Sie sind auf Seite 1von 33

225

Chapter 12
Measurement and modelling of surface runoff on
bench terraces
Abstract: Infiltration-excess overland flow is an essential component of many erosion and
catchment runoff models. The spatially variable infiltration (SVI) model of Yu et al. (1997c) and
a set of derivative equations relating runoff depth and maximum effective runoff rate to storm
depth, depth-averaged rainfall intensity, average maximum infiltration capacity and an additional
amount of initial infiltration, were validated and tested on back-sloping bench terraces in volcanic
upland West Java, Indonesia. Measurements were used of runoff rate and depth from 31 small
(1.0-8.2 m
2
) bounded plots representing sections of terrace beds or risers and from six larger (53-
231 m
2
) terrace units with hydrologically defined boundaries. Runoff rates modelled using
rainfall intensity data corresponded well with observed patterns and the storm-based equations
were used successfully to model event runoff depths and maximum effective runoff rates.
Empirical relationships were found between measured effective runoff rate and modelled
(maximum) values. Resulting values for maximum average infiltration rate (I
m
) varied between
18 and 443 mm h
-1
and reflected effects of vegetation or mulch cover and soil compaction. The
amount of initial additional infiltration was 0-13 mm and exhibited an inverse relationship with
I
m
. It is concluded that the SVI model and the derived equations provide a robust and accurate
method for predicting runoff at the investigated (plot) scale.
Parts of this chapter are submitted to Journal of Hydrology with L.A. Bruijnzeel as co-author.
12.1 Introduction
Soil erosion, downstream flooding and siltation pose a major challenge to watershed
managers, particularly in the humid tropics with their high rates of deforestation and
intense rainfall. Knowledge of the volume and rates of runoff generated in response to
rainfall is very important, if not quintessential, to predicting soil losses. Although runoff
may be generated in a number of ways (Ward, 1984; Brammer and McDonnell, 1996),
Hortonian infiltration-excess overland flow may well be the dominant mechanism on
bare, degraded soils (Kirkby, 1978; Hudson, 1995). A simple model describing the
generation of infiltration-excess runoff was presented by Green and Ampt (1911). The
model basically includes an initial amount of infiltration (needed to bring the soil to
steady-state conditions) after which infiltration is assumed to proceed at a maximum
final rate. The Green-Ampt infiltration model is widely used, including in the more or
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 12 - RUNOFF FROM BENCH TERRACES
226
less physically-based WEPP (Water Erosion Prediction Project) model, aimed to replace
the traditional Universal Soil Loss Equation (USLE) (Laflen et al., 1995). One
adjustment made to the Green-Ampt model in the WEPP context relates to the
observation that the apparent final infiltration rate varies with the magnitude of rainfall
rates (Risse et al., 1995). Such findings already raise some doubt as to the physical basis
of the Green-Ampt model, whereas Yu (1999) demonstrated further inconsistencies
between rates of infiltration-excess runoff observed on small erosion plots and
simulations by Green-Ampt theory. He argued that these inconsistencies were probably
due to the implicit assumption in the Green-Ampt model that the final infiltration rate is
uniform over the entire area. This implies that an entire area suddenly starts to generate
runoff once rainfall rate exceeds the final infiltration rate, a phenomenon that was not
observed in reality. In response to this deficiency, Yu et al. (1997c) advanced a different
infiltration model that assumes an exponential spatial distribution of maximum
infiltration rates. This so-called spatially variable infiltration (SVI) model involves an
infiltration rate that increases non-linearly with rainfall intensity and eventually
approaches an average maximum infiltration rate (I
m
in mm h
-1
) when almost the entire
area is generating runoff. The model also includes an initial amount of infiltration (F
0
in
mm) similar to the Green-Ampt model. Besides spatial variations in infiltration capacity,
another argument in favour of this model formulation is the highly uneven distribution of
throughfall and stemflow below a (forest or crop) vegetation canopy (e.g. Lloyd and
Marques-Filho, 1988; Chapter 5). Yu (1999) also compared the performance of the SVI
model with that of the Green-Ampt model using rainfall-runoff rate data from various
locations in Australia and South-east Asia. In both cases, a kinematic wave
approximation was used to simulate the attenuation of runoff peaks over travelled
distance. The performance of the SVI model was superior to that of the Green-Ampt
model (Yu, 1999).
A feature of any process-based infiltration model is that, ideally, rainfall intensity
data at short intervals (i.e. in the order of one minute) are needed to obtain accurate
results (Yu, 1999; Chapter 11). In response to this problem, in Chapter 11 an exponential
distribution of rainfall intensity versus rainfall depth was assumed during individual
storms and analytical expressions for the SVI model were derived which only required
depth-averaged rainfall intensity () and event rainfall depth (P) to estimate runoff depth.
During a test of the usefulness of the exponential rainfall depth-intensity model using
rainfall data for 30 storms (range 33-81 mm) the SVI model equations yielded runoff
depth estimates that were very close to values calculated using the original high-
resolution rainfall data. However, a number of assumptions that was made await further
testing. Firstly, the problem of simulating the initial infiltration amount (F
0
in the SVI
model) in the analytical storm-based equations developed in Chapter 11 was overcome
by replacing this term by an amount of excess infiltration (S
I
in mm), i.e. in addition to
the calculated amount. In addition, the effect of the choice of the amount of initial
infiltration or the attenuation of runoff peaks on calculated Q
e
values were not included,
although it was hypothesised in Chapter 11 that for large storms initial infiltration would
not have much of an effect on calculated Q
e
values. The Q
e
values calculated with the
analytical equation should probably be considered maximum values (denoted by Q
e
in
Chapter 11) that would need to be downscaled empirically to actual values.
In this chapter we aim to validate and test the mathematical theory developed in
Chapter 11 and the various assumptions referred to. To do this measurements of rainfall
intensity, runoff rate and depth are used collected within the context of a research project
focusing on runoff and erosion processes on bench-terraced hillslopes in volcanic upland
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
227
West Java, Indonesia. Each back-sloping bench terrace comprises a terrace riser and a
terrace bed, with a toe drain in between (Fig. 12.1), each of which can be expected to
have reasonably homogeneous, but contrasting, infiltration characteristics. The SVI
model theory assumes an exponential spatial distribution of I
m
over the considered area.
However, if the overall runoff comes from a number of contributing areas whose
individual infiltration characteristics are well-described by the SVI theory but having
contrasting average values then, strictly speaking, the infiltration behaviour of the area at
large in many cases will not be accurately described anymore by an exponential spatial
distribution. This makes the investigated environment a good test case for evaluating the
ability of the SVI model to simulate runoff from an area with contrasting contributing
zones. The consequences of the present findings for the modelling of sediment transport
by surface runoff are dealt with in Chapter 13.
12.2. Study area
The research was conducted in the 1 km
2
upper catchment of the Cikumutuk river
near the town of Malangbong, about 40 km East of Bandung, West Java, at an altitude of
580-610 m a.s.l. (7-03S, 108-04W). Slopes are generally fairly steep at about 15- and
mostly terraced The underlying Quaternary volcanic tuffs have weathered to kaolinitic
Oxisols that consist of several decimetres of highly permeable, well-aggregated soil on
top of a less permeable, massive subsoil (Table 12.1). The area experiences a humid
tropical climate with a drier season (average monthly rainfall less than 60 mm) generally
extending from July until September and a mean annual rainfall of about 2650 mm.
Further site information is provided in Chapter 3.
Fig. 12.1. General layout of back-sloping bench terraces, the three components sections and
their intersection with the subsoil. A hydrologically defined terrace unit is shown lightly
shaded.
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
228
A typical terrace unit consists of a terrace riser, a compacted toe drain that also acts
as an access path, and a terrace bed that slopes backwards towards the toe drain (Fig.
12.1). The terrace risers typically have slopes of ca. 35-50- (55-75%) and a projected
width of 0.5 to 1.3 m, whereas the terrace beds have slopes of 0-12- (0-20%) and a width
of 0.8 m or more, depending on the original slope (cf. Chapter 14). Despite the contrast
in permeability between topsoil and subsoil no subsurface stormflow emerged from the
terrace risers (cf. Fig. 12.1). Further support for this was provided by a tracer experiment
and by tensiometer observations (unpublished). The toe drains are typically 0.3-0.4 m
wide and have a slight gradient of up to 3- (5%) towards the end of the terrace, where
runoff and sediment from the terraces drain into a ditch or onto a trail. Terraces often
drain towards both sides with a water divide approximately in the middle, at 15 to 40 m
from either end. Crops planted on the rainfed terrace bed usually include inter-cropped
cassava and maize, sometimes with a third crops such as upland rice, groundnuts or
another legume. The crops are sown in or planted a few weeks after the first frequent rains,
normally at the end of October or beginning of November, and harvested in February or
March, except cassava which is usually left to grow until the end of the dry season. Further
details about cropping practices and seasonal crop development can be found in Chapter 5.
12.3. Materials and Methods
12.3.1. Plot characteristics
Runoff was measured from small (1.0-6.1 m
2
) plots encompassing sections of terrace
risers and beds (Artificial Boundary Erosion Plots; ABEPs) and from entire terrace units
(Non-imposed Boundary Erosion Plots; NBEPs). The terrace riser and bed plots had
Location Double ring infiltrometry Permeameter core testing Tension
infiltrometry
K
sat
Average
K
sat
Range (N)
K
sat
Average
K
sat
Range (N)
K
unsat
Range (N)
mm h
-1
mm h
-1
mm h
-1
mm h
-1
mm h
-1
Terrace bed topsoil
A, B 558 128-1020 (13) 326 9-474 (5) 16-18 (2)
C, D 459 117-895 (18) 426 8-639 (3) 4-20 (2)
E, F 877 800-954 (2) 260 18-625 (4) 16-64 (2)
Terrace toe drain / subsoil
a
A, B 86 14-218 (5) 71 61-80 (2) 2.9 (1)
C, D 56 19-120 (5) 14 1-21 (3) 5.0 (1)
E, F 2.5 (1) 4.9 3-9 (3) 4.6 (1)
a
Infiltrometer measurements pertain to terrace toe drains only, permeameter measurements on cores
pertain to both toe drains and subsoil.
Table 12.1. Measurements of saturated (K
sat
) and unsaturated conductivity (K
unsat
) of the
terrace bed topsoil and subsoil or terrace toe drain soil, as derived from double ring
infiltrometry, laboratory permeameter tests on soil cores and tension infiltrometry at 3 cm
suction.
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
229
parallel sheet metal boundaries oriented downslope at a spacing of 0.9 to 3.8 m. All plots
were laid out close to the six terrace unit plots (cf. Fig. 3.1; Tables 12.2 and 12.3). The
upper boundaries of the plots were defined hydrologically, but in some cases were
stabilised with sheet metal strips. The treatment of the risers varied from bare or
unweeded soil to newly planted or already well-established grass or shrubby herb
(Gendarussa vulgaris Ness). The terrace bed plots either had bare soil, mixed cassava
and maize (sometimes with groundnut or rice), mono-cropped cassava (in its first or
second year, clean-weeded or not), ginger planted on tied ridges, or soil left fallow and
invaded by weeds. Full details of the dimensions and treatments on the various terrace
bed and riser plots can be found in Tables 12.2 and 12.3, respectively.
The locations of the three pairs of terrace units, coded A to F, were chosen to
represent steep (C and D, 25-26- or 42-44%), intermediate (E and F, 14-17- or 24-29%)
Treatment Location, code Area W S C
v
C
s
rc
m
2
m % % % %
15 Dec 1998 - 11 May 1999: P=1775 mm
a
cassava-maize-peanut
b
AB B -I-1 6.1 4.7 7 54 (33-87) 35 4.8
cassava-maize-rice CD B -I-8 1.6 1.7 17 79 (43-95) 42 23.1
cassava-maize EF B -I-5 1.7 1.8 19 65 (15-94) 51 4.2
cassava-maize
a
EF B -I-12 4.3 0.9 2 78 (27-99) 43 0.1
1 Nov 1999 - 31 Oct 2000: P=2871 mm
bare EF B -II-3 3.8 2.5 16 - - 18.9
bare EF B -II-6 3.5 2.3 15 - - 16.4
ginger EF B -II-4 3.8 2.5 16 5 (0-9) - 10.4
ginger EF B -II-5 3.5 2.3 21 5 (0-10) - 10.6
cassava (weeded) EF B -II-9 3.8 2.5 11 36 (0-59) - 12.6
cassava (weeded) EF B -II-10 3.6 2.4 14 26 (0-41) - 16.5
cassava-maize-rice EF B -II-1 3.2 2.1 11 55 (0-91) - 13.6
cassava-maize-rice EF B -II-2 3.5 2.3 17 67 (0-92) 40 9.2
cassava-maize-peanut EF B -II-8 3.6 2.4 8 73 (0-96) - 5.7
cassava-maize-peanut EF B -II-7 3.5 2.3 9 70 (0-92) 29 9.2
1 Nov 2000 - 6 Apr 2001: P=1754 mm
bare EF B -III-4 3.8 2.5 16 - - 26.1
bare EF B -III-5 3.5 2.3 21 - - 24.0
cassava - fallow EF B -III-7 3.5 2.3 9 n.a. - 26.7
cassava - fallow EF B -III-8 3.6 2.4 8 n.a. - 15.3
cassava (weeded)- fallow EF B -III-1 3.2 2.1 11 n.a. - 34.2
cassava (weeded) - fallow EF B -III-2 3.5 2.3 17 n.a. - 28.4
cassava (weeded) - fallow EF B -III-9 3.8 2.5 11 n.a. - 28.9
cassava (weeded) - fallow EF B -III-10 3.6 2.4 14 n.a. - 30.6
fallow EF B -III-3 3.8 2.5 16 n.a. - 18.7
fallow EF B -III-6 3.5 2.3 15 n.a. - 9.5
a
plots measured from 4 January 1999 onwards only; corresponding rainfall 1418 mm.
b
runoff rate measurements used in runoff hydrograph simulation.
Table 12.2. Characteristics of the terrace bed plots (ABEPs). Listed are treatment, area, width
(W), slope gradient (S), canopy cover (C
v
), surface mulch cover (C
s
) and runoff coefficient
(rc) for the entire measurement period, together with average runoff coefficient (rc).
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
230
and relatively gentle (A and B, 9-12- or 16-21%) original slopes. The contributing areas
of the NBEPs were determined by field survey and varied in size between 53 and 231 m
2
,
with a terrace width of 2.3 and 6.4 m, depending primarily on original slope steepness
(Table 12.4). During the first wet season (1998/99) terraces A to D had a varying amount
of grass cover on their risers, while the risers of terraces E and F were left bare. The
terrace beds were planted with a mixture of cassava and maize in all cases, with peanuts
as a third crop on terraces A and B and with upland rice on terraces C and D. After
harvesting the maize (in January, February and March 1999, on terraces A/B, C/D and
E/F, respectively), the stalks were left on the bed and in the central drain. Before the
1999/2000 rainy season the partial grass cover was removed from the risers of terraces A
and C, whereas Gendarussa shrub cuttings were planted on the riser of terrace F. In
addition, on terraces B, D and F, silt pits (LxWxH0.5x0.3x0.2 m) were dug at the end of
Treatment Location, code Area W S C
v
C
s
rc
m
2
m % % % %
15 Dec 1998 - 11 May 1999: P=1775 mm
a
unweeded CD R -I-7 2.1 1.2 62 11 (4-18) - 12.0
unweeded
b
EF R -I-4 2.2 0.9 61 25 (10-39) - 20.0
unweeded
a
EF R -I-13 1.6 1.5 61 7 - 19.7
unweeded AB R -I-3 2.0 0.9 68 27(13-42) - 7.6
unweeded / shrub
c
CD R -I-9 2.5 1.3 73 2 / 44 - 13.4
grassed
a
CD R -I-11 2.0 1.2 70 10 (7-13) - 3.9
grassed
a
AB R -I-10 1.0 0.8 59 96 - 0.6
1 Nov 1999 - 31 Oct 2000: P=2871 mm
bare EF R -II-2 2.0 0.6 57 - - 22.6
bare EF R -II-3 1.9 0.5 52 - - 24.5
bare CD R -II-6 3.8 1.1 55 - - 11.8
grass planted EF R -II-1 1.3 0.3 58 54 - 10.8
grassed CD R -II-8 1.4 0.8 74 15 - 5.5
shrub planted CD R -II-5 3.8 1.0 55 15 - 10.0
shrub sparse CD R -II-7 2.4 1.2 68 22 - 10.0
shrub dense EF R -II-4 1.5 0.5 83 85 - 15.6
1 Nov 2000 - 6 Apr 2001: P=1754 mm
bare EF R -III-2 2.0 0.6 57 - - 21.4
bare EF R -III-3 1.9 0.5 52 - - 27.6
bare CD R -III-6 3.8 1.1 55 - - 13.0
grassed EF R -III-1 1.3 0.3 58 n.a. - 7.8
grassed CD R -III-8 1.4 0.8 74 n.a. - 11.3
shrub CD R -III-5 3.8 1.0 55 n.a. - 6.6
shrub CD R -III-7 2.4 1.2 68 n.a. - 4.6
shrub EF R -III-4 1.5 0.5 83 n.a. - 12.5
a
plots measured from 4 January 1999 onwards only; corresponding rainfall 1418 mm
b
runoff rate measurements used in runoff hydrograph simulation
c
shrubby herb planted as late as 9 February 1999
Table 12.3. Characteristics of the terrace riser plots (ABEPs). Listed are treatment, area, width
(W), slope gradient (S), canopy cover (C
v
), surface mulch cover (C
s
) and runoff coefficent (rc)
for the entire measurement period, together with average runoff coefficient (rc).
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
231
the toe drains, while a bunch of cassava stems was put just below these pits. An identical
mixture of cassava, maize and rice was planted on all terrace beds. Treatments were the
same in the 2000/01 season, but without conservation measures in the toe drain (Table
12.3).
Terrace unit NBEP A B C D E F
Terrace dimensions
Width m 6.4 4.2 2.8 2.3 2.7 3.5
Length m 35.8 34.8 19.0 27.0 22.9 26.0
Area m
2
231 147 53 63 63 91
Original slope gradient % 11 7 29 43 24 22
Fraction of subsoil exposed % 11 16 21 46 25 22
Projected width
bed m 4.9 3.2 1.2 0.8 1.3 1.7
riser m 1.2 0.7 1.4 1.2 1.0 1.4
drain m 0.35 0.30 0.29 0.35 0.42 0.44
Fraction of terrace
bed % 77 76 41 36 48 48
riser % 18 17 48 49 37 39
drain % 5 7 10 15 15 13
Slope gradient
bed % 8 17 24 25 19 21
riser % 69 76 64 75 68 65
drain % 3.5 3.7 5.3 5.8 1.5 2.0
Measurement characteristics
2 Dec 1998 -11 Apr 1999 (130 days): P = 1697 mm
Cover: riser C
v
% 30 98 25 0.75 - -
bed C
v
% 57 (33-87) 74 (35-95) 71 (37-99)
mulch C
s

a
% 35 30 43
Q
tot
mm 120 119 271 441 115 138
rc % 7 7 16 26 7 8
25 Oct 1999 - 24 Oct 2000 (365 days): P = 2567 mm
b
Cover: riser C
v
% - 98 - 75 - 35
bed C
v
% 55 (1-88) 52 (1-86) 59 (1-92)
Q
tot
mm 142 154 397 331 192 240
rc % 6 6 15 13 7 9
25 Oct 1999 - 6 Feb 2001 (104 days): P = 1367 mm
c
Q
tot
mm 312 n.a. 319 n.a. 208 63
rc % 23 n.a. 23 n.a. 15 5
a
mulch left after harvest
b
conservation measures in drains of terraces B, D and F
c
cover fractions not measured but similar to 1999/00
Table 12.4. Dimensions of the six terrace unit plots (NBEPs) and details of treatment and
runoff during the three consecutive measurement periods. Listed are average cover provided
by vegetation (C
v
) or surface mulch (C
s
), runoff depth (Q
tot
) and coefficient (rc) for the entire
measurement period.
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
232
12.3.2. Rainfall and runoff measurements
Rainfall rate was measured using a custom-built tipping bucket-logger system which
recorded the time of each tip to the nearest second. Rainfall intensity was measured using
five instruments at different locations in the catchment between 17 November 1998 and 8
February 2000, while daily rainfall totals continued to be measured afterwards. Because
of occasional malfunctioning of individual instruments, data collected by three different
tipping bucket units were compiled. Rainfall intensity data were used both in their
original form (i.e. with variable time steps corresponding to equal rainfall amounts of
0.07-0.10 mm) and resampled into time intervals of 2 to 3-minutes. If there were more
than four hours between two consecutive tips, these were taken to represent different
storm events. The first tip for each individual storm was assumed to represent rain falling
at an intensity equal to that causing the second tip. Further details on calibration of the
instrument and (re-) sampling procedures can be found in Chapter 11 (Appendix 11.B).
Runoff from the ABEPs and NBEPs was measured daily for almost two-and-a-half
year, for the periods of 3 December 1998 to 11 April 1999 and 23 October 1999 to 6
February (NBEPs) or 6 April 2000 (ABEPs) (see also Tables 12.2, 12.3 and 12.4).
Runoff from the ABEPs was collected in (halved) oil drums placed below a gutter
extending along the downslope end of the plot; the volume was measured using a graded
cylinder (for volumes less than two litres) or a graded bucket (for larger volumes).
Runoff from the NBEPs was first collected in a rectangular basin with an effective
capacity of 0.20-0.27 m
3
. When runoff volume exceeded the capacity of the basin a
divider system directed one-fourth or one-fifth of the flow into a drum with a capacity of
0.13-0.18 m
3
. Whenever required one-seventh of the excess flow was further diverted
into a second drum of similar capacity. Height-volume rating curves were established for
all basins and drums to allow runoff volume to be determined from measured water level.
In all cases runoff volumes were divided by projected plot area to derive runoff depths.
In addition, runoff rates were measured during selected periods using a pressure
transducer-logger system developed at the Vrije Universiteit Amsterdam. The system
measures water level at pre-set time intervals with an accuracy typically better than 0.1
mm. Data were collected at 2 or 3-minute intervals in the collecting basins of two ABEPs
on a terrace bed and riser, respectively, from 18 January until 6 March 1999. Rating
curves were established to calculate runoff rates from water level changes. Water levels
were also measured in the stilling basins of NBEPs, at 3-minute intervals, between 6
December 1998 and 12 April 1999 using one logger unit for each terrace pair. Roughly
every two weeks the systems were switched between terraces.
12.4. Runoff modelling
12.4.1. The spatially variable infiltration (SVI) model
In the spatially variable infiltration model (SVI; Yu et al., 1997c) rainfall excess (Q
in mm h
-1
) is calculated from rainfall intensity (R) assuming an exponential spatial
distribution of maximum infiltration rates that is represented by the average maximum
infiltration rate (I
m
in mm h
-1
):
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
233
( )
1
]
1

,
_


m
m
I
R
I R R I R Q exp 1 [12.1]
Theoretically I
m
is reached when the entire area under consideration generates runoff.
Alternatively, it can be stated that 37% of the area has a maximum infiltration capacity in
excess of I
m
(Yu et al., 1997c). It is implicitly assumed in Eq. [12.1] that infiltration-
excess water from one area does not run onto an adjacent area with a higher infiltration
capacity. To simulate the higher infiltration rates occurring at the beginning of storms an
initial amount of infiltration (F
o
in mm) was introduced, taking place before runoff is
generated according to Eq. [12.1].
Yu et al. (1997d) suggested the use of a simple routing scheme, as an approximation
of the kinematic wave model, to account for the fact that generated runoff needs to travel
to the point at which it is measured. It reads:
( )
i i i
Q Q Q +

1
1
[12.2]
where Q
i
and Q
i-1
(both in mm h
-1
) are modelled runoff rates for time intervals i and i-1,
respectively, Q
i


(in mm h
-1
) the modelled rate of infiltration-excess rainfall for interval i,
and a dimensionless routing parameter that has a value between 0 and 1 (cf. Yu et al.,
1997c, 1997d, 1999). The parameter can be related to time lag by:
t K

1
or
t K
K
+
[12.3]
where K (in h) is the time lag and t (in h) the duration of the time interval used in
modelling. The modelled degree of runoff rates attenuation depends on the contrast
between the time lag (K) and the rainfall resampling interval (t; cf. Yu, 1999).
Runoff rates simulated using observed rainfall rates can be used to calculate the
effective runoff rate (Q
e
in mm h
-1
) used by the GUEST erosion model to compute the
sediment concentration in runoff for the event (Rose, 1993; see Chapter 13). It is defined
as:
5 . 2
4 . 1
1
1
]
1


tot
e
Q
Q
Q [12.4]
where Q (in mm h
-1
) denotes instantaneous runoff rate and Q
tot
(in mm) the runoff depth
associated with the considered event.
12.4.2. Event-based model equations
To avoid the need for high time-resolution rainfall intensity data, an exponential
depth-rainfall intensity distribution was assumed for individual storms in Chapter 11. It is
characterised by storm depth P (in mm) and depth-averaged rainfall intensity (in mm
h
-1
). The latter is calculated as:
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
234
( ) ( )
P
t R
P
PR
R
n
i
i
n
i
i



1
2
1

[12.5]
The time intervals (t) do not have to be of equal duration in order to calculate . On the
basis of the exponential rainfall distribution, an expression was derived that expresses
storm runoff depth (Q
tot
) as a function of P, and I
m
(cf. Chapter 11):

'

1
]
1

,
_

+ 0 , 1 ln 1 max
I
m
m
tot
S
I
R
R
I
P Q

[12.6]
where S
I
(in mm) is initial excess infiltration. The definition of S
I
necessarily was
conceptually somewhat different from the initial amount of infiltration (F
0
) used the
original SVI model; it was assumed that at the beginning of a storm, the apparent rate of
runoff generation (as calculated by Eq. [12.1] or the first term of Eq. [12.6]) is decreased
because an additional (rather than initial) amount of infiltration has to take place before
runoff actually occurs. The symbol S was used to indicate that this term behaves like a
storage. It may include storage in surface depressions and on the vegetation canopy and
an effective matric potential that may need to be overcome. In the current study
interception losses were neglected.
To derive event-based expressions for effective runoff rate two additional
assumptions had to be made, viz.: (i) initial additional infiltration (S
I
) does not affect the
effective runoff rate; and (ii) no runoff peak attenuation occurs. The first assumption
results in an over-estimation of Q
e
if higher than average rainfall rates occur at the
beginning of the storm, and (vice versa) an under-estimation if early rainfall intensities
are lower than average. Neglecting the effect of runoff routing may result in an over-
estimate. As such, the effective computed. The resulting effective runoff rate may be
denoted Q
e
(in mm h
-1
) to distinguish it from Q
e
proper and is calculated as (Chapter
11):
( ) ( ) [ ]
5 . 2
0
4 . 1
4 . 1
5 . 2
0
4 . 1
exp 1 exp
1
1
1
]
1


,
_


1
1
1
1
1
]
1

du u u u
R
I
u Q R
PI
Q
dt Q
Q
m
tot
m
tot
e
[12.7]
where u=R/I
m
. There is no analytical solution for the integral in the second part of Eq.
[12.7], but it is easily solved with mathematical software (cf. Chapter 11). It was tested
whether the errors caused by the cited two assumptions are systematic, and in that case
may be corrected with an empirical factor (f

), so that actual effective runoff rate Q


e
can
be simulated as:
e e
Q f Q

[12.8]
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
235
12.4.3. Runoff from contrasting contributing areas
The bench terraces examined are comprised of three segments with contrasting
infiltration characteristics (cf. Fig. 12.1). In principle, the spatially averaged theoretical
runoff response Q
~
(in mm h
-1
) from the entire terrace can therefore be described by
weighing the infiltration characteristics of each segment by its corresponding area (cf.
Eq. [12.1]):

,
_

1
1
]
1


n
i
I
R
i m i A
i m
e I R F Q
1
, ,
,
1
~
[12.9]
where F
A,i
is the relative area and I
m,i
(in mm h
-1
) the maximum average infiltration rate
associated with segment i. Overall total runoff depth ( Q
~
tot
in mm) is given by (cf. Eq.
[12.6]):

'

,
_

1
1
]
1

,
_

+
n
i
i I
i m
i m
i A tot
S
I
R
R
I
P F Q
1
,
,
,
,
0 , 1 ln 1 max
~

[12.10]
where S
I,i
(in mm) is the initial additional infiltration depth on segment i. Furthermore, if
the resulting distribution of maximum infiltration rates over the entire terrace would
remain exponential the corresponding spatially-averaged value
m
(in mm h
-1
) is given
by:
( )

n
i
i m A m
I F I
1
~
[12.11]
However, in many cases the distribution of infiltration rates of the larger area will no
longer conform to an exponential distribution. In such cases, using Eqs. [12.1] or [12.6]
to predict instantaneous runoff rates or event runoff totals using
m
according to Eq.
[12.11] will result in rainfall-infiltration relationships that are different from those
calculated with Eqs. [12.9] and [12.10]. This is illustrated in Fig. 12.2a and b where
rainfall-infiltration relationships resulting from Eqs. [12.9] and [12.10] are shown for
combinations of (relative) areas and I
m
for the three segments of a representative bench
terrace in this study (cf. Section 12.5). Instead of using an I
m
value calculated by
Eq.[12.11], an apparent value of
m
may be chosen such that it best fits infiltration rates
over the range of observed rainfall intensities (cf. Fig. 12.2a and b). Total runoff depth
from the composite area is hence described by (cf. Eq. [12.6]):
+

'

1
1
]
1

,
_

+ 0 ,
~
~
1 ln
~
1 max
~
I
m
m
tot
S
I
R
R
I
P Q

[12.12]
where S
~
I
(in mm) denotes an effective spatially-averaged storage term and is the
error that is introduced by the approximation. In the example illustrated in Fig. 12.2b this
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
236
error is less than 7% for values between 5 and 150 mm h
-1
. It should be noted that
m
can always be chosen such that Eqs. [12.6] and [12.12] produce identical results but that
the actual value varies with .
Another problem is encountered when calculating S
~
I
as the area-weighed value of S
I
in situations were not all contributing zones produce runoff in excess of their respective
S
I
values. Moreover, in the context of the current study, runoff from the terrace riser and
0
50
100
0 25 50 75 100 125 150
R (mm h
-1
)
I
(
R
)

(
m
m

h
-
1
)
terrace bed
terrace riser
terrace toe drain
using Eq. [12.11]
Best fit
overall cumulative response (a)
0.00
0.10
0.20
0.30
0.40
0 25 50 75 100 125 150
R (mm h
-1
)
r
c

terrace toe drain
terrace riser
terrace bed
using Eq. [12.11]
Best fitting Eq. [12.12]
Overall cumulative response (Eq. [12.10])
(b)
Fig. 12.2. The combined response of the three terrace sections expressed as the relationship
between (a) rainfall intensity (R) and infiltration rate (I(R); Eq. [12.9]) and (b) depth-averaged
rainfall intensity () and runoff coefficient (rc; Eq. [12.10]). Cumulative contributions by the
three sections are shown by dashed lines, whereas solid lines indicate the relationships for the
combined response using area-weighed average values of
m
. Relative area values of 5, 16 and
79% and I
m
values of 20, 70 and 200 mm h
-1
were used for the toe drain, riser and bed,
respectively.
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
237
bed enters the central toe drain before eventually leaving the terrace. Part of this runoff
may therefore help to fill the S
I
of the toe drain. For these reasons S
~
I
was not readily
calculated from the values of the respective contributing areas and it was optimised
instead. Finally, because Eq. [12.12] is conceptually and mathematically identical to Eq.
[12.6], it follows that Eq. [12.7] can be used to calculate the effective runoff rate Q
e

(using
m
and S
~
I
instead of I
m
and S
I
). An approach similar to Eq. [12.8] may
subsequently be followed to estimate actual Q
e
values from this.
12.4.4. Model validation and testing
To validate the SVI model for the simulation of instantaneous runoff rates and event
runoff totals in the studied environment, the measured storm hydrographs for both
ABEPs and NBEPs were simulated using the original model equations and rainfall
intensity measurements. Runoff hydrographs were measured on a terrace bed (code B-I-1
in Table 12.2) and a riser plot (R-I-4 in Table 12.3) between 19 January and 6 March
1999, and on terrace units A to F between 15 February and 30 March 1999. The number
of runoff hydrographs was limited by the duration of the runoff event; measured runoff
hydrographs that had less than three observations (i.e. 6 to 9 minutes) were not used.
Moreover, large storms occasionally produced enough runoff to fill the first collecting
basin of terrace units, causing it to overflow into the second compartment. Because water
level was not recorded automatically in the second or third compartment this produced
incomplete hydrographs and these were discarded. The values of I
m
and S
I
were
optimised by fitting storm-based equations to measured daily runoff depths (see below)
by Levenberg-Marquardt minimisation of the residual variance. The time lag (K) value
that fitted the measured runoff rates best was optimised in a similar way using the
measured runoff hydrographs. Subsequently, the effective runoff rate (Q
e
) was computed
in four different ways, viz.: (i) from observed runoff rates (Q
e
-obs), (ii) from runoff rates
simulated using an optimised value for time lag (Q
e
-K), (iii) from runoff rate simulated
without using a routing procedure (Q
e
-sim), and using Eq. [12.7] (Q
e
). The influence of
runoff routing was investigated by comparing Q
e
-K and Q
e
-sim values.
Furthermore, the event-based equations in Section 12.4.2 and 12.4.3 were used to
model runoff from the ABEPs and NBEPs during the periods that rainfall intensity was
measured, i.e. from December 1998 to 10 April 1999 and from 23 October 1999 to 7
February 2001. Firstly, depth-averaged rainfall intensity () was calculated for all
available storms using Eq. [12.5] using the original (not resampled) data. The resulting
time series for P and were used in combination with Eqs. [12.6] and [12.12] to estimate
event runoff depths. The values of I
m
and S
I
for each plot-period combination were again
found the Levenberg-Marquardt method. It was observed that runoff from the terrace bed
plots decreased dramatically if the maize harvest residue was left on the soil. Therefore a
distinction was made between pre- and post-harvest parameters were appropriate. In all
cases model efficiency (ME) sensu Nash and Sutcliffe (1970) was calculated as an
indicator of model performance.
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
238
12.5. Results
12.5.1. Rainfall and runoff
Measured seasonal or annual rainfall totals and corresponding surface runoff
coefficients are listed in Table 12.2 for the terrace risers and beds (ABEPs) and in Table
12.3 for the terrace units (NBEPs). Rainfall was measured from 17 November 1998 until
6 April 2001. As shown in Fig. 12.3a and b, the first two rainy seasons received more or
less average amounts of rain but the third season exhibited a more irregular pattern and
was rather dry on the whole. Simultaneous rainfall intensity and runoff data were
available for the period 17 November 1998-8 February 2000, except for the period 12
April-10 May 1999. Depth-averaged rainfall intensity () was calculated for 121 storms
using Eq. [12.15]. The resulting values are plotted against corresponding rainfall depths
(P) in Fig. 12.4, showing a reasonable correlation between the two (r
2
=0.62).
0
20
40
60
80
100
N D J F M A M J J A S O N D J F M A M J J A S O N D J F M A
D
a
i
l
y

r
a
i
n
f
a
l
l

(
m
m
)
1998 1999 2000 2001
(a)
0
100
200
300
400
500
600
N D J F M A M J J A S O N D J F M A M J J A S O N D J F M A
M
o
n
t
h
l
y

r
a
i
n
f
a
l
l

(
m
m
)
1998 1999 2000 2001
(b)
Fig. 12.3. (a) Daily and (b) monthly rainfall during the measurement period, compared with
average monthly rainfall between 1994 and 2001.
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
239
During the 147 days that runoff from the ABEPs was measured in the 1998/99 rainy
season (127 days for four out of eleven plots) a total of 1775 mm (or 1418 mm) of
rainfall was recorded, of which between 0.1 and 23% left the plots as runoff (Tables 12.2
and 12.3). The 1999/2000 measurement period comprised an entire agronomic year, with
a total rainfall of 2871 mm, of which 6-25% was recorded as runoff. The 2000/01 period
covered 156 days with a total rainfall of 1754 mm, of which 5-31% ran off. Despite the
difference in slope gradient and subsoil exposure, runoff coefficients associated with bare
terrace riser sections (12-28%) were not higher than those for bare terrace beds (16-
26%). However, runoff from terrace beds with more than 5% vegetation cover was
generally less than that from bare plots, especially in the case of mixed cropping (0.1-
24%; Table 12.2). Nevertheless, plots that were planted with cassava only showed
consistently high runoff coefficients of 13-34%, with little difference between weeded
and non-weeded plots (Table 12.2). Similarly, at 0.6-13% runoff from terrace risers that
had some form of deliberately planted cover (grass, herb) was generally less than half
that from bare risers, whereas unweeded risers with variable degrees of cover exhibited
intermediate values (8-20%; Table 12.3).
Runoff from the six terrace units was measured during approximately the same
periods as for the ABEPs (Table 12.4). In the 1998/99 period (130 days), 7-26% of the
1697 mm rainfall left the terraces as surface runoff. In 1999/2000 (2567 mm of rain in
365 days) this percentage was 6-15%, whereas in 2000/01 (1367 mm in 104 days) runoff
coefficients for four remaining terraces ranged between 5 and 23%. Runoff from the two
terraces on the steeper part of the slope (CD) was consistently higher than from the other
terraces (rc=5-23% versus 13-26%; Table 12.3).
12.5.2. Runoff hydrograph simulation
A total of 13 suitable runoff hydrographs was available for the terrace bed
representing storms of 15-56 mm with values of 14-79 mm h
-1
. Of these 13 events, five
R = 1.50 P
0.87
r
2
= 0.62
0.01
0.1
1
10
100
1000
0.1 1 10 100
P (mm)
R

(
m
m

h
-
1
)
-
)
Fig. 12.4. The relationship between storm depth (P) and depth-averaged rainfall intensity ().
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
240
storms occurred before the harvesting of the maize, while the remaining storms
represented conditions with cassava only and a surface mulch of maize stalks (cf. Table
12.5). Measurements of runoff from the terrace riser also produced 13 events, having
rainfall depths of 10-59 mm and with values of 14-79 mm h
-1
. Model results for the
two ABEPs are listed in Table 12.5 and shown in Fig. 12.5.
ABEPs Observed Modelled Obs Modelled
Date P Q
tot Q
tot
ME Q
e
-obs Q
e
-K Q
e
-sim Q
e

mm mm h
-1
mm mm mm h
-1
mm h
-1
mm h
-1
mm h
-1
Bed before harvesting (S
I
=0.85 mm, I
m
=72 mm h
-1
, K=0.28 min)
26-Jan-99 34 14 1.7 1.5 0.77 2.7 2.1 2.2 3.8
27-Jan-99 41 34 6.7 6.9 0.83 15.4 13.6 14.3 16.3
29-Jan-99 45 40 8.4 7.6 0.49 17.3 12.9 13.5 20.9
30-Jan-99 25 16 1.3 1.7 0.80 2.5 2.3 2.4 4.7
31-Jan-99 15 31 1.7 0.8 0.21 4.8 1.9 2.1 14.1
Bed after harvesting (S
I
=0.24 mm, I
m
=443 mm h
-1
, K=0.28 min)
17-Feb-99 56 79 3.1 3.8 0.55 9.3 11.9 12.2 19.9
21-Feb-99 18 18 0.1 - n.a. 1.2 - - 1.2
22-Feb-99 26 15 0.2 0.2 0.55 0.9 0.5 0.5 0.9
24-Feb-99 22 26 0.4 0.4 0.60 1.7 1.4 1.5 2.5
26-Feb-99 20 14 0.1 - n.a. 1.5 - - 0.8
27-Feb-99 22 57 1.9 1.0 0.47 7.2 4.2 4.6 11.0
2-Mar-99 29 53 1.2 1.1 0.66 5.0 3.5 3.6 9.7
3-Mar-99 44 32 5.6 1.4 0.07 18.2 3.8 4.0 3.8
Riser (S
I
=0 mm, I
m
=53 mm h
-1
, K=0.91 min)
26-Jan-99 33 14 1.9 2.7 0.57 3.0 2.0 2.3 4.6
27-Jan-99 40 34 8.9 9.0 0.74 18.7 15.2 17.5 18.8
29-Jan-99 44 40 14.9 10.1 0.71 26.6 14.6 16.5 24.3
30-Jan-99 24 16 4.8 2.7 0.38 6.0 3.6 3.9 5.8
31-Jan-99 16 31 3.8 1.9 0.59 12.4 5.0 6.9 16.4
14-Feb-99 10 18 1.3 1.2 -0.07 4.2 6.9 8.8 7.0
17-Feb-99 59 79 20.2 21.8 0.97 49.6 49.8 53.8 58.8
21-Feb-99 19 18 4.2 1.3 0.20 10.1 2.0 2.2 6.9
22-Feb-99 26 15 3.7 2.6 0.27 3.2 2.8 3.3 5.5
24-Feb-99 22 26 3.5 3.7 0.30 14.5 9.3 10.3 12.9
27-Feb-99 23 57 10.1 7.7 0.69 32.0 21.4 24.6 38.6
2-Mar-99 36 53 12.2 11.6 0.88 28.3 27.9 30.0 35.4
3-Mar-99 39 32 11.3 7.6 0.85 11.8 15.1 16.5 17.6
Table 12.5. Results of model validation for a terrace bed and riser section plot (ABEPs B-I-1
and R-I-4; cf. Tables 12.2 and 12.3). Listed are rainfall depth (P), depth-averaged rainfall
intensity (), observed and modelled runoff depth (Q
tot
) with associated model efficiency
(ME) and effective rates of runoff calculated from runoff rates observed (Q
e
-obs), simulated
using rainfall intensity data with (Q
e
-K) and without (Q
e
-sim) a routing procedure,
respectively, and rates calculated using Eq. [12.7] (Q
e
).
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
241
0
1
2
3
4
14:30 15:00 15:30 16:00 16:30 17:00
Q
'

(
m
m

h
-
1
)
0
25
50
75
100
125
150
R

(
m
m

h
-
1
)
(a)
0
20
40
60
80
100
120
140
16:30 17:00 17:30 18:00 18:30 19:00 19:30
Q
'

(
m
m

h
-
1
)
0
100
200
300
400
R

(
m
m

h
-
1
)
(b)
0
5
10
15
20
12:00 12:53 13:47 14:41
Q
'

(
m
m

h
-
1
)
0
25
50
75
100
125
150
R

(
m
m

h
-
1
)
(c)
Fig. 12.5. Observed (open dots) and modelled (solid line) runoff (Q) hydrographs and rainfall
pattern (R, inverse vertical axis) for selected events on terrace section plots (ABEPs): (a)
approximating the average model performance (ME=0.55; terrace bed, 22 February 1999); (b)
the best model performance (ME=0.88; terrace riser, 2 March 1999); and (c) the worst model
performance (ME<0; terrace riser, 14 February 1999). Note that scales vary.
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
242
In general, the simulated runoff hydrographs were fairly close to those observed
(average model efficiency 0.59; Table 12.5). For the bed plot the model simulated no
runoff for two events whereas some was measured in reality. The hydrographs
corresponding to the closest-to-average (ME=0.55), as well as the best (ME=0.97) and
worst model efficiency (ME<0) are shown in Fig. 12.5. The most important reason for
the poor performance of the model in Fig. 12.5c is that the first runoff peak was not
simulated because the initial infiltration term (optimised at S
I
=0.85 mm for all storms)
was not filled yet at this time. The fit in Fig 12.5a was also improved by decreasing S
I
somewhat and inspection of the other hydrographs suggested that these, too, could often
Terrace unit Observed Modelled Obs Modelled
Date
P Q
~
tot Q
~
tot ME Q
~
e-
-obs Q
~
e
-K Q
~
e
-sim Q
~
e

mm mm h
-1
mm
mm mm h
-1
mm h
-1
mm h
-1
mm h
-1
Terrace A: S
~
I
=0.38 mm,
m
=228 mm h
-1
, K=1.23 min
21-Feb-99 18 10 0.4 0.2 0.29 2.3 0.4 0.5 0.7
22-Feb-99 25 13 0.5 0.4 0.72 0.9 0.4 0.5 1.2
27-Feb-99 20 51 0.8 1.6 0.68 8.4 5.9 8.3 15.2
Terrace B: S
~
I
=0.0 mm,
m
=304 mm h
-1
, K=3.02 min
24-Mar-99 25 38 0.5 0.7 0.96 2.0 1.7 2.7 7.2
26-Mar-99 17 22 0.3 0.6 0.46 1.8 0.9 1.5 2.6
Terrace C: S
~
I
=1.73 mm,
m
=56 mm h
-1
, K=3.12 min
15-Mar-99 36 18 1.9 3.1 0.84 6.5 3.8 7.1 6.9
26-Mar-99 17 22 0.7 1.1 0.16 2.3 2.5 3.1 9.5
Terrace D: S
~
I
=1.52 mm,
m
=27 mm h
-1
, K=1.48 min
21-Feb-99 19 10 1.5 1.6 0.79 4.7 3.3 3.8 4.2
24-Feb-99 22 39 3.3 7.3 0.12 15.5 19.8 23.4 28.8
27-Feb-99 23 60 5.7 11.3 0.38 57.7 31.2 40.3 48.0
03-Mar-99 39 31 7.5 11.2 0.74 34.8 20.9 23.8 21.6
Terrace E: S
~
I
S
I
=0.08 mm,
m
=299 mm h
-1
, K=1.62 min
21-Feb-99 19 10 0.3 0.4 0.39 2.4 0.2 0.5 0.6
22-Feb-99 26 13 0.5 0.6 0.53 0.9 0.6 0.7 1.0
24-Feb-99 22 39 1.3 3.1 0.84 4.7 1.4 3.6 7.6
27-Feb-99 23 60 3.0 6.0 0.83 9.1 2.1 8.1 16.5
03-Mar-99 39 31 3.1 3.7 0.75 8.0 1.9 4.4 5.0
24-Mar-99 23 34 0.3 4.1 -2.03 2.5 1.1 6.2 5.9
15-Mar-99 36 18 0.7 1.1 0.69 2.8 1.0 1.6 1.8
26-Mar-99 17 22 0.4 0.9 0.48 1.1 0.5 1.2 2.6
Table 12.6. Results of model validation for terrace unit plots A-E in 1998/99 (cf. Table 12.4).
Listed are rainfall depth (P), depth-averaged rainfall intensity (),observed and modelled
runoff depth (
Q
~
tot
) with associated model efficiency (ME) and effective rates of runoff
calculated from runoff rates observed (
Q
~
e
-obs) simulated using rainfall intensity data, with
(
Q
~
e
-K) and without (
Q
~
e
-sim) a routing procedure, and rates calculated using Eq. [12.7] (
Q
~
e
).
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
243
be much improved by small increases or decreases in S
I
(not shown). The values of lag
time (K) that produced the best fits were 0.28 and 0.91 minutes, for the terrace bed and
riser, respectively (Table 12.4).
The hydrograph simulation results for terrace units are summarised in Table 12.6. A
total of 19 runoff events was available for analysis, with rainfall amounts of 17-39 mm
and values of 10-51 mm h
-1
. All runoff storms occurred after the harvesting of maize
and rice from the terraces, leaving only the cassava crop with the maize harvest residue
on the terrace bed and in the toe drain. As observed earlier for the terrace riser and bed
plots, simulated runoff rates generally corresponded fairly well with observations, with
an identical average model efficiency (ME) of 0.59. The runoff hydrographs
corresponding to the best (ME=0.96), worst (ME<0) and closest-to-average (ME=0.53)
performances are shown in Fig. 12.6. Again, an important source of discrepancy was
associated with the magnitude of S
~
I
. Model fits in Fig. 12.6a and particularly in Fig.
12.6c were greatly improved when a slightly greater value of S
~
I
was used. Small
changes in S
~
I
also improved model performance for a number of other events (not
shown). The value of time lag (K) producing the best agreement between simulated and
observed runoff rates varied from 1.23 minutes for terrace A to 3.12 minutes for terrace
C (Table 12.6).
12.5.3 Modelling effective runoff rate
The values for effective runoff rate derived in four different ways are summarised
for the terrace riser and bed plots in Table 12.5 and for the terrace units in Table 12.6.
The Q
e
values computed using Eq. [12.7] overall appear higher than those calculated
from simulated runoff rates. Perhaps surprisingly, omitting the routing procedure did not
result in an appreciable difference between values of Q
e
-K and Q
e
-sim, respectively
(Table 12.5). Rather, two other phenomena are responsible for the difference between
simulated Q
e
and Q
e
values. Firstly, it was already suggested that initial infiltration (S
I
)
can reduce peak runoff, particularly if the highest rainfall rates occur at the start of the
storm (an example of this is discussed further on). Secondly, rainfall intensity data were
resampled into 2 or 3-minute intervals before simulating the runoff hydrographs. This
already decreased peaks rainfall intensities and so lead to lower runoff peaks even
without a routing procedure. On the other hand, Eq [12.7] used a value of that was
calculated from the original (not resampled) tipping-bucket data to calculate Q
e
(cf.
Section 12.3). This effect of resampling also explains why overall total runoff amounts
are on average 15% lower than those measured (cf. Chapter 11). As a result Q
e
values in
many cases agree similarly well or even better to values calculated from observed
hydrographs (Q
e
-obs) than do those calculated from simulated hydrographs (Fig. 12.7a,
b). For example, the coefficient of determination between Q
e
and Q
e
-obs for the terrace
riser was r
2
=0.96 versus r
2
=0.92 between Q
e
-K and Q
e
-obs.
Results for the terrace units largely confirm those obtained for terrace section plots,
although in this case the omission of time lag (K) results in Q
~
e
values that are 28%
higher on average. However, these simulated hydrograph-based values (Q
~
e
-K) are still
substantially lower than those observed. Values calculated with Eq. [12.7] correspond
very well to those based on observed runoff rates (Fig. 12.7d). Again an important reason
for the under-estimation of simulated runoff peaks is related to rainfall data resampling.
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
244
0
1
2
3
4
14:00 14:30 15:00 15:30 16:00 16:30 17:00
Q
'

(
m
m

h
-
1
)
0
20
40
60
80
100
R

(
m
m

h
-
1
)
(a)
0
2
4
6
17:00 17:30 18:00 18:30 19:00
Q
'

(
m
m

h
-
1
)
0
50
100
150
200
250
300
R

(
m
m

h
-
1
)
(b)
0
2
4
6
8
10
12
14
17:00 17:30 18:00 18:30 19:00
Q
'

(
m
m

h
-
1
)
0
50
100
150
200
250
300
R

(
m
m

h
-
1
)
(c)
Fig. 12.6. Observed (dashed line with dots) and modelled (solid line) runoff hydrographs and
rainfall pattern for selected events on terrace units (NBEPs): (a) approximately average model
performance (ME=0.53; terrace E, 22 February 1999; (b) the best model performance
(ME=0.96; terrace B, 24 March 1999; and (c) the worst model performance (ME<0; terrace E,
24 March 1999).
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
245
Use of Eq. [12.7] to some extent compensates for this under-estimation. However, the
longer travel time associated with the terrace units when compared with the riser and bed
plots (1.2-3.1 min versus 0.28-0.91 min, respectively) may have resulted in an over-
compensation in some cases. Overall, Q
~
e
values were somewhat better correlated with
simulated ( Q
~
e
-K) than with observed values ( Q
~
e
-obs), at r
2
=0.91 and r
2
=0.82,
respectively. Nevertheless, the performance of Eq. [12.7] is still considered good and
suggests that, in the studied environment, calculated Q
~

e
values may not need to be
corrected for the effects of runoff peak attenuation.
12.5.4. Modelling event runoff from terrace risers and beds
The storm-based infiltration model (Eq. [12.6]) was used to simulate runoff from a
total of 31 runoff plots comprising sections of terrace risers or beds. The length of the
various modelling periods and the associated rainfall amounts and depth-averaged
intensity are listed in Table 12.7. Simulation periods varied between 96 and 117 days,
during which 1207-1390 mm of rainfall was recorded with associated overall values of
32.1-36.2 mm h
-1
. Because a distinct change in runoff characteristics was observed on the
bed: Q
e
-K= 0.74 Q'
e
r
2
= 0.92
riser: Q
e
= 0.56 Q'
e
r
2
= 0.76
0.1
1
10
100
0.1 1 10 100
Q
e
' (mm h
-1
)
Q
e
-
K

(
m
m

h
-
1
)
(a)
(c) terrace units:
Q
e
-K = 0.64 Q
e
'
r
2
= 0.91
0.1
1
10
100
0.1 1 10 100
Q'
e
(mm h
-1
)
Q
e
-
K

(
m
m

h
-
1
)
bed: Q
e
-obs = 0.86 Q
e
'
r
2
= 0.96
riser: Q
e-obs= 0.69 Q
e
'
r
2
= 0.34
0.1
1
10
100
0.1 1 10 100
Q
e
' (mm h
-1
)
Q
e
-
o
b
s

(
m
m

h
-
1
)
(b)
(d) terrace units
Q
e
= 1.00 Q'
e

r
2
= 0.82
0.1
1
10
100
0.1 1 10 100
Q'
e
(mm h
-1
)
Q
e
-
o
b
s

(
m
m

h
-
1
)
Fig. 12.7. Maximum effective runoff rate for the terrace bed and riser resulting from Eq.
[12.8] (Q
e
) compared with values calculated from (a) simulated (Q
e
-K) and (b) observed (Q
e
-
obs) hydrographs. Closed dots refer to terrace bed, triangles to terrace riser. (c-d) The same,
but for terrace units, compared with values calculated from (c) simulated ( Q
~
e
-K) and (d)
observed ( Q
~
e
-obs) hydrographs.
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
246
Treatment Location, code Observed Modelled Pre-/post-harvest
N Cum
Q
tot
rc Cum
Q
tot
Cum
diff.
ME S
I
I
m
mm % mm % mm mm h
-1
1998/99: P=1390 (1207) mm
a
Bed sections
cassava-maize-peanut AB B -I-1 39 83 5 112 34% 0.45 0.9/0.2 72/443
cassava-maize-rice CD B -I-8 47 411 23 370 -10% 0.87 4.8/0.3 18/35
cassava-maize EF B -I-5 25 75 4 55 -26% 0.76 13.0/4.1 103/62
cassava-maize
a
EF B -I-12 4 1 0 1 0% n.a.
c
7.8/7.8 233/224
Riser sections
unweeded CD R -I-7 40 213 12 196 -8% 0.67 1.7 64
unweeded EF R -I-4 53 355 20 333 -6% 0.85 0.4 53
unweeded
a
EF R -I-13 39 279 20 235 -16% 0.75 0 65
unweeded AB R -I-3 37 134 8 125 -6% 0.67 2.6 82
unweeded / shrub
b
CD R -I-9 40 239 13 225 -6% 0.78 0.3/3.9 86/35
grassed
a
CD R -I-11 28 56 4 54 -4% 0.68 0.3 298
grassed
a
AB R -I-10 3 9 1 9 0% n.a.
c
3.9 250
1999/00: P= 1390 (1281) mm
Bed sections
bare EF B -II-3 44 205 15 193 -6% 0.74 4.1 35
bare EF B -II-6 37 180 14 163 -10% 0.67 8.2 25
ginger EF B -II-4 41 165 12 154 -6% 0.76 4.2 44
ginger EF B -II-5 39 134 10 123 -8% 0.75 6.7 39
cassava (weeded) EF B -II-9 35 178 13 170 -5% 0.79 5.9 32
cassava (weeded) EF B -II-10 51 226 17 208 -8% 0.84 4.9 29
cassava-maize-rice EF B -II-1 38 144 11 140 -2% 0.65 1.4 78
cassava-maize-rice EF B -II-2 40 110 8 107 -3% 0.70 0.9 115
cassava-maize-peanut EF B -II-8 33 59 4 58 -3% 0.66 0.7 223
cassava-maize-peanut EF B -II-7 44 80 6 78 -3% 0.83 0.1 213
Riser sections
bare EF R -II-2 54 282 20 284 1% 0.75 0.03 52
bare EF R -II-3 52 291 21 302 4% 0.76 - 50
bare CD R -II-6 66 167 12 176 5% 0.80 - 105
grass planted EF R -II-1 56 120 9 144 20% 0.67 - 157
grassed CD R -II-8 24 64 5 64 1% 0.81 2.4 128
shrub new planted CD R -II-5 74 141 10 161 14% 0.71 - 130
shrub sparse CD R -II-7 28 99 7 104 6% 0.66 3.0 80
shrub dense EF R -II-4 74 158 11 186 18% 0.66 - 113
a
shorter period with 1207 mm of rain;
b
herb planted on 9 February 1999;
c
too few runoff events measurements
Table 12.7. Results of application of the event-based infiltration equation for terrace bed and
riser plots. Listed are the observed number of runoff events (N), runoff coefficient (rc),
cumulative observed and modelled runoff amounts (Q
tot
), the relative difference between
cumulative amounts, model efficiency (ME) and optimised values of the model parameters
average maximum infiltration rate (I
m
) and initial additional infiltration (S
i
), for pre- and post-
harvest phases where applicable. Model efficiencies associated with plots used for hydrograph
simulation are printed bold.
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
247
terrace beds after harvesting of maize and rice the 1998/99 bed plot data were divided
into pre- and post harvest periods which separate optimised S
I
and I
m
values. The same
procedure was followed for terrace riser plot R-I-9, which was planted with shrub
cuttings on 9 February 1999 (Table 12.3). No crops were harvested at the end of the
0
5
10
15
20
25
30
0 5 10 15 20 25 30
Q
tot
observed (mm)
Q
t
o
t

m
o
d
e
l
l
e
d

(
m
m
)
(a)
-
50
100
150
200
250
- 50 100 150 200 250
Cumulative observed Q
tot
(mm)
C
u
m
.

m
o
d
e
l
l
e
d

Q
t
o
t

(
m
m
)
(d)
0
10
20
30
0 5 10 15 20 25 30
Q
tot
observed (mm)
Q
t
o
t

m
o
d
e
l
l
e
d

(
m
m
)
(b)
-
50
100
150
- 50 100 150
Cumulative observed Q
tot
(mm)
C
u
m
.

m
o
d
e
l
l
e
d

Q
t
o
t

(
m
m
)
harvest
(e)
0
10
20
30
0 5 10 15 20 25 30
Q
tot
observed (mm)
Q
t
o
t

m
o
d
e
l
l
e
d

(
m
m
)
(c)
-
100
200
300
400
0 100 200 300 400
Cumulative observed Q
tot
(mm)
C
u
m
.

m
o
d
e
l
l
e
d

Q
t
o
t

(
m
m
)
harvest
(f)
Fig. 12.8. Observed and modelled event runoff depth totals from terrace section plots
(ABEPs) illustrating (a) the average model performance (ME=0.75, bed plot B-II-3), (b) the
worst model performance (ME=0.45, bed plot B-I-1) and (c) the best model performance
(ME=0.87; bed plot B-I-8). (d-f) Cumulative runoff for the same plots; dashed line in (e)
shows agreement when first events are neglected (see text). Times of maize harvest indicated
where applicable.
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
248
1999/2000 season runoff (8 February 2000). Table 12.7 lists the number of runoff events,
total amounts of runoff for each runoff plot-period data set, modelled runoff amounts,
differences between observed and modelled total amounts, Nash-Sutcliffe model
efficiency (ME) and optimised parameter values, for the two phases distinguished where
applicable.
Model performance was generally good (ME was 0.46-0.87 with an average of 0.74).
Plot runoff totals that were modelled with an accuracy that was closest to the average
model efficiency are shown in Fig. 12.8a (individual events) and d (cumulative totals),
with the worst and best fit equivalents in Fig. 12.8b,e and in Fig. 12.8c,f, respectively.
Runoff depth was modelled within a few millimetres for the majority of events, although
the difference was occasionally up to 10 mm. The relatively poor performance shown in
Fig. 12.8b and e is caused by an over-estimation at the beginning of the rainy season.
Similar initial over-estimations were also observed other data sets (including some
pertaining to terrace units, see below) and may be explained by the influence of the
preceding dry season on infiltration characteristics. Furthermore, runoff measurements
from one particular plot (B-I-1, Fig. 12.8b) probably experienced technical problems at
the start of the measuring period; if the first ten events are not taken into account model
efficiency is increased to 0.81 and the cumulative modelled amounts differ by 2 mm
(3%) only. Such technical problems did not occur in the other cases and model
efficiencies were all in excess of 0.65. Omitting the earlier-mentioned runoff plot (B-I-1)
modelled and observed runoff totals differed from observed depths by -26% to +20% and
often considerably less (cf. Table 12.7).
Values of I
m
and S
I
derived for the different plots by optimisation varied widely and
not always consistently (Table 12.7). Maximum infiltration capacity ranged between 18
and 442 mm h
-1
. The lowest values were found for bed plots kept bare or planted with
cassava or ginger and regularly weeded (25-44 mm h
-1
), followed by (almost) bare
terrace risers (50-105 mm h
-1
). Bed sections with mixed crops usually demonstrated a
0
5
10
15
20
25
30
0 100 200 300 400 500
S
I
(mm)
I
m

(
m
m

h
-
1
)
terrace units
bed sections
riser sections
Fig. 12.9. Relationship between maximum average infiltration rate (I
m
and
m
for terrace
risers/beds and units, respectively) and initial additional infiltration (S
I
or S
~
I
, respectively).
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
249
higher I
m
value, at 72 to more than 233 mm h
-1
. The highest I
m
values were associated
with low vegetation types and surface mulching. Maximum infiltration rate increased
where the crop was left on the terrace bed after harvesting, e.g. from 72 to 443 mm h
-1
for
plot B-I-1 and from 18 to 35 mm h
-1
for plot B-I-8. Similarly, the application of mulch at
the start of the 1999/2000 wet season increased I
m
from 78 to 115 mm h
-1
in mixed
cropping of cassava, maize and rice, although the already high infiltration rates below a
mixture of cassava, maize and peanut did not seem to be affected (223 and 213 mm h
-1
for plots with and without mulch, respectively; Table 12.7). Maximum average
infiltration rates on the risers were considerably higher when the riser was vegetated with
grass (128-298 mm h
-1
) than if shrubby herb was planted (80-130 mm h
-1
). Interestingly,
infiltration was even reduced from 86 to 35 mm h
-1
after shrub cuttings were planted on
bare terrace riser (R-I-9). To plant the cuttings one had to rest ones hand on the riser for
support, causing some compaction. Similarly, the exceptionally low I
m
value of 18 mm h
-
1
for a relatively short, steep terrace bed (B-I-8) may be related to the relatively thin top
soil, but also to compaction by people passing. The two terrace bed plots established near
terraces E and F demonstrated a gradual decrease in infiltration rate; infiltration rates
decreased after harvesting in 1999, from 103 to 62 mm h
-1
(B-I-5) and from ca. 233 to ca.
224 mm h
-1
(B-I-12), contrary to the increase observed after harvesting on other beds.
The next season again a lower value of I
m
=78 mm h
-1
was determined for a nearby plot
with comparable crop cover.
For initial storage amount (S
I
) optimised values ranged between zero and 13.0 mm
(Table 12.7; Fig. 12.9). Despite considerable scatter, the highest I
m
values (>200 mm h
-1
)
generally seemed to be associated with low S
I
values (0.2-0.7 mm), whereas I
m
values of
<100 mm h
-1
often corresponded to S
I
values in excess of 0.7 mm (cf. Fig. 12.9).
12.5.5. Modelling event runoff from terrace units
The storm-based runoff model equation for terrace units (Eq. [12.12]) was used to
simulate event runoff depths from six terrace units during two seasons. Lengths and
characteristics of the simulation periods are listed in Table 12.8. In 1998/99, combined
rainfall intensity and daily runoff measurements were available for a 130 day period (3
December 1998-11 April 1999) with a total rainfall of 1684 mm (=34.1 mm h
-1
). In
1999/2000, measurements were available for 106 days (26 October 1999-8 February
2000) with 1335 mm rainfall ( =32.3 mm h
-1
). The 1998/99 period was divided into
storms representing the pre- and post-harvest periods, respectively; this was not
necessary in 1999/2000. Table 12.8 lists the model results for each terrace and simulation
period. Overall model efficiency ranged between 0.67 and 0.82, with an average
performance that was better than for the terrace bed and riser plots at 0.81 versus 0.75,
respectively. The closest-to-average, best and worst results are illustrated in Fig. 12.10a
to f. Differences between modelled and observed cumulative runoff amounts were less
than 5%, except for a difference of +15% for terrace B in 1999/2000 which can again be
attributed to the preceding dry season (Fig. 12.10f).
Optimum values for apparent average composite maximum infiltration rate (
m
) for
the respective terrace units were 21-186 mm h
-1
before harvesting and 27-299 mm h
-1
thereafter. The lowest values were found again for the terraces of the steepest part of the
slopes (CD) at 21-51 mm h
-1
. Infiltration rates on the other four terraces were very
similar in the first year (124-145 mm h
-1
before and 82-304 mm h
-1
after harvesting) but
values were markedly different in the second year, at 183-186 mm h
-1
for terraces A and
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
250
B versus 76-81 mm h
-1
for terraces E and F. The magnitude of
m
increased after
harvesting on terrace E but decreased on terrace F (from 132 to 82 mm h
-1
), in line with
observations on nearby terrace beds (cf. Table 12.5). The other four terraces (A-D) were
cropped for at least four seasons before 1998, whereas terraces E and F had been left
fallow. Soil surface conditions started to deteriorate when the two terraces were taken in
use again and this may go some way to explain the decrease of
m
.
Inferred pre-harvest S
~
I
values were 0.04-3.51 mm but were reduced in all cases after
harvesting to 0-1.73 mm (Table 12.8). As found earlier for the terrace beds and risers,
there appeared to be an inverse relationship between average maximum infiltration rate
and initial excess infiltration (Fig. 12.9).
12.6. Discussion
12.6.1. Overall model performance
Runoff rates measured on terrace beds and risers and on terrace units were simulated
in the current study with an average model efficiency of 0.59 (Tables 12.7 and 12.8). Yu
et al. (1997c, 1997d) tested the SVI model using rainfall and plot runoff data from six
different locations in Australia and South-east Asia and obtained model efficiencies of
0.72-0.94. However, they used values for average maximum infiltration rate (I
m
), initial
infiltration (F
0
) and time lag (K) that were optimised for each individual event, whereas
in the present study sets of single parameter values derived from longer-term
Terrace Observed Modelled Pre-harvest Post-harvest
Q
~
tot
rc
Q
~
tot
Diff. ME
S
~
I

m S
~
I

m
mm % mm % mm mm h
-1
mm mm h
-1
1998/1999
A 117 7 115 -2% 0.79 0.47 124 0.38 228
B 116 7 116 0% 0.91 0.04 145 - 304
C 268 16 266 -1% 0.88 3.51 29 1.73 56
D 434 26 432 -1% 0.94 2.76 21 1.52 27
E 114 7 107 -5% 0.94 0.75 138 0.08 299
F 136 8 134 -1% 0.82 0.66 132 - 82
1999/2000
A 93 7 93 0% 0.85 0.13 186 - -
B 84 6 97 15% 0.67 0.07 183 - -
C 242 18 244 1% 0.74 2.23 36 - -
D 159 12 166 5% 0.68 2.78 51 - -
E 130 10 136 4% 0.73 1.75 76 - -
F 133 10 132 -1% 0.77 1.59 81 - -
Table 12.8. Results of application of the event-based infiltration equation for terrace unit
plots. Listed are runoff coefficient (rc), cumulative observed and modelled runoff amounts
(
Q
~
tot
), the relative difference between cumulative amounts, model efficiency (ME) and
optimised values of the model parameters effective average maximum infiltration rate (
m
) and
initial additional infiltration ( S
~
I
), for pre- and post-harvest phases where applicable.
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
251
measurements of event runoff depth were used. For comparative purposes runoff totals
for the storm events listed in Tables 12.5 and 12.6 were also simulated with the three
model parameters (I
m
or
m
, S
I
or S
~
I
, and K) optimised for each storm. Optimisation of S
I
in particular produced better fits: resulting model efficiencies were markedly higher,
from an average ME=0.59 to 0.70(0.20) and 0.82 (0.15) for the terrace beds and risers,
respectively, and from 0.59 to 0.84(0.13) for the terrace unit data. These values are
0
5
10
15
0 5 10 15
Q
tot
observed (mm)
Q
t
o
t

m
o
d
e
l
l
e
d

(
m
m
)
(a)
0
50
100
150
0 50 100 150
Cum. observed Q
tot
(mm)
C
u
m
.

m
o
d
e
l
l
e
d

Q
t
o
t

(
m
m
)
harvest
(d)
0
10
20
30
40
0 10 20 30 40
Q
tot
observed (mm)
Q
t
o
t

m
o
d
e
l
l
e
d

(
m
m
)
(b)
0
100
200
300
400
500
0 100 200 300 400 500
Cum. observed Q
tot
(mm)
C
u
m
.

m
o
d
e
l
l
e
d

Q
t
o
t

(
m
m
)
harvest
(e)
0
5
10
15
0 5 10 15
Q
tot
observed (mm)
Q
t
o
t

m
o
d
e
l
l
e
d

(
m
m
)
(c)
0
50
100
150
0 50 100 150
Cum. observed Q
tot
(mm)
C
u
m
.

m
o
d
e
l
l
e
d

Q
t
o
t

(
m
m
)
(f)
Fig. 12.10. Observed and modelled event runoff depth from terrace units: the most average
performance (ME=0.82) on terrace F in 1998/99 (a and d); the best performance (ME=0.94)
on terrace D in 1998/99 (b and e); and the worst performance (ME=0.67), on terrace B in
1999/00 (c and f). Times of maize harvest are indicated where applicable.
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
252
much more similar to the range cited by Yu et al., but the optimised parameter values
also exhibited correspondingly high coefficients of variation (cf. Yu et al., 1997c) of
typically around 0.5 and more than unity in some cases.
Model efficiency, both for simulated runoff hydrographs and event runoff depths,
was generally greatest for plots with high runoff coefficients (such as those with a bare
surface, little vegetative cover and, in particular, compacted surfaces). This phenomenon
was also observed by Yu et al. (1997c, 1997d), who reported ME to increase from 0.72
for a plot with rc=6% (I
m
=444 mm h
-1
) to 0.94 for a plot with rc=53% (I
m
=13.9 mm h
-1
),
and they suggest that an important reason for this tendency related to the increased
noise/signal ratio associated with overall low runoff rates. The same may also be true for
the current study; for example, the 2- or 3-minute water level and daily runoff
measurements may have been affected by any measurement errors which have their
greatest influence on low runoff rates or volumes. Resampling rainfall intensity data
before modelling hydrographs also introduced errors caused by time-averaging. In
addition, rainfall intensity measurements were occasionally made at some distance away
from the runoff plots (cf. Section 12.2) and this may have introduced discrepancies in the
timing of cloud bursts. The resulting lack of synchronicity reduced model efficiency even
when the pattern of simulated runoff rates was still very similar to that observed.
12.6.2. Infiltration and runoff generation
Derived values of average maximum infiltration rate (I
m
) exhibited a wide range:
from 18 mm h
-1
on a compacted, steep bed to 443 mm h
-1
for a mulched and well-
vegetated bed. Values for (almost) bare terrace beds and risers were 25-105 mm h
-1
in the
present study (rc=8-21%) with an average of 53 mm h
-1
(rc=15%) (Table 12.7). Such
values are within the range of 4.1 (3.6) mm h
-1
to 443.8 ( 309.1) mm h
-1
reported by
Yu et al. (1997d) for six locations in South-east Asia and Australia on a variety of soil
types. The highest I
m
was derived for a bare, steep plot on a well-aggregated clay-silt
Inceptisol developed from basaltic parent material on Leyte in the Philippines,
corresponding to a runoff coefficient of 2% (P=2800 mm yr
-1
). For bare, aggregated clay-
soils in volcanic Alfisols at Los Baos in the Philippines a maximum infiltration rate of
44.1(46.5) mm h
-1
(rc=19%) was found, whereas a value of 97.9(113.1) mm h
-1
(rc=2%) was determined for an Ultisol developed in shales in northern Thailand (Yu et
al., 1997)
The presently derived I
m
values also agree well with those inferred for small (0.6x0.3
m) trays filled with the same soil type and exposed to natural rainfall (70-151 mm h
-1
,
rc=9-24 %; Chapter 10). The determined maximum infiltration rates are intermediate
between saturated hydraulic conductivity values obtained with double-ring infiltrometry
and unsaturated conductivities at low suction determined by tension infiltrometry (cf.
Table 12.1). However, it should be noted that these measurements were made on terrace
beds with a crop cover and therefore the associated infiltration rates may represent under-
estimates. Taking Table 12.1 at face value, average saturated infiltration rates of 260-877
mm h
-1
would be expected over larger areas on the terrace beds, compared with only 4-64
mm h
-1
at a matric suction of -3 cm (pF=0.5). Comparing these values to I
m
values listed
in Table 12.7 suggests that in practice many larger pores in the terrace bed soil are not
functional during rainfall infiltration. Nearby soil water content measurements confirmed
this notion (Chapter 6).
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
253
Not surprisingly, the model results suggest that the presence of a good vegetation or
mulch cover has a positive effect on infiltration rates. However, the effect varies between
the various types of cover (Tables 12.2, 12.3 and 12.7). When compared with a bare
surface, frequently weeded cassava and ginger crops did little or nothing to increase
infiltration and I
m
values remained low (29-44 mm h
-1
; rc=10-17%; Table 12.7). Low
vegetation types generally promoted infiltration: a cropped bed invaded by weeds
produced less runoff (8-9%) than its clean-cultivated counterpart. Similarly, adding
maize and rice or peanut to the cassava crop increased I
m
to values of 72->233 mm h
-1
(rc<11%; Table 12.7). The effect of a dense vegetative cover was even more pronounced
on the terrace risers: I
m
increased from 50-105 mm h
-1
(rc=13-28%) on risers with
(almost) no cover to 80-130 mm h
-1
(rc=5-16%) at 15-85% herb cover and up to 128-
>298 mm h
-1
(rc=0.6-11%) for a dense grass cover (Tables 12.3 and 12.7). Mulching also
enhanced infiltration in most cases. This was observed both on plots were mulching was
applied at the start of the rainy season and, more suddenly, when maize harvest residues
were left on the terrace beds. The importance of a good vegetation and mulch cover was
also observed by Yu et al. (1997d) at various sites in South-east Asia, who observed that
clean cultivation farmers practice with mono-cropping or mixed cropping of maize
generally had little or even a negative effect on infiltration, while application of
hedgerows, mulch or alley cropping resulted in a considerable increase. Similarly,
Suwardjo and Abujamin (1988) reported that mulching and minimum tillage on volcanic
Haplorthox soils elsewhere in West Java decreased runoff coefficients to 3-16%,
compared with 19-21% and 15-21% for bare and non-mulched, traditionally plowed and
cropped plots, respectively.
The strong contrast in infiltration rates between the well-aggregated (top-)soil and
the more massive (sub-) soil apparent from field hydrological measurements (Table 12.1)
was confirmed by the current study. Compaction of terrace soil bed soil by foot traffic
decreased the infiltration rate to well below average, at 18 mm h
-1
versus more than 72
mm h
-1
for the other plots (B-I-8; Table 12.7). Similarly, the mere leaning on a riser
surface when planting shrubs more than halved the maximum infiltration rate, from 86 to
35 mm h
-1
(R-I-9, Table 12.7). Such findings illustrate the vulnerability to compaction of
the studied light volcanic soil.
The six terrace units (A-F) showed effective apparent average maximum infiltration
(
m
) rates that were roughly intermediate between I
m
values associated with the risers and
beds. Although comparing
m
and I
m
values is difficult for reasons outlined in Section
12.4.3, an approximate comparison of runoff characteristics can be made by comparing
terrace unit runoff coefficient to the area average runoff coefficient of the three
contributing zones (riser, bed and toe drain). For a specific value of , a value for
m
can
be chosen such that the areal average runoff coefficient results from Eq. [12.11]; this
value should be similar to that derived from the terrace unit runoff measurements
themselves. The overall average value of 34.1 mm h
-1
for the 1998/1999 pre-harvest
period was used in combination with relative areas of the three terrace components and
corresponding I
m
values based on plot runoff measurements (cf. Table 12.7); the results
are summarised in Table 12.9. The approach produced reasonable values for terraces C
and D but values were too low for the other four terraces. The limited number of
replications is probably an important reason for the discrepancies. For example, runoff
from only one terrace bed plot was measured for terraces A and B and terraces C and D
which, in addition, yielded strongly contrasting I
m
values of 72 and 18 mm h
-1
,
respectively. Using a value of about 223 mm h
-1
(as measured on several terrace bed
plots near terraces E and F) for terraces A and B gave
m
estimates of 136-209 for terrace
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
254
A and 120-208 mm h
-1
for terrace B, which were much closer to observed values (Table
12.9). Finally, an important reason for the low
m
estimates for terraces E and F may
related to the relatively low I
m
estimates for the toe drain.
12.6.3. Initial excess infiltration
There appeared to be an inverse relationship between average maximum infiltration
rate (I
m
or
m
) and initial excess infiltration (S
I
or S
~
I
; Fig. 12.9). The greater exposure of
the bare or poorly covered soils often associated with low infiltration rates may have led
to higher soil evaporation and therefore greater soil water suction. Probably more
importantly, however, the high rate of normal infiltration that occurs in a permeable
soils may be already able to replenish any soil moisture deficits at the start of the storm
and so reduces the additional amount of water needed for this.
When comparing observed and simulated runoff hydrographs it was noted that often
a considerable portion of the discrepancy was related to the applied value of S
I or S
~
I
. To
explore the reasons for these variations the optimised values for individual storm events
were compared with daily measurements of surface soil water suction on a terrace bed
near terraces E and F (cf. Chapter 6). Although variations in soil matric suction were
rather modest, some of the highest optimised S
I
values did indeed occur during a drier
period (1-16 February 1999). However, much lower values were found as well and no
systematic relationship between S
I
and soil water suction could be established. The S
I
term also includes any additional water stored on the canopy or in surface depressions.
The storage capacity of mixed crops was determined in a rainfall interception study at the
same location and was estimated to reach a maximum of 0.25-0.30 mm at the height of
crop development (Chapter 5). Surface depression storage was not measured, but
application of the empirical relationship based on soil surface roughness (estimated at 5-
20 mm) and slope gradient developed by Onstad (1984) suggested a depression storage
decreasing from 0.6-3.5 mm for a horizontal surface to zero for plots at gradients above
10%. However, canopy and surface depression storage have entirely different effects:
Terrace Relative area I
m
of contributing areas (N)
m
bed riser drain bed riser toe drain estimated observed
% % % mm h
-1
mm h
-1
mm h
-1
mm h
-1
mm h
-1
A 77 18 5 72 (1) 250 (1) 3-86 66-85 124
B 76 17 7 72 (1) 250 (1) 3-86 62-85 145
C 41 48 10 18 (1) 64-105 (3) 5-56 27-42 29
D 36 49 15 18 (1) 128-298 (2) 5-56 32-57 21
E 48 37 15 78-223 (6) 50-65 (4) 3-5 37-58 138
F 48 39 13 78-223 (6) 50-65 (4) 3-5 40-63 132
Table 12.9. Average and ranges of I
m
values derived for terrace beds and risers near the six
bench terrace units derived from modelling and effective average maximum infiltration rates
(
m
) for the terrace units as estimated from these values and derived from runoff modelling,
respectively (see text for explanation). The number of corresponding data sets (N) is given
between brackets (values for toe drains based on hydraulic conductivity measurements, cf.
Table 12.1).
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
255
while the canopy may intercept rainfall that otherwise could have infiltrated easily,
surface depressions will retain water that has failed to infiltrate. The last can therefore be
expected to be more important for S
I
or S
~
I
. It can be concluded that the variation in S
I
values for terrace risers (0-3.9 mm, cf. Table 12.7) is most likely related to drying out of
the top soil between events, whereas high S
I
values (up to 13.0 mm) occasionally found
for terrace beds may also include a certain amount of surface depression storage (cf.
Table 12.7). The terrace unit plots produced apparent S
~
I
values that were similar (0.04 to
3.51 mm; Table 12.8).
Average S
I
and S
~
I
values found in this study cannot be compared directly with
values for the initial infiltration amount (F
0
) reported by Yu et al. (1997c, 1997d)
because of their conceptually different definition. A case can be made for both
approaches, but the one followed in the present study has clear advantages for the
mathematical storm-based description of infiltration (cf. Chapter 11). An approximate
indication of F
0
values that correspond to the determined S
I
values may be derived by
assuming that rainfall intensity at the beginning of the storms was, on average, equal to
the depth-averaged rainfall intensity () for the entire storm. In that case F
0
is equal to
the threshold amount of rainfall (P in mm) needed to realise S
I
and can be estimated as
(cf. Eq. [12.6]):
1
0
1 ln 1

1
]
1

,
_

+
m
m
I
I
R
R
I
S P F

[12.13]
where S
I
and I
m
are to be replaced by S
~
I
and
m
, respectively, for terrace units. Values of
F
0
were calculated using an value of 34 mm h
-1
, representing the range of values
obtained for the various measurement periods (32.1-36.2 mm h
-1
), in combination with
the parameters listed in Tables 12.7 and 12.8. Resulting estimates of F
0
were on average
5.8 mm (0-21 mm) for the terrace risers, 14.8 mm (0-96 mm) for the beds and 5.3 mm (0-
12 mm) for terrace units. Unlike S
I
( S
~
I
) values the calculated F
0
values did not appear to
have any significant relationship with I
m
, which lends some support to explanation for the
S
I
-I
m
relationship proposed earlier. The calculated F
0
values are within the range of 5.6-
18.4 mm reported by Yu et al. (1997c, 1997d) for the earlier-mentioned clay soils in the
Philippines and Thailand.
12.6.4. Runoff peak attenuation and effective runoff rates
The original SVI model was used with 2 to 3-minute rainfall intensity interval data
to simulate observed runoff rates, involving a kinematic wave approximation for runoff
routing. Provided that the flow is hydraulically turbulent and uniform, Yu et al. (1997d)
suggested that the resulting time lag (K) values is related to Mannings roughness
coefficient (n in s m
-1/3
), flow distance (L in m), slope gradient (S) and volumetric flow
rate per unit slope width (q in m
2
s
-1
) as:
5 2
5 3
2
1

1
]
1

q
S
nL
K [12.14]
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
256
Eq. [12.14] was used in combination with the dimensions of the plots from which runoff
rate was measured, whereas a value of n=0.017 s m
-1/3
was derived from controlled flume
studies (Chapter 13). Observed effective runoff rates for the terrace bed (0.9-18.2 mm
h
-1
), terrace riser (3.0-49.6 mm h
-1
) and terrace unit (0.9-57.7 mm h
-1
) were taken as
representative q values. In the last case, the value was corrected for the fact that the
runoff was accumulated in the terrace toe drain before leaving the terrace unit, through
multiplication times the ratio of terrace width over toe drain width (cf. Table 12.4).
For the terrace bed this approach yielded a time lag of 0.29-0.96 minutes, which
comes close to the value of 0.28 minutes derived by hydrograph modelling. At 0.07-0.22
minutes, the time lag estimated for the terrace riser by Eq [12.14] is much lower than the
fitted value of 0.91 minutes. The time lag derived for the terrace riser by hydrograph
modelling was greater than that for the less steep, longer bed section. Part of the
explanation may be related to the fact that runoff from the riser, unlike that from the bed,
needed to travel a short lateral distance through a low-inclined collecting gutter before
entering the runoff container. This may have introduced an additional time lag, which
should then be considered uncharacteristic of the natural situation. For the terrace unit a
time lag of 0.13-0.99 minutes was found, which is less than the 1.23-3.12 minutes
derived by modelling. Part of this difference may be explained by the fact that Eq.
[12.14] implicitly assumes that all runoff flows directly towards the outlet of the terrace,
whereas in reality runoff first flows towards the toe drain before diverting to the drainage
point. As such, the simulated time lag of 1.2-3.1 minutes seems plausible after all.
The storm-based equations produced estimates of effective runoff rate (Q
e
and Q
~
e
)
that agreed reasonably well with observed values, often better so than did values
calculated from hydrographs simulated with (resampled) rainfall intensity data. Under-
estimation of runoff rates associated with using resampled rainfall data was avoided by
using values (based on original tipping bucket data) and this presumably partially
compensated errors introduced by neglecting the attenuation of flow rates. Use of a
reduction factor proved not necessary for the terrace units, for which coefficient of
determination between Q
~
e
and the observed value was still good. The relative
agreement between observed and calculated values increased with their actual magnitude.
Only storms of intermediate size could be used for the terrace units, suggesting that
agreement could have been even better for larger storms. Because of high intensity, large
storms are the most important in runoff (and sediment generation) this finding is an
encouraging result in terms of modelling.
12.7. Outlook for further model application
Some difficulties remain when modelling infiltration for an area that comprises
contributing zones with contrasting values of S
I
and I
m
, particularly if this results in a
spatial distribution of I
m
that differs significantly from the exponential distribution of
maximum infiltration rates assumed by the SVI model. An empirical method was
employed in the current study to overcome this problem. A more robust mathematical
approach based on a different type of distribution is feasible but would require additional
parameters; the availability and resolution of the runoff rate measurements were not
considered sufficient to attempt such an approach at this stage, however.
Relationships between the infiltration model parameters initial additional infiltration
depth (S
I
) and average maximum infiltration rate (I
m
), and between S
I
and initial
CHAPTER 12 - RUNOFF FROM BENCH TERRACES
257
infiltration (F
0
) as defined by Yu et al. (1997c, 1997d) were only briefly discussed and
deserve more attention. Although the relative importance of initial infiltration is least for
the large events that dominate runoff and sediment generation processes, initial (excess)
infiltration does constitute an important phenomenon in general hydrological modelling.
The present study has demonstrated that both runoff rates and event runoff depth can
be modelled with reasonable accuracy by a relatively simple infiltration model requiring
few parameters. While the present modelling concept is the same as the SVI model of Yu
et al. (1997bc), the reduction of storm characteristics to only two parameters, event
rainfall depth (P) and depth-averaged rainfall intensity (), facilitates the application of
the model considerably and provides scope for the predictive use of runoff and erosion
models. The analysed rainfall data suggested that a well-defined statistical relationship
exists between storm rainfall depth (P) and depth-averaged rainfall intensity (). Similar
relationships may occur at other locations and may vary according to local climate.
Exploration of the potential of available rainfall intensity records or, alternatively, depth-
duration curves may prove particularly fertile (cf. Chapter 11).
The model and model results on runoff generation presented here are extended in
Chapter 13, focusing on the modelling soil loss from terrace risers, beds and units in the
same humid tropical volcanic upland environment.

Das könnte Ihnen auch gefallen