Table 13.5). The value that best fitted the combined measurements was 0.0159 s m-1/3 (r2=0.86, see Fig. 13.3). Sediment concentrations measured during individual experiments showed considerable variation; concentrations were usually highest at the beginning of the experiment and decreased towards the end, except for experiments IIIa and IIIb, during which the concentrations again slightly rose after in initial decrease. This increase coincided with the formation of shallow rills. The highest concentration (117 kg m-3) was measured at the start of experiment Va, but had already dropped to 33 kg m-3 five minutes later (Table 13.5). the lowest concentration (0.203 kg m-3) was measured after 50 minutes in experiment Ia. The depositability of individual sediment samples was calculated (assuming a value of 0.0038 m s-1 for the fraction smaller than 0.15 mm; cf. Table 13.4) and tended to increase with stream power: no sediment coarser than 0.15 mm was transported below =0.023 W m-2, but depositability increased via 0.004-0.035 m s-1 at =0.050-0.053 W m-2 to 0.056-0.072 m s-1 at =0.198-0.205 W m-2 (Table 13.5). The single highest depositability (0.097 m s-1, measured at the start of experiment VI) was still less than for the original material.

273

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES

0.40

0.30

v i (m s -1 )

0.20

0.10

0.00 0.1 1 10 100

Aggregate size class mean (mm)

1.0 0.8
Fraction slower

0.6 0.4 0.2 0.0 0.0001


sediment soil

0.001

0.01
v i (m s )
-1

0.1

Fig. 13.1. Settling velocity of soil and sediment aggregates (a) coarser than 0.84 mm diameter (bars show range) and (b) finer than 0.84 mm diameter.

Size class mm >4.0 2.0-4.0 1.19-2.0 0.59-1.19 0.25-0.59 0.15-0.25 <0.15

vi m s-1 2.17)10-1 1.61)10-1 1.24)10-1 9.15)10-2 5.03)10-2 2.12)10-2 3.81)10-3

Weight fraction % 7.3 1.8 19.4  1.3 22.1  4.5 23.5  1.5 18.1  3.9 7.1  2.2 2.6  1.1

Table 13.4. Settling velocity and wet-sieved fraction of aggregate size classes for sediment used in flume experiments. The corresponding bulk settling velocity is =0.106 m s-1 (3.7%).

274

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES

0.30 0.25 0.20

V (m s )

-1

0.15 0.10 0.05 0.00 0 1 2 3


4

4
2 -1

q (x10 m s )

Fig. 13.2. Results of discharge (q) and flow velocity (V) measurements in the flume experiments. Best fitting value of Mannings n was 0.0159 s m-1/3 (r2=0.86).

Parameter a Hydraulic characteristics Discharge (q x10-4 m2 s-1) Flow velocity (V, m s-1) Flow depth (D, mm) Mannings n (s m-1/3) Stream power (, W m-2) Sediment characteristics Number sed. of samples Average c (kg m-3) C. V. (%) Last four samples Average conc. C. V. (%) Depositability ( , m s-1) Average C. V. (%) 0.29 0.029 1.0 0.063 0.010

I b 0.26 0.019 1.4 0.123 0.009 a 0.72 0.103 0.7 0.014 0.025

II b a

III b a

IV b 3.35 0.209 1.6 0.012 0.115 a

V b 5.96 0.257 2.3 0.013 0.205

0.67 1.44 1.54 3.15 0.075 0.161 0.111 0.126 0.9 0.9 1.4 2.5 0.023 0.011 0.021 0.027 0.023 0.050 0.053 0.108

5.77 0.225 2.6 0.016 0.198

9 0.6 119% 0.2 5%

7 0.7 95% 0.3 23%

7 11.3 127% 4.6 24%

8 2.6 44% 2.0 11%

7 9 8 9 19.6 2.1 14.8 15.0 28% 47% 45% 87% 21.2a 1.6 19.1a 9.1 23% 15% 19% 26%

7 8 57.5 19.0 48% 108% 51.4 6.4 15% 29%

<0.004 <0.004 <0.004 <0.004 0.035 0.004 0.063 0.053 0.072 0.056 89% 44% 55% 53% 43% 49%

Effective stream power fraction (F) Average 0.015 0.043 0.137 0.042 0.149 0.023 0.142 0.087 0.309 0.089 C. V. (%) 119% 95% 127% 44% 28% 47% 45% 87% 48% 108% 6.1 9.3 9.0 12.3 12.6 16.8 17.2 21.4 21.6 Modelled conc. (kg m-3)b 6.5 a concentrations did not decrease towards the end of the experiment. b using parameters given in text.

The variation in determined F values was considerable, yielding values between 0.0130.016 and 0.2670.128 (Table 13.5). Despite the variation during individual experiments, there appeared to be a tendency for F to increase with discharge.

275

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES

70 60 50

c (kg m -3 )

40 30 20 10 0 0.00

0.05

0.10

0.15

0.20

0.25

5  (W m -2 )
Fig. 13.3. Comparison of sediment concentrations (c) measured in the flume experiments at different values of stream power (5) and values predicted by Eqs. [13.9] and [13.10] on the basis of discharge measurements and using values of =1680 kg m-3; =0.104 m s-1; n=0.0159 s m-1/3 and an optimised value of F=0.115 (see text for explanation).

The average concentrations measured at different stream powers are shown in Fig. 13.3. Concentrations predicted by Eqs. [13.9] and [13.10] agreed best, if still rather poorly, to observed values when a value of F=0.115 was used (r2=0.41).
13.5. Modelling sediment yield from terrace riser and bed plots and terrace units

13.5.1. Modelling procedure Terrace riser and bed plots


Measurements of storm runoff depth and sediment yield from the terrace riser and bed plots were used to calibrate and test the components of the TEST model. This could only be done for the periods for which rainfall intensity data were also available, leaving the periods of 17 December 1998-11 April 1999 (116 days, P=1532 mm) and 21 October 1999-8 February 2000 (110 days, P=1390 mm). Most of the model parameters either could be measured directly or were derived elsewhere: parameters that determine the relationship between slope gradient and apparent downslope splash length (%, & and p in Eq. [13.2]) were based on field studies (Chapter 9) and soil tray experiments (Chapter 10). Width (W) and gradient (S) of the plots was measured at the beginning of each season in field surveys. Storm erosivity (E20) was determined for all storms using Eq. [13.6] with values for depth-averaged rainfall intensity ( ) and rainfall depth (P) calculated from tipping-bucket measurements (cf. Chapter 12). Storm runoff depth (Qtot) was calculated using the runoff model presented in Chapter 12 with appropriate Im and SI values.

276

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES


The remaining unknown parameters are soil detachability (D20), the washed fraction (j) and the coefficients relating soil loss to vegetation (v) and soil contact cover (s). One data set collected on a relevant terrace riser or bed plot was used to calibrate each parameter. Parameters were derived separately for terrace beds and risers and because of the apparent differences between clean-weeded cassava or ginger cultures and mixed crops (cf. Section 13.4.1) separate values of v were also determined for these two categories. The values for D20 and j were optimised through non-linear LevenbergMarquardt optimisation by least squares, using data sets collected from plots with bare soil (B-II-3 and R-II-2), mixed crops (B-II-2), clean-weeded cassava crop (B-II-9), a plot that received 40% mulch cover at the start of the season (B-II-2), one on which the maize and rice litter was left after harvest (B-I-8) and a riser plot with partial grass cover (B-RII-8). Model performance was expressed by the model efficiency (ME; Nash and Sutcliffe, 1970; cf. Chapter 12)

Terrace unit plots (NBEPs)


After calibrating the terrace bed and riser model components, event sediment yield measurements from six terrace unit plots were used to evaluate the performance of the TEST model. Again, simultaneous rainfall intensity and sediment yield data were available for two periods, i.e. 3 December 1998-11 April 1999 (130 days, P=1684 mm) and 26 October 1999-8 February 2000 (106 days, P=1335 mm). The dimensions, slope gradients and cover characteristics of the bed, riser and drain sections were established in field surveys (Table 12.3, Chapter 12). Estimates for initial amounts of sediment in the toe drain (m) were based on detailed field surveys of sediment storage in the toe drains of the respective terraces made on 10-12 November 1998, 17-26 January 1999 and 8-9 April 1999 (Wiegman, 1999). Because of the lack of measurements for the 1999/2000 season, measurements at the end of the 1998/99 period were used as initial values for this period. Three additional parameters needed to be derived by calibration using the existing data sets, viz.: (i) the coefficient H, relating H to m (cf. Eq. [13.15]); (ii) the coefficient s predicting the reduction in soil loss because of harvest litter left in the toe drain (1998/99 season); and (iii) the reduction in soil loss caused by the conservation measures (silt pits and mulching in the toe drains) on terraces B, D and F during the 1999/00 season. The last was expressed in an equivalent soil cover (CS*), i.e. the fraction of mulch cover that would result in an equal reduction of sediment yield. s and CS* were determined using one calibration data set each by non-linear optimisation. The coefficient H appeared to be a particularly sensitive parameter, however, and therefore it was decided to use the minimum total residual variance from all twelve data sets as a criterion instead of using one data set for calibration. It was accepted that this precluded fully independent calibration and testing of the model and the implications of this are discussed in Section 13.6.4. Again, Nash-Sutcliffe model efficiency (ME) was used as an indicator of model performance.

277

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES

Parameter Fixed W (m) Cv (%) Cs (%) % (m) & (-) p (-) Calibrated D20 (g J-1) j (%) v (-) s (-)

bed plots 0.9-4.7 0-96 0-51 0.010 9.35 0.44 11.7 3.98 1.36 or 0.0 a 0.38 or 2.45 b

Value

riser plots 0.3-1.5 0-96 n.a. 0.010 9.35 0.44 15.9 2.15 2.02 n.a.

referring to a cover of mixed crops and cassava, respectively. referring to mulch applied before and after harvest, respectively.

Table 13.6. Values of parameters used in the TEST model. Fixed parameters were derived elsewhere (Chapters 9 and 10) or relate to plot dimensions; calibrated parameters were determined using one data set each (n.a.=not applicable).

13.5.2. Terrace risers and beds


All parameters used to simulate soil loss for the terrace beds and risers are summarised in Table 13.6. Measured and modelled sediment yield are listed in Table 13.7 along with measured runoff coefficient (rc), modelled percentage of washed sediment and observed percentages of suspended sediment. The absolute and relative differences between cumulative observed and modelled soil losses and model efficiency (ME) are also given. For the bare bed and riser calibration data sets, respective D20 values of 11.7 and 15.9 g J-1 were found, while for j values of 3.98 and 2.15% resulted (Table 13.6). The associated model efficiencies for these data sets were 0.53 and 0.83, respectively (Table 13.7). For mixed crops planted on the terrace bed a cover coefficient (v) of 1.36 (ME=0.54) was found (Table 13.6), resulting in a relationship between relative erosion and cover fraction that is reasonably close to proportional (e.g. MF=0.51 for 50% cover, cf. Eq. [13.9]). Conversely, the fact that clean-weeded cassava and ginger cultures did nothing to reduce erosion is reflected by v=0 (ME=0.49). The lower grass and shrub vegetation on some of the risers appeared more effective in reducing erosion (v=2.02; ME=0.39) and resulted in a mulch factor MF=0.36 for 50% cover. Even more effective was the application of mulch: the model suggests that 50% mulch cover after harvesting reduced erosion rates from the terrace bed to 29% of non-mulched soil (s=2.45, ME=0.40). When applied at the beginning of the cropping season mulching appeared to be considerably less effective, reducing erosion over the entire period to an average 82% only (s=0.38, ME=0.54, cf. Tables 13.6 and 13.7).

278

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES

Treatment

Plot code

rc Mobs Mmod Difference % t ha-1 t ha-1 t ha-1 %

ME Obs. Mod frac. frac. susp. wash % %

Mod wash conc g l-1

Terrace bed section plots Calibration plots Bare soil Cassava (weeded) Mixed crops Mixed crops, mulch start mixed crops, mulch later Bare soil Cassava (weeded) Ginger on ridges Ginger on ridges Mixed crops, mulch later Mixed crops, mulch later Mixed crops, mulch later Mixed crops Mixed crops, mulch start Average Standard deviation Terrace riser section plots Calibration plots No cover Grass cover No cover No cover No cover No cover No cover No cover Shrub planted Grass Grass Grass Shrub Shrub Shrub Average Standard deviation
a

II-3 II-9 II-1 II-2 I-8 II-6 II-10 II-4 II-5 I-1 I-5 I-12 II-8 II-7

19 13 14 9 24 16 16 10 11 5 4 0.1 6 9 11 6

26.4 31.9 11.4 7.1 14.1 25.8 32.5 21.1 12.9 6.8 3.7 0.3 4.5 5.2 14.5 11.0

25.7 23.3 10.6 8.1 14.1 24.8 27.3 21.9 21.9 3.7 7.7 4.1 4.9 4.9 14.5 9.2

-0.7 -8.6 -0.8 1.0 0.0 -1.0 -5.1 0.8 9.1 -3.1 4.0 3.8 0.3 -0.4 2.8

-3 -27 -7 13 0 -4 -16 4 70 -45 108

0.53 0.49 0.54 0.54 0.40 0.59 0.63 0.58 0.07 0.22 <0 a <0 7 0.21 -7 0.31 19 0.38

22 17 34 37 29 22 23 29 23 16 29 20 36 48 28 9

62 57 47 39 57 59 59 51 42 65 13 1 32 36 39 19

8.3 7.8 1.3 2.0 2.2 8.4 7.7 7.3 7.5 3.2 1.9 2.9 2.6 2.2 4.7 2.9

II-2 II-8 I-3 I-4 I-7 I-13 II-6 II-3 I-9 II-1 I-10 I-11 II-7 II-4 II-5

23 5 8 22 13 21 12 25 15 11 1 4 10 16 10 13 7

152.9 66.2 27.7 47.4 88.3 65.3 86.6 127.5 107.2 99.0 0.6 50.0 46.4 49.8 66.4 72.1 39.2

134.2 72.7 55.3 63.4 61.1 50.3 70.9 147.2 53.8 68.8 15.2 61.8 43.4 32.8 53.5 65.6 34.0

-18.7 6.5 27.6 16.0 -27.2 -15.0 -15.7 19.6 -53.4 -30.2 14.5 11.9 -3.0 -17.0 -12.9 19.3

-12 10 100 34 -31 -23 -18 15 -50 -30


a

24 -6 -34 -19 27

0.83 3.6 0.39 3.9 0.12 8.7 0.17 12.8 0.14 6.9 0.31 4.5 0.66 3.6 0.52 5.6 0.03 6.3 0.04 1.3 <0 24.1 0.12 3.9 0.30 10.4 0.17 2.4 0.45 5.5 0.30 6.9 5.7

11 4 7 15 14 22 12 10 15 3 0.6 4 9 4 10 9 6

5.0 4.4 3.2 2.9 4.4 4.7 5.1 5.0 3.5 1.7 1.1 4.8 3.8 0.9 3.8 3.6 1.4

difference more than order of magnitude.

Table 13.7. Results of TEST model application to measurements of sediment yield from terrace bed and riser plots during two seasons. Listed are observed runoff coefficient (rc; cf. Chapter 12), cumulative sediment yield observed (Mobs) and modelled (Mmod) for the simulation period, absolute and relative difference between these, associated model efficiency (ME), fraction of sediment still in suspension at the time of sampling, and modelled fraction and concentration of washed (fine) sediment. Note that sediment yields may vary from those listed in Tables 13.1 and 13.2 depending on the period for which rainfall intensity measurements were available (see text for further details).

279

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES

1000

Modelled soil loss (t ha )

-1

100 R-I-10 10

B-I-12 1

0.1 0.1 1 10 100


-1

1000

Measured soil loss (t ha )

Fig. 13.4. Correlation between observed and modelled soil losses from terrace bed (triangles) and riser (dots) section plots (overall ME=0.83). Codes of outliers refer to the corresponding plots (see text for explanation).

In most cases application of the calibrated model yielded estimates for total soil loss that agreed reasonably with observed amounts, although model efficiencies were sometimes rather low. Modelled terrace bed sediment yields were within 19% of observed values on average, whereas the difference for most individual plots ranged between -27% and +70%. An even greater over-estimation resulted for two erosion plots from which sediment yield was observed to be particularly low (Table 13.7, Fig. 13.4). Associated model efficiencies varied from 0.07 to 0.63 (ME=0.36 on average), with values well below zero for the latter two plots. The differences between cumulative modelled and observed sediment yields from terrace riser plots were somewhat larger (27% on average), and ranged between -50% and +100% for individual plots. Again, one particularly well-protected plot (R-I-10) had a very low sediment yield that was severely over-estimated by the model. Model efficiencies ranged between 0.03 and 0.83 (lower than zero for the mentioned plot), with an average of 0.30 (Table 13.7, Fig. 13.4). Fig. 13.5 compares cumulative modelled and observed sediment yields for the bed and riser plots used for calibration, plus those showing the best, worst and closest-to-average model efficiencies. Despite the sometimes poor simulation of individual storms, as expressed by the generally low model efficiencies, agreement between cumulative values is still reasonable when compared to contrasts between plots, as is illustrated in Fig. 13.4. If the agreement between observed and simulated overall sediment yields for individual plots is calculated conform to the formulation of model efficiency, a high value of ME=0.83 results.

280

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES


40

200

(a)
30

(d)
Cum. modelled SY (t ha ) 150

Cum. modelled SY (t ha )

-1

-1

20

100

10

B-II-3* B-II-9* B-II-10

50

R-II-2* R-II-8*
0

0 0 5 10 15 20 25 30 35

50

100 Cum. observed SY (t ha )


-1

150

200

Cum. observed SY (t ha -1 )

15

100

(b)
Cum. modelled SY (t ha )
75

(e)

10 time of harvest

Cum. modelled SY (t ha )

-1

-1

50

25

B-II-1* B-I-8*
0 0 5 10 -1 Cum. observed SY (t ha ) 15
0 0 25 50 Cum. observed SY (t ha )
125
-1

R-II-6 R-I-13
75 100

10

(c)
8 Cum. modelled SY (t ha )
Cum. modelled SY (t ha )
-1
-1

(f)
100
time of planting

75

50

25

B-II-2* B-II-12
0 0 2 4 6 -1 Cum. observed SY (t ha ) 8 10
0 0 25 50 75
-1

R-II7 R-I-9

100

125

Cum. observed SY (t ha )

Fig. 13.5. Comparison of modelled and observed cumulative sediment yield for selected data sets collected on (a-c) terrace bed section plots and (d-f) terrace riser section plots. Asterisks (*) indicate data sets used for calibration. Additional data shown pertain to bed section datasets for which the model performed best (B-II-10 in (a), ME=0.63,), closets-to-average (B-I-8 in (b), ME=0.40); and worst (B-I-12 in (c), ME<0). For riser section data, the best (R-II-6 in (e), ME=0.66), closest-to-average (R-I-13 in (e), ME =0.31, and R-II-7 in (f), ME=0.30) and worst (R-I-9 in (f), ME =0.03) model performance are also illustrated.

281

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES 13.5.3. Terrace units


Results obtained by applying the TEST model to simulate event sediment yields from bench terrace units are summarised in Table 13.8, while cumulative modelled and observed sediment yields are shown for all data sets in Fig. 13.6a-f. Values of s=3.34 and CS*=0.50 were obtained from calibration data sets, while the best overall model performance was obtained with H=0.0072 m2 kg-1. With these parameter values the absolute difference between modelled and observed sediment yields ranged between -34 t ha-1 for terrace A in the 1999/2000 season and +10 t ha-1 for terrace C in the 1998/99 season (Table 13.8). The model explained 63% of the total variance contained within the twelve data sets. However, model efficiencies in excess of 0.60 were found for only five out of the twelve data sets (Table 13.8). In particular, sediment yields from terraces E and F in 1998/99 were predicted poorly, with model efficiencies less than zero and overestimations of 200-327% (Table 13.8, Fig. 13.6e-f). An important reason for this is probably related to the profile of the toe drains of terraces E and F. The model used the average gradient of the gutter over the last 6-metre (as measured at the beginning of the 1998/99 season), whereas in actual fact the toe drains of these terraces had a horizontal or slightly negative gradient over the last few metres. This will have reduced runoff sediment transport much, particularly of coarser material. Indeed very little bed load was found in the collecting basins for these two terraces (Table 13.8). Observations and predictions for terraces E and F agreed better during the 1999/00 period (Fig. 13.6e and f), corresponding with an increase of the bed load fraction over the two seasons. This may be related to a gradual filling of the toe drain, increasing its gradient. A similar phenomenon probably caused the initially poor model performance for terrace D in the 1998/99 season (Fig. 13.6d). The under-estimation of sediment yield from terrace A in 1999/00 may well be related to the fact that the partial grass cover on the risers of terrace A (and C) was removed during the 1999 dry season, causing considerable amounts of riser soil to fall into in the toe drain. The amount of sediment in the toe drain will therefore have been more than the 70 kg m-2 assumed on the basis of measurements at the end of the 1998/99 season (cf. Table 13.7). Increasing m to 300 kg m-2 increased model efficiency from 0.32 to 0.60 and decreased the difference between modelled and observed cumulative sediment yield from -64% to -33%. Model performance was also rather poor for terrace B in both seasons, corresponding with a number of break points in the cumulative curves (ME=0.29-0.40, Fig. 13.6b). No satisfactory explanation could be found for this phenomenon but it corresponds to the general finding that predictions were less good for bench terraces experiencing low soil losses overall. Similar observations were made with respect to modelling runoff from these plots (cf. Chapter 12). Table 13.8 also compares modelled amounts of sediment transported by runoff entrainment and wash to observed amounts of bed load and suspended load. Despite the fact that these cannot be exactly equated (see Section 13.6.3), there appears to be generally good agreement between the respective amounts. This agreement is also evident in Fig. 13.8, in which total yields of entrained/bed load material and washed/suspended sediment are compared, as well as overall sediment yields. Model efficiencies calculated on the basis of these cumulative amounts were 0.83, 0.86 and 0.54 for total, entrained/bed load and washed/suspended sediment yield, respectively (Fig. 13.8). The most notable exceptions are the already explained over-estimates of coarse sediment for terraces E and F.

282

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES

A Runoff m (mm h-1) a Runoff coefficient (%)

1998/99 C D

1999/2000 C D 36 15 51 13

E 76 10

F 81 11

124- 145- 29- 21- 138- 132186 183 228 304 56 27 299 82 7 7 17 26 7 8 7 7 1.3 5.6 4.3 327 <0 1.9 5.8 3.9 200 <0

Terrace sediment yield Observed (t ha-1) 13.4 7.0 43.3 22.0 11.2 8.4 53.1 27.4 Modelled (t ha-1) -2.2 1.4 9.8 5.4 Difference (t ha-1) Difference (%) -17 20 23 25 Model efficiency (ME) 0.52 0.29 0.62 0.57

53.1 5.2 116.4 8.4 10.0 7.4 19.0 8.6 89.6 8.5 8.9 5.9 -34.0 3.4 -26.8 0.1 -1.0 -1.5 -64 65 -23 1 -10 -20 0.32 0.40 0.68 0.67 0.69 0.62 76.2 3.7 1.3 0.3 13.4 4.8 7.6 5.5 6 2 10 3 1.91 0.36 0.21 0.26 95.1 21.3 59.9 46.1 52 14 21 13 1.6 6.9 5.3 5.4 29 4 51 16 4.6 5.3 7.6 6.8 61 7 22 10 0.0 7.4 5.5 4.4 43 6 34 17

Modelled transport components entrainment (t ha-1) 7.9 5.8 44.6 19.4 0.8 0.2 13.9 4.3 3.3 2.6 8.5 7.9 4.8 5.6 5.1 4.3 wash (t ha-1) 2 2 6 6 11 2 2 1 from drain (%) b Sed. delivery ratio (SDR) 0.88 0.91 1.57 1.11 0.13 0.16 1.01 0.69 Observed sediment components 'bed load' (t ha-1) 11.2 4.4 37.4 15.7 0.2 0.1 41.7 1.1 2.2 2.6 5.9 6.3 1.2 1.8 11.3 4.1 suspended (t ha-1) Sediment concentration observed ctot (g l-1) modelled ctot (g l-1) 11.2 5.9 16.0 5.0 1.1 1.4 64.3 6.2 9.3 7.1 19.6 6.2 4.9 4.2 23.1 10.2 60 5 14 20 33 2 39 26 56 13 21 11 35 10 34 20 74 5 16 5 69 8 16 7 58 5 16 22 19 1 46 34

Modelled sediment sources splashed from riser (%) washed from riser (%) splashed from bed (%) washed from bed (%)

Toe drain deposition layer characteristics m initial (kg m-2) 39 22 98 32 m final, model (kg m-2) 42 23 79 30 70 41 150 27 m final, obs. (kg m-2) H average modelled 0.24 0.14 0.45 0.19
a b

153 84 70 41 178 33 70 46 124 101 ? ? 0.69 0.13 0.39 0.27

150 27 124 101 109 37 145 114 ? ? ? ? 0.60 0.21 0.62 0.54

98/99 numbers refer to pre- and post-harvest conditions, respectively. expressed as percentage of washed sediment detached from the toe drain.

Table 13.8. Results of TEST model application to measurements of sediment yield from terrace units during two seasons. Listed are effective maximum infiltration rate (m) and observed runoff coefficient (rc) (cf. Chapter 12), cumulative sediment yield (M) observed and modelled for the simulation period, absolute and relative difference between these, associated model efficiency (ME), modelled and observed sediment transport components and sediment concentration. Also listed are modelled sediment delivery ratio (SDR, expressing the amount of sediment from beds and risers leaving the terrace), initial transport mechanisms and characteristics of the layer of sediment deposited in the terrace toe drain, including the observed and modelled amount of sediment per unit toe drain area (m) at the start and end of the simulation period (where available) and the associated time-averaged fraction of toe drain covered with sediment (H). Note that sediment yields may vary from those listed in Tables 13.3 depending on the period for which rainfall intensity measurements were available (see text for further details).

283

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES

60

10

(a)
50

(b)
Cum. modelled SY (t ha -1 ) 8

Cum. modelled SY (t ha -1 )

40

30

20

10

2
1998/99 1999/2000

1998/99 1999/2000

10 20 30 40 Cum. observed SY (t ha -1 ) 50 60

2 4 6 Cum. observed SY (t ha -1 ) 8 10

140 120 Cum. modelled SY (t ha-1 ) 100 80 60 40 20 20 40 60 80 100 -1 Cum. observed SY (t ha ) 120 140 1998/99 1999/2000

40

(c)
Cum. modelled SY (t ha -1 ) 30

(d)

20

10 1998/99 1999/2000 10 20 -1 Cum. observed SY (t ha ) 30 40

25

10

(e)
20 Cum. modelled SY (t ha -1 )
Cum. modelled SY (t ha )
-1

(f)
8

15

10

5 1998/99 1999/2000 5 10 15 -1 Cum. observed SY (t ha ) 20 25


-

2 1998/99 1999/2000 2 4 6 -1 Cum. observed SY (t ha ) 8 10

Fig. 13.6. (a-f) Comparison of modelled and observed cumulative sediment yield for the six terraces during the 1998/99 (line with dots) and the 1999/2000 (dashed line with open circles) modelling periods. Figure code refers to respective terrace unit plot codes.

The TEST model suggested that wash transport was responsible for 10-35% of total terrace soil loss, resulting in overall average wash sediment concentrations of 0.5-12.5 kg m-3 (cf. Table 13.7). The remaining sediment consisted almost exclusively of material in the toe drain entrained by the flow (re-entrainment sensu Rose, 1993). Specific D20 and j values for sediment in the toe drain were not available, but calculations with values found for the terrace risers (likely to represent over-estimates) suggested that the

284

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES

1000
entrained / 'bed load' washed / suspended load Total sediment yield

Modelled SY (t ha )

100

-1

10

0.1 0.1 1 10 100


-1

1000

Observed SY (t ha )

Fig. 13.8. Correlation between observed and modelled sediment yields from the six terraces during the two consecutive modelling periods in 1998/99 and 1999/2000. Modelled entrained and washed sediment are compared to sediment settled (bed load) and in suspension at the time of sampling.

transport of material re-detached by rainfall and subsequently washed away accounted for less than 1.7% of all sediment (cf. Table 13.7). Splash could be ignored as a means of transport at the scale of bench terraces: calculations indicated that it provided less than 0.06% (<8.7A10-3 t ha-1) of sediment yield in all cases, although this only involved material splashed out of the toe drain, whereas in reality the edges of terraces may also contribute some sediment to adjacent trails and gullies. The overall concentration of reentrained sediment was predicted to be 3-34 kg m-3. This range is much greater than that found for washed sediment, which is explained by the influence that toe drain gradient (Sdrain), fractional cover of the deposition layer (H) and the presence of mulching all had on modelled flow-driven transport.

13.5.4. Parameter sensitivity and uncertainty analysis


The sensitivity of the TEST model to the various parameters was evaluated using the parameter values derived for terrace A during the 1998/99 season. Parameter values were increased and decreased independently by 50%; the original values and the relative changes in total sediment yield associated with these perturbations are listed in Table 13.9, grouped into parameters relating to runoff prediction, flow-driven erosion and rainfall-driven erosion components and terrace dimensions, but otherwise sorted in order of decreasing sensitivity. Not surprisingly, one of the most important parameters is the average maximum infiltration rate (m). Runoff amounts and effective runoff rates were predicted by the model presented in Chapter 12 and a discussion may be found there. For this particular data set flow-driven transport was responsible for 75% of total sediment yield (cf. Table 13.9), and therefore parameters relating to this model component were generally more important than those related to the rainfall-driven component. Of
285

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES

Parameter

Value

Change (%) in soil loss after change of +50% -50% -36b 157c 27 25 14 -10 -10 11 -7.5 -6.0 4.2 17 13 -4.1 -4.1c -4.0 3.9 -3.0 1.1 0.2 -3.7 0.3 -0.8 82 b -61 -31 -29 -20 21 19 -16 13 11 -6.8 -17 -13 7.0 6.7 6.2 -4.0 4.0 -1.5 -0.3 3.7 -1.5 0.8

Runoff generation

m
Flow -driven erosion m(initial) H Sdrain F , C (mulch in drain) n L Rainfall-driven erosion D20 j (crop cover) C (crop cover) , C (mulch on bed) %S , C (riser cover) Sriser Sbed Terrace dimensions Wbed Wdrain Wriser
a b

124 / 228 mm h-1 a 1.0 39 kg m-2 0.00720 m2 kg-1 3.5 % 1680 kg m-3 0.106 m s-1 0.115 3.34, 0.35 0.0159 m1/3 s-1 35.8 m 11.7 / 15.9 g J-1 d 3.98 / 2.15 % d 1.36 and 0.0 e 0.33-0.87 2.45, 0.35 0.031 / 0.089 m d 2.02, 0.30 69.1 % 8.0 % 4.93 m 0.35 m 1.16 m

values refer to time before and after harvest, respectively total runoff changed by -35% and +90%, respectively c 50% increase led to values in excess of unity d values refer to terrace bed and riser sections, respectively e referring to a pre-harvest mixed crops and post-harvest mono-cropped cassava, respectively

Table 13.9. Sensitivity analysis of the TEST model parameters, used parameter values for terrace A during the 1998/99 simulation period. Listed is the percentage change in cumulative modelled sediment yield after a 50% increase or decrease in the tested parameter value.

particular importance in this respect were erodibility (), the initial amount of deposited sediment in the toe drain (m) and the coefficient that relates the latter to the area covered (H; Table 13.9). Other sensitive parameters appeared to be the gradient of the toe drain (Sdrain), the sediment density (), sediment depositability () and the effective fraction of stream power (F). Of the parameters pertaining to rainfall-driven erosion, soil detachability (D20) and the fraction of detached sediment remaining in suspension (j) were of similar importance. Other parameters, including geometric aspects such as terrace length and the width and gradient of the bed and risers, appeared to be less significant (Table 13.9).

286

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES

Plot

Original ME 0.87 0.52 1.80 2.47 1.01 0.64 0.85 0.72 0.3 0.68 0.99 66 %

and H optimised H m2 kg-1 2.1)10-2 5.0)10-2 9.3)10-5 4.4)10-6 2.3)10-1 4.1)10-1 4.1)10-1 3.3)10-2 1.9)10-2 1.9)10-1 1.4)10-1 120 % ME 0.59 0.33 0.66 0.67 <0 <0 0.73 0.48 0.75 0.67 0.72 0.66

Only H optimised H m2 kg-1 1.1)10-2 4.8)10-3 3.9)10-3 2.9)10-3 0 0 1.8)10-1 7.2)10-3 8.8)10-3 7.2)10-3 1.3)10-3 6.3)10-3 2.3)10-2 236 % ME 0.59 0.32 0.65 0.59 <0 <0 0.73 0.45 0.74 0.67 0.71 0.65

1998/99

1999/2000

A B C D Ea Fa A B C D E F

0.55 0.29 0.48 0.21 <0 <0 0.36 0.45 0.74 0.67 -0.29 0.65 )

Average coeff. var.


a

best fit was obtained without any entrainment, see text for explanation.

Table 13.10. Values of and H for individual data-set derived by Levenberg-Marquardt optimisation, using model efficiency (ME) as a criterion. Model efficiency values associated with the use of original parameter values (=1.0, H=5.2)10-3 m2 kg-1) are also listed. Modelefficiencies that were increased by more than 5% of initial variance are printed bold.

To investigate the uncertainty surrounding the parameters and H, these were optimised separately for each of the twelve data sets. Resulting optimum values and the effect on model performance are listed in Table 13.10. Although widely varying values resulted for both parameters, these did little to improve model performance in eight out of twelve data sets. Of the remaining four data sets, one resulted in a value for that was very close to unity (1.01) and two others resulted in unlikely high values (1.80 and 2.47). The fourth pertained to terrace E, for which important reasons for poor model performance were already discussed in Section 13.5.3. The theory-based assumption that equalled unity is therefore not likely to be the most important source of model errors. The optimised value for H was highly variable between data sets, even when it was assumed that =1. Possible reasons for this are discussed in Section 13.6.4.
13.6. Discussion

13.6.1. Soil loss from bench terraces


In an attempt to interpret runoff and sediment yield amounts from terrace riser, bed and unit plots measured during the various periods, the amounts measured during the incomplete years of 1998/99 and 2000/01 were compared to amounts observed during the corresponding period in the (fully measured) 1999/2000 year. For example, if sediment yield from a bench terrace was measured between 2 December 1998 and 11 April 1999, the amount of sediment generated between 2 December 1999 and 11 April 2000 was
287

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES

Treatment Terrace riser little or no protective cover newly planted cover well-established cover Terrace bed bare soil clean-weeded cassava ginger weeds mixed crops Terrace unit with bare riser with riser protection with riser protection and toe drain conservation

Runoff rc in % 19 (8-30) 12 (10-14) 8 (1-16) 22 (16-27) 25 (13-36) 10-11 10-20 9 (0.1-24) 13 (6-23) 12 (4-26) 9 (6-13)

Annual soil loss t ha-1 yr-1 200 (59-393) 149 (118-209) 62 (1-100) 94 (71-118) 80 (30-123) 30-34 11-24 15 (1-31) 59 (2-148) 31 (8-74) 9 (6-13)

Fraction fines % 6 (3-13) 8 (2-24) 8 (2-24)

Annual transport kg m-1 yr-1 16 (5-22) 14 (4-26) 5 (0.1-11)

Plotyears 10 3 10 4 8 2 2 8 8 5 3

21 (19-24) 22 (16-30) 20 (17-23) 19 (7-29) 23-29 8 37-39 3-6 31 (16-48) 4 (0-8) 43 (16-95) 25 (14-37) 89 (81-100)

Table 13.11. Average and range of estimated annual runoff (expressed as runoff coefficient, rc), annual soil loss, the percentage of fine sediment and annual transport (where applicable, per unit length of toe drain) associated with terrace risers, beds and units under various forms of management, as based on erosion plot measurements (see text for explanation).

determined and this amount was compared with the sediment yield for the entire year (25 October 1999 to 24 October 2000). Comparison of the latter two values yielded a correction factors of 1.09 to 1.57 or the various data sets and these were used to estimate annual soil loss. A similar approach was followed to estimate annual runoff coefficient. Clearly, the derived runoff and sediment yield estimates should be considered approximate only, but at least they allow an initial comparison of annual sediment yields. The resulting estimated averages and ranges of runoff coefficient and sediment yield from terrace risers, beds and units under various forms of management are listed in Table 13.11. Despite the wide ranges of estimates of soil loss from bench terraces with unprotected risers (2-148 t ha-1 yr-1), these figures compare reasonably well to earlier measurements on bench terraces A to D: Purwanto (1999) reported sediment yields of 97-242 t ha-1 during the 1994/95 and 1995/96 wet seasons. Similar values of 28-87 t ha-1 (5 months, P.1500 mm) have been reported for bench terraces in volcanic and sedimentary soils in South-central Java (Bruijnzeel and Critchley, 1996) whereas annual soil loss of 26 t ha-1yr-1 (P=2515 mm) was observed on level bench terraces in Eutropept soils in Central Java (Haryati et al., 1995). However, the wide ranges also demonstrates the limited usefulness of such comparisons without taking into account terrace dimensions, slope, rainfall characteristics and vegetation and soil cover. This is where a process model can be of assistance. The current study has made clear why the bench terraces are still eroding at high rates: important reasons are the poor cover that is maintained on the terrace risers, the limited protection provided by clean-cultivated cassava crops and, to a lesser extent,
288

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES


other crops, and the lack of soil conservation measures in the toe drain. In principle, large reductions in soil loss are feasible by ways that may be classified as aiming to: (i) decrease surface runoff; (ii) decrease the transport of material to the toe drain; or (iii) decrease the transport of sediment through and out of the toe drain. The first possibility was already discussed in Chapter 12 and essentially involves denser vegetation and soil surface cover on the terrace risers and beds, which also appears to be the most effective way to stop erosion from terrace beds and risers. A well-established vegetative riser cover reduces runoff from an average 19 to 8% of annual rainfall, whereas estimated erosion rates are reduced from 200 to 62 t ha-1 yr-1 (69% decrease; Table 13.11). Riser protection also somewhat reduced runoff from entire bench terraces from 13 to 12% of rainfall and soil loss from 59 to 31 t ha-1 yr-1 (a 47% reduction). Similarly, the presence of crops on the terrace bed generally helped reduce runoff and erosion, from 22-25% of rainfall and 80-94 t ha-1 yr-1, respectively, on bare or poorly protected soil, to 9% and 14 t ha-1 yr-1 under mixed cropping. An exception is provided by the clean-weeded cassava crop, which did little or nothing to reduce soil losses. Results from the current study confirm the widely held notion that cover close to the soil is more effective in interfering with erosion processes than cover by of a canopy well above the soil surface (e.g. Rose et al., 1997b). The canopy breaks the erosive force of rainfall but also creates bigger throughfall drops, which may obtain considerable erosive power as they fall from increasing heights. The cassava crop in particular has a high canopy. This may have created throughfall drops that were more erosive than the original rainfall, particularly at low rainfall intensities, possibly explaining why erosion under this type of cover appeared to be even more severe than on bare soil. The less pronounced effect of mulch application at the start of the season, when compared to mulch applied after harvest, corresponds with splash measurements on the same terrace bed reported in Chapter 9. While the effect of ca. 46% mulch cover was to reduce splash transport by as much as 84% during the first two weeks, its effectiveness quickly decreased, and after two months splash transport on the mulched plot even exceeded that on the bed without mulch cover. As the rate of erosion decreased generally the percentage of fine particles increased. For example, coarse material could constitute as much as 71-81% of sediment from bare or poorly protected beds and 87-97% on bare risers but was in most (if not all) cases considerably less important when soil losses were lower. It illustrates that loss of coarse material is prevented more easily than that of fine particles. In combination with adequate riser cover, the use of silt pits and mulching in the toe drain stopped almost all coarser particles from leaving the terrace and also retained more of the generated runoff; compared to poorly managed bench terraces runoff was estimated to be reduced by onethird (to 9% of rainfall) whereas soil loss was decreased more than six-fold (to 9 ton ha-1 yr-1). However, fine particles in suspension will not settle out easily, and conservation measures that interfere with runoff in the toe drain will be less efficient in retaining this material. It is this fine sediment that plays an important role in the nutrient loss by erosion (cf. Palis et al., 1990; Hashim et al., 1997). For similar reasons, constructions beyond the scale of terrace units such as gully plugs or check dams will be partially effective at best, apart from being rapidly filled. The current study therefore provides strong support to the contention that soil conservation measures are most effectively applied at the grass root level, i.e. focused on preventing initial runoff generation and rainfall detachment, rather than trying to stop sediment that is already on the move.

289

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES 13.6.2. Sediment characteristics and flume experiments
The measured sediment density of 2.53A103 kg m3 is slightly lower than the 2.60A1032.65A103 kg m3 normally assumed for kaolinite and quartz, the two main sediment constituents (Chapter 3). The slightly lower values may have been caused by some air trapped in the pycnometer or, alternatively, by the few percents of organic matter present in the soil. At 1.68A103 kg m3, the measured density of wet aggregates was within the range suggested by Knisel (1980) for use in the CREAMS model (1.6A103-1.8A103 kg m3). Comparison of the densities of wet aggregates and primary particles suggests that the aggregates have a porosity of ca. 24% compared to 64% for the bulk topsoil (Chapter 6). The depositability of sediment used in the flume experiments (=0.106 m s-1) was at the high end of the range of values reported in the literature (0.022 m s-1 for a Thai clay-loam to 0.094 m s-1 for a Chinese silty clay-loam; Yu et al., 1999). For material dominated by clay and silt, the size of aggregates is clearly important in determining soil depositability. The sediment was well-aggregated into particles with a median water-stable size of ca. 1.1 mm (Chapter 3). At the same time, the material also had a very fine fraction that was possibly more easily mobilised than the coarser aggregates and certainly easier transported. For example, all sediment transported at low stream power (approximately <0.05 W m-2) was sized less than 0.15 mm. With increasing stream power aggregate size and depositability increased to 0.072"0.031 m s-1 but never reached the value of the original sediment. These observations are consistent with the findings of Rose et al. (1997a). As a consequence, sediment left behind in the flume will have become gradually depleted of fine particles. This process also occurs in field situations (possibly enhanced by rainfall detachment) as also appears from the contrast in the percentage of fine particles in wet-sieved soil (>40%, Chapter 3) and toe drain sediment (<2.6%, cf. Table 13.4). Such selective processes also explain the rapid decrease of sediment concentrations after the start of the experiment, comparable to the effects of shielding well-documented in laboratory experiments (Heilig et al., 2001). The GUEST runoff entrainment theory used here assumes steady state equilibrium conditions and cannot yet account for such phenomena. Velocity measurements made at different discharges suggested an average Mannings roughness coefficient (n) of 0.0159 s m-1/3, which is in the range of 0.0130.015 s m1/3 reported for comparable flume experiments by Proffitt (1993a). However, individual values appeared to decrease from 0.6-0.12 s m-1/3 at very low discharges to 0.013-0.016 at the highest discharge (ca. 6A10-4 m2 s-1). A decrease in roughness coefficient with increasing velocity was also documented by Rose et al. (1997a) and the apparent partial failure of Mannings equation may be caused by the concentration of runoff into shallow rills, particularly at higher discharges (cf. Takken et al., 1998). Furthermore, the applicability of Mannings equation has recently been questioned for situations with thin sheetflow, where the depth of flow may be in the same order of magnitude as surface roughness (e.g. Lawrence, 1997; Govers et al., 2000). Given these restrictions to theory applicability, it is not surprising that calculated effective stream power fractions (F) varied considerably. An optimised value of F=0.115 still predicted the observed concentrations rather poorly. The variation of sediment concentration during experiments and the contrasts between replicate experiments were related to the incidence and severity of rilling, as was also observed by Proffitt et al. (1993a) and Mosely (1974). Indeed, Proffitt et al. (1993a) suggested that real transport limited conditions only exist directly after sediment has been added to the flow, for example by

290

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES


rill wall collapse. This clearly creates problems in using the entrainment theory for modelling purposes. In principle the existence of rills can be included in the theory (cf. Hairsine and Rose, 1992b), but this requires additional information that was not routinely collected during the experiments and not at all available in the field. Having said this, the derived average value for F is within the documented range of 0.08 (Hairsine and Rose, 1992b) to 0.15 (Hairsine, 1988), providing some additional support for the usefulness of the entrainment theory.

13.6.3. Modelling erosion on terrace risers and beds


The current study amply demonstrated that rainfall-driven erosion processes are the crucial first steps in soil loss from the back-sloping bench terraces. It suggested that erosion on the terrace bed and riser occurs almost exclusively through rain splash and wash of fine particles. In the field no rills were observed on back-sloping bench terraces, although they were seen on some forward-sloping terraces. From the perspective of GUEST theory, flow-driven processes are likely to be absent if stream power is low (<0.008 W m-2; cf. Proffitt et al., 1993a). Using the roughness coefficient found in the flume study (n=0.0159 s m-1/3), characteristic values for riser and bed slope length and gradient of 1.0 m, 2.5 m, 60% and 10%, respectively, and an somewhat arbitrary likely maximum runoff rate of 100 mm h-1, the corresponding stream power was estimated at 0.07 and 0.16 W m-2 for the bed and riser, respectively (cf. Eqs. [13.10]). At first sight this suggests that the erosive power of sheetflow may occasionally have been enough to entrain particles. Associated flow depths would be very low, however, at 0.5 and 0.2 mm for the bed and riser, respectively. On terrace risers and beds transport by rain splash appeared to be more important than wash transport, particularly on bare risers: as much as 91"6% of all material coming from the risers appeared to be transported by splash, whereas this was still 56"19% on the terrace bed (Table 13.7). The small width of risers and beds is the most important reason for this dominance of rain splash (cf. Eq. [13.4]). The steepness of the risers also promotes splash transport, while the generally good soil infiltration characteristics limit transport by runoff. These findings correspond with field and laboratory studies on the same soil, although these also indicated that gravity-driven transport such as splash-creep and rolling of aggregates provided an additional 6-50% of material on steep risers (Chapters 9 and 10). The derived erodibility values (D20=11.7 and 15.9 g J-1 for the beds and risers, respectively) are in the same order of magnitude as values found in earlier studies in this environment. Splash cup experiments on terrace beds with little vegetation cover produced values of 3.5-18.5 g J-1, while approximate values of 11.0-21.7 g J-1 were measured on 30-40E terrace risers (Chapter 9). Measurements of splash on soil trays suggested detachability values of 7.5-20 g J-1 (Chapter 10). In all cases there was considerable temporal variability and a tendency for detachability to decrease during the course of the season. Such temporal changes are not taken into account by the model and may well be responsible for much of the unexplained variation. In addition, D20 values will presumably have varied between plots and thus the calibrated values represent a very limited sample only. For example, if the data set for plot B-II-6 had been used for calibration instead of B-II-3, a value of D20=12.1 g J-1 (i.e. 4% higher) would have resulted.

291

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES


Many of the earlier remarks about spatial and temporal variability also apply to the wash fraction (j). The calibrated values of 4.0% and 2.2% for the terrace bed and riser, respectively, are similar to those based on soil tray measurements: values of 2.3% to 4.5% were found for bare soil, with no apparent relationship with slope gradient (Chapter 10). An indication of the approximate settling velocity of the washed sediment can be obtained by comparing the depth of flow (estimated earlier at a maximum of 0.2-0.5 mm) and the time needed for runoff to travel to the end of the plot (estimated to be at least 1060 seconds; cf. Chapter 12), which suggests settling velocities of less than 10-5-10-6 m s-1. Estimates of the settling velocity of sediment still in suspension at the time of sampling can be made by considering the water depths in the erosion plot collecting drums and basins (up to ca. 40 cm) and the time passing between the end of the storm and the time of sampling (usually some 5 to 20-hours). This leads to settling velocities of less than 10-6-10-5 m s-1 and therefore the suspended sediment will have comprised mainly individual primary clay and silt particles or small clusters of them (cf. Proffitt et al., 1993b). This was confirmed by laser particle size analysis (Konert and Vandenberge, 1997) of undispersed runoff sediment samples from the erosion plots: median particle sizes of 5-20 m were measured with very few particles coarser than 50 m. In principle, the washed sediment cannot be equated to suspended sediment: fine material may be transported by splash as well, whereas conversely some coarser material dislodged close to the end of the plot may have been transported by runoff. Nevertheless, average modelled concentrations of washed sediment (1-8 g l-1 for terrace bed plots and 1-5 g l-1 for terrace riser plots) corresponded well with observed concentrations of suspended sediment (1-5 g l-1 for the terrace bed plots and 1-4 g l-1 for the risers, cf. Table 13.7). One of the reasons for the contrast in washed fraction (j) values found for terrace beds and risers may be related to the structure of the risers. The upper part of the riser typically consisted of loose, well-aggregated topsoil while the lower part was excavated in a more massive, less erodible subsoil. This subsoil amounted to 0-50% of total projected riser width, the fraction increasing with the height of the riser and, therefore, with the gradient of the original hillslope (cf. Chapter 12). While more runoff will have been generated on the less permeable subsoil, detachment will have been less on this area. As a consequence, less sediment will be carried in suspension which explains the lower wash fraction. The presence of this subsoil also has some consequences for transport by splash. Field studies of splash transport were made on the looser topsoil and sediment splashed from this part of the riser onto the subsoil needs to be transported to the toe drain (Chapter 9). It was observed in the field that the bulk of sediment splashed onto the massive subsoil was rapidly transported to the underlying toe drain by a combination of splash and runoff transport. At the same time, very little sediment was splashed from the subsoil upslope onto the topsoil. The subsoil therefore neither provided nor stored sediment and downslope splash transport from the topsoil section may be considered a net loss from the riser, justifying the approach followed. Another conceptual problem ensued from the installation of erosion plots on the terrace beds: material splashed beyond the downslope border of the erosion plot represented a net loss of sediment, whereas sediment in the toe drain normally may be splashed back onto the terrace bed. Support for the approach taken comes from the fact that a slight but steep ramp between the terrace bed and toe drain is created by hoeing at the start of the season. Material splashed down the ramp presumably far outweighs the small amount that may be splashed upwards, although the ramp itself introduces additional complexity. The above not only serves to illustrate simplifications and abstractions that need to be made when modelling erosion, but also some of the
292

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES


interpretation difficulties associated with measurement of sediment yield from relatively small plots with artificial boundaries (cf. Chapter 8). Finally, the mulch coefficients found in the present study are somewhat lower than the S=4.4 that described the decrease in splash and wash transport observed in soil tray experiments after application of 55% surface mulch cover (Chapter 10). This may be related to the longer time of observation in the present study, giving the mulch more opportunity to decompose and be buried with the sediment (cf. Chapter 9). It is difficult to compare values derived in this study with those reported in the literature, as the latter vary with respect to the type of erosion that was measured (splash, wash, runoff or gross plot sediment yield) and the scale and manner of measurement.

13.6.4. Modelling erosion on terrace units


Despite the satisfactory overall performance of the TEST model, some problems were encountered with regard to modelling flow-driven transport, in particular with the parameter H. Because sediment in the toe drain had already been detached it was noncohesive and easily re-entrainable and hence the erodibility coefficient () was assumed to equal unity (i.e. it was essentially omitted, cf. Rose, 1993). Testing showed that this value produced a reasonable fit with data from the flume experiments, while optimising its value using sediment yield data in most cases did little or nothing to improve model performance. In the flume experiments any unexplained variation in sediment concentrations ended up in the effective stream power fraction (F), while in field applications of the GUEST model these variations usually end up in the erodibility coefficient ( ; e.g. Rose et al., 1997a; Yu and Rose, 1999b; Yu, in prep.). In the subsequent field application this role was assumed by the parameter H that defines the relationship between the amount of sediment (m) accumulated in the toe drain and its fractional cover (H). Unfortunately, the (optimised) parameter H and the initial value of m appeared to be among the most sensitive. Reasonable estimates of initial m were available for the first simulation period, but had to be estimated for the second. Also, observed values were based on surveys of the entire toe drain, whereas for modelling purposes conditions at the end of the toe drain may have been more important. Optimising H lead to a remarkably low value of 7.2A10-3 m2 kg-1, although for individual data sets the optimum value for this parameter showed a wide range. This H value predicts that 100 kg m-2 of sediment would still cover only 51% of the toe drain. Given the sediment bulk density (ca. 950 kg m-3), this implies that the deposited layer had an average thickness of ca. 20 cm between uncovered areas. Although much sediment was deposited against the riser base, this does not agree very well with field observations, which suggested that H is more likely to have been an order of magnitude greater at 0.02-0.07 m2 kg-1. In many cases the flow was concentrated in shallow channels that would occasionally cut through the deposition layer to the subsoil. As a consequence, the value of H for the channel bed would have been lower than that for the toe drain as a whole and this may go some way to explain the low H value. In addition, many of the discussed reasons for the variations in F found in the flume experiments will also have had a bearing on H. It is difficult to envisage a way to describe the concentration of flow in channels without greatly adding to the conceptual and mathematical complexity of the model. Nonetheless, in its current form the model provided estimates of event and total sediment yield that showed reasonable accuracy.

293

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES


13.7. Implications for erosion process modelling

The TEST model was able to predict runoff and soil loss from backsloping bench terraces with encouraging accuracy. At the same time, the number of parameters used remained limited, while preserving the physical consistency of the model theory. The model and its components were tested over a range of scales, from the decimetre or even centimetre scale (Chapters 9 and 10) to terrace units of 53-231 m2. The processes affecting sediment delivery from progressively larger areas were identified and welldescribed and this provides scope for up-scaling the model: one such attempt is made for two bench-terraced 4 ha hillside sub-catchments in Chapter 14. Model performance was possibly enhanced by the limited spatial variations in micro-relief, soil characteristics and vegetation and soil cover. Testing of the model in a more complex environment presents a future challenge. The explicit physical description of the various processes responsible for soil loss from the terraces provides obvious advantages over the more empirical (R)USLE approach. At the same time, the (R)USLE has already gained wide acceptance among workers in the field of soil conservation. Within the limits of the simulated conditions, the model presented here may assist in attempts to capture the effects of cover management and terrace design on soil loss in CP factors in the context of the (R)USLE for use in Indonesia's uplands (cf. Agus et al., 1998). An important aspect identified here relates to the contrasts in transport mechanisms between coarse and fine sediment particles. Whereas runoff transport of coarse material may be described well by GUEST theory, assuming an equilibrium between entrainment and deposition, the difference between travel velocity and settling velocity for fine (e.g. primary silt and clay) particles, whether detached by rainfall or entrained by the flow, may well be too great to reach equilibrium during transport. The different behaviour of the two types of sediment, evident from the fact that part of the sediment in the collecting vessels had not settled even after many hours, is also acknowledged with regard to sediment transport in rivers (e.g. Einstein and Chien, 1953). A distinction between the two corresponding forms of transport was realised here by equating them to rainfalldetachment and runoff-entrainment dominated transport, respectively. While this produced good results, it is readily admitted that conceptual objections can be made against this approach and a more rigorous approach seems feasible. Work in this area, describing the (im-) balance between material going into transport and that settling out as a function of travel distance and/or time is currently in progress (Heilig et al., 2001; Rose et al., 2002; Hairsine et al., 2002; Sander et al., 2002) In any case, the apparent importance of rainfall-driven flow transport in the current study confirms the findings by Yu and Rose (1999b), who suggested that rainfall-driven runoff transport could have played a significant role on erosion plots (25-36 m, 30-40%) on Chinese and Thai clay loam soils with crop cover. The relative importance of rainfalldetached sediment in runoff will depend, inter alia, on soil dispersivity, the amount of runoff generated, slope length and gradient. More research is needed to quantify these effects (cf. Proffitt and Rose, 1991; Rose et al., 1997a). Other issues requiring attention include the hydraulics of thin sheet flow (cf. Govers et al., 2000) and the sources of the variability contained in the parameters H, F and/or in field applications of the GUEST theory. Finally, there is scope for attempts to relate important soil erosivity parameters, such as splash detachability (D20), wash fraction (j) and runoff erodibility () in the current context, to measurable properties. For example, splash detachability and transportability have been related to soil shear strength measured by a Swedish fall cone
294

CHAPTER 13 - EROSION PROCESSES ON BENCH TERRACES


device (Bradford et al., 1986) and parameters calculated from the particle size distribution (Poesen and Savat, 1981; Bradford et al., 1987). Similarly, temporal patterns in soil erodibility () were found to correlate with changes in shear strength measured with a Torvane (Rose et al., 1997a). It is likely that the wash fraction (j) depends to a large extent on soil dispersivity, whereas the effective stream power fraction (F) and Mannings roughness coefficient (n) may also bear a relationship to soil characteristics. Combined with the obvious dependence of depositability () on the aggregate size distribution, this suggests that there may be only a limited number of sets of values for the above-mentioned parameters and this could prove a fruitful area of further research.

295

You might also like