Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

133

Chapter 8
Exponential distribution theory and the interpretation
of splash detachment and transport experiments
Abstract Soil detachment and transport by rain splash is usually the first step in soil loss and
sediment transport. It can be measured using a variety of approaches, including splash cups,
trays, and boards. However, the results of splash experiments are affected by their geometry and
not readily translated into generally applicable parameters. In this paper we develop a theory that
can be used to interpret splash experiments. It is based on the assumption that the spatial
distribution of particles splashed from a point source can be described by an exponential decay
function, for which there is considerable support in the literature. The theory is evaluated for the
cited experimental techniques, partly with the use of a numerical model. It is made clear that
conventional measurements of splash and the true rate of detachment by splash are two different
entities that can be linked if the average splash length is known. In principle, the theory is not
valid for a sloping surface, but analysis of the magnitude of the error involved indicates that in
many cases good estimates of detached amounts can still be obtained.
Published as: Van Dijk, A.I.J.M., Meesters, A.G.C.A., Bruijnzeel, L.A., 2002. Exponential
distribution theory and the interpretation of splash detachment and transport experiments. Soil
Science Society of America Journal 66 (in press)
8.1. Introduction
The detachment and transport of soil particles ensuing from the impact of rain drops,
or splash for short, is usually considered an important first step in the chain of processes
leading to loss of soil and subsequent sediment transport. Falling rain drops are able to
detach much larger amounts of soil particles than unconcentrated overland flow, after
which the detached particles may be entrained and carried off by the flowing water
(Hudson, 1995). In addition, splash may result in significant net transport of sediment on
sloping soils (e.g. Moeyersons and de Ploey, 1976; Wan et al., 1996). To quantify splash
detachment and transport, a range of techniques has been developed over the years.
Sreenivas et al. (1947) were among the first to use round cups or funnels embedded
in the soil to catch splashed particles. They considered the mass of collected material
divided by the surface area of the cup (with radius R in m), denoted here by m
R
, as an
indicator of detachment rate on the surrounding soil. This method has been applied in
many subsequent studies, a number of which were reviewed by Poesen and Torri (1988).
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 8 - EXPONENTIAL SPLASH DISTRIBUTION THEORY
134
However, the cup splash rate (m
R
) is not independent of cup size, as already put forward
by, inter alia, Rose (1960) and Bolline (1975). Furthermore, the relationship between cup
size and splash rate is also influenced by the distribution of distances over which the
splashed particles travel: given a particular amount of soil detached per unit of area
outside the cup (i.e. detachment rate , as mass per unit area) more sediment will end up
in the cup as the distance over which splashed particles travel increases. On the basis of
the reciprocity principle, the same will be true for a similar experimental technique in
which a cup filled with soil is subjected to (artificial) rain and the sediment ejected from
the cup is collected (Ellison, 1947; Rose, 1960; Al-Durrah and Bradford, 1981, 1982;
Kinnell, 1982; Riezebos and Epema, 1985; Sharma and Gupta, 1989). If the cup is small
compared with the average splash length, it can be argued that the bulk of sediment
splashed from the cup will end up outside the cup (Rose, 1960). However, the use of very
small cups gives rise to practical problems and therefore normally cups are used with a
diameter of a few to more than ten cm. Under such conditions, the resulting amount of
collected sediment can no longer be equated to detachment rate, as is still commonly
practised.
Alternatives to circular cups include splash boards (Ellison, 1944; Kwaad, 1977) and
trays (Quansah, 1981; Savat and Poesen, 1981; Wan et al., 1996), in which case the
amount of splash is expressed either per unit tray area or length. The results are again not
readily converted to actual detachment rate () because the amounts of sediment ending
up in the collectors themselves depend on the distribution of distances over which the
splashed particles travel. In addition, unless the dimensions of the tray are several orders
of magnitude greater than the average splash length, tray size will also have a significant
effect on measured splash rates, as does the size of splash cups.
The problems outlined above complicate the interpretation of experimental results
much more than is often acknowledged. Farmer (1973), for example, subjected a soil-
filled tray of 46 by 122 cm to artificial rainfall. From the difference in particle size
distributions between the original and splashed material, he inferred preferential
detachment of certain size classes. A comparable study by Gabriels and Moldenhauer
(1978) used trays of 30 by 45 cm. However, as Poesen and Torri (1988) demonstrated,
small particles are splashed over greater distances than larger particles. Consequently, a
greater fraction of detached particles will end up outside the tray for small size classes
than for larger size classes. Thus, wrong conclusions may be drawn from the
measurements (regardless of the possibility that smaller particles are indeed detached
more easily).
Summarising, the (apparent) detachment rates derived from cup, tray, or splash
board experiments must be considered ill-defined, experiment-specific and not applicable
to field situations, unless the geometry of the experiment and the spatial distribution of
the splashed particles around their source are taken into account. Recognising this,
Farrell et al. (1974) advanced a theory describing the influence of experimental geometry
on splash measurements. The practical application of their theory was limited initially,
because little information existed about the spatial distribution of splashed particles.
However, the spatial distribution of splash has been measured in later studies, such as
those of Poesen and Savat (1981), Savat and Poesen (1981), Riezebos and Epema (1985)
and Torri et al. (1987).
The objective of the current paper is to develop a mathematical distribution theory
that can be used to assess the influence of geometry and splash length in the type of
experiments described above. In doing so, it is initially assumed that the spatial
distribution of particles splashed from a point of rain drop impact can be described by an
CHAPTER 8 - EXPONENTIAL SPLASH DISTRIBUTION THEORY
135
exponential function, which will be called the fundamental splash distribution function
(FSDF). It will be shown that there is considerable support for this in the literature. On
the basis of the FSDF, the relationship between observed amounts of splash and actual
detachment rates will be evaluated in terms of a characteristic splash length and the
geometry of the experiment. In some cases, this is done analytically; in others, a
numerical model is used. A list of all symbols used is provided in Appendix 8.A for
reference.
8.2. Theory of an exponential fundamental splash distribution function (FSDF)
Let the rate of detachment (in g m
-2
) be defined as the mass per unit source area
that has been detached by rain splash. Consider the splash from a small source area onto
a small target area at radial distance r; let the splash density m
point
(r) (in g m
-4
) be the
splashed mass per unit of source area, per unit of target area. It is subsequently assumed
that m
point
(r) is described by the following exponential function:

,
_

r
r
r m
po
exp
2
) (
int

[8.1]
where (in m) is the average splash length, i.e. the mass-weighted average radial
distance over which the particles are splashed. Within the context of the theory presented
in this paper, Eq. [8.1] may be called the fundamental splash distribution function
(FSDF), since all other equations are derived from it by integration.
There is empirical support for the assumption that the FSDF conforms to an
exponential equation. Riezebos and Epema (1985) conducted laboratory experiments
using a soil cup surrounded by concentric rings at 5 cm intervals that collected sediment
splashed from the cup. Individual artificially produced water drops fell from different
heights on the centre of the soil-filled cup. In all cases, the experimental results for the
mass splashed beyond a distance r were described very well by an exponential function
(Fig. 8.1). It can be shown that this conforms to the assumed FSDF by considering the
amount of material M
ring
(r) (in g m
-3
) that is splashed from a small source area at the
centre of impact, onto a very narrow ring at a distance r from the source, per unit of
radial distance and per unit of source area:

,
_


r
r m r r M
po ring
exp ) ( 2 ) (
int

[8.2]
Since integration of Eq. [8.2] between r and infinity also yields an exponential function,
these results are in accordance with the hypothesis. It should be noted, however, that
because Riezebos and Epema (1985) used a cup diameter of 5 cm, the spatial distribution
of particles close to the impact centre remains unknown.
In principle, the amount of material detached (splash mass, M in g) by a single
impacting drop can be calculated once the area of impact (a, in m
2
) is known:
a dr M a M
ring

0
[8.3]
CHAPTER 8 - EXPONENTIAL SPLASH DISTRIBUTION THEORY
136
which demonstrates that the use of in Eq. [8.1] is consistent. In reality, the exact
definition of a and may create some problems. One reason is that Eq. [8.3] is only truly
valid if the area of impact a is of infinitely small size. However, a falling rain drop will
create an impact crater of up to a few millimetres in diameter and then break up into
droplets that carry soil particles (Ghadiri and Payne, 1988). Therefore, the highest density
of transported particles will occur around the perimeter of the impact crater. This does
not pose a major obstacle to using Eq. [8.1], provided that the average splash length is
not of the same order of magnitude as the size of the impact crater. Assuming, for
example, that the impact of a rain drop leaves a crater with a radius of 2 mm and the
average splash length () has a value of 0.1 m (a representative value, see below), then
the amount of material implicitly assumed by Eq. [8.1] to be inside the crater would be
only 2% of the total amount of splashed material. Another point to be made in this
respect is that splash detachment itself is difficult to determine and even to define. Some
soil particles may be detached but returned to their original position, while others may be
pushed away to the perimeter of the impact crater without actually leaving the soil
surface, as has been demonstrated in splash cup experiments (Kinnell, 1974, 1982) and in
numerical simulations of rain drop impact (Huang et al., 1982, 1983). For pragmatic
reasons, therefore, detachment is defined here as the amount of soil that appears to have
been detached based on measurements of splash. In some cases, this may include some
pushed material as well (Kinnell, 1974, 1982).
8.3. Application of the distribution theory to splash measured in one dimension
The distribution theory developed above can be applied to a situation in which
material is splashed across an (effectively) infinitely long straight boundary. This is
0.001
0.01
0.1
1
10
0 20 40 60 80 100
r (cm)
M
a
s
s

s
p
l
a
s
h
e
d

b
e
y
o
n
d

r

(
g
)
0.5 m 1 m 2 m
4 m
13 m
Fig. 8.1. The mass of soil splashed beyond a radial distance r from its point source (data from
Riezebos and Epema, 1985). Exponential functions are fitted to the data, which represent
splash resulting from artificial drops falling from indicated heights.
CHAPTER 8 - EXPONENTIAL SPLASH DISTRIBUTION THEORY
137
important to interpret measurements made with, for example, splash trays and splash
boards. To this end, the radial form of Eq. [8.1] has to be converted to a two-dimensional
co-ordinate system first and eventually to a single dimension. One way of doing this is to
imagine an array of very small unit source areas, forming a narrow source strip. First, the
splash density pattern resulting from one such source strip will be considered. Taking the
y-axis along the source strip, it is easy to see that splash density will be a function of the
distance |x| (perpendicular to the source strip) alone. This function can be determined by
considering an element of the source strip (having length dy and width dx) positioned at
(0, y) and a target element positioned at (x, 0). The contribution of the source element to
the splash density in (x, 0) is then given by m
point
(r)dxdy where r is the Euclidean distance
(Eq. [8.1]). The contribution of all elements forming the source strip to the splash density
at (x,0) will be denoted by v
strip
(x)dx. Because this involves the entire length of the strip,
v
strip
(x) has the dimension of mass per unit width of strip per unit target area (i.e. g m
-3
). It
can be found by integration along the length of the strip:
( )


+
2 2
int
) ( y x m dy dx dx x v
po strip
[8.4]
Thus, v
strip
(x) is the one-dimensional equivalent of the radial splash density function
represented by Eq. [8.1]. It can be shown that the solution of Eq. [8.4] involves a zero-
order modified Bessel function denoted by K
0
(Eq. [9.6.24] in Abramowitz and Stegun,
1965):
( ) dx
x
K y x
y x
dy dx dx x v
strip

,
_


+
+


0
2 2
2 2
exp
2
) (

[8.5]
This result implies that when the exponential FSDF is applied to a one-dimensional strip
source, it does not retain its exponential form. However, it does resemble an exponential
function at larger distances from the source (Fig. 8.2). Another characteristic of Eq. [8.5]
is that, like Eq. [8.1], it attains an infinite value for x=0. Mathematically this is not a
problem, since integration across a distance larger than zero will always yield a finite
value. In further calculations, it is important to know what amount of material splashed
from the source strip ends up beyond (a boundary at) a distance x. This can be found by
integration of Eq. [8.5], between x and infinity. The result may be called the transport
function, denoted by v
beyond
(x). It represents the splashed mass ending up beyond a
boundary line at distance x, per unit width and per unit of boundary length. Therefore,
v
beyond
(x) has the dimension of mass per area (in g m
-2
). The integration is written as:
( )

x
strip beyond
dx x v x v ' ' ) ( [8.6]
where x is the integration variable. There is no simple analytical solution to this integral
but its numerical solution is relatively straightforward, except for very small values of x
(cf. Abramowitz and Stegun, 1965, Eq. [11.1.9]; see Fig. 8.2). Taking things one step
further, a source area of soil is considered that is infinitely large at one side of a
boundary. The amount of material transported across this boundary may be called the
CHAPTER 8 - EXPONENTIAL SPLASH DISTRIBUTION THEORY
138
rate of transport, denoted by q and expressed in mass per unit boundary length (g m
-1
). It
represents the cumulative amounts of v
beyond
(x) for the entire assembly of possible source
strips positioned between zero and an infinitely large distance from the boundary. The
value of q may be found by single integration of Eq. [8.6] between 0 and infinity:
( ) ( ) ( ) ( )x x v x d x d x v x d x v x d dx x x v x d dx q
strip
x
strip
x
strip strip
+


0 0 0 0 0 0
[8.7]
where x is defined by x=x-x. The solution of this complicated integration is
surprisingly simple (cf. Abramowitz and Stegun, 1965, Eq. [11.4.22]):

q [8.8]
Eq. [8.8] describes the relationship between splash detachment rate () and splash
transport rate (q) across a straight line. Provided that is known, Eq. [8.8] can, in
principle, be used to calculate detachment rates from measurements of splash across a
straight boundary, for example from a soil tray.
However, the source area of an experimental tray is not infinitely large. Momentarily
assuming a tray of (effectively) infinite length but of finite width (w in m) the transport
rate over one side of the tray [q(w) in g m
-1
] can still be obtained using Eq. [8.6], which is
now integrated across 0<x<w, resulting in:
0.0001
0.001
0.01
0.1
1
0 1 2 3 4 5
|x|/ %
v
s
t
r
i
p
(
x
)
0.0001
0.001
0.01
0.1
1
v
b
e
y
o
n
d
(
x
)
Dimensionless density function
Dimensionless transport function
Fig. 8.2. One-dimensional distribution of particles splashed from a source strip expressed as a
density function of perpendicular distance x (v
strip
(x); Eq. [8.7]) and as a transport function
indicating the fraction of detached material splashed beyond distance x (v
beyond
(x); Eq. [8.6]).
All variables have been made dimensionless; dashed lines represent approximate exponential
functions. =average splash length.
CHAPTER 8 - EXPONENTIAL SPLASH DISTRIBUTION THEORY
139

,
_

w
x x
K dx dx w q
0 0
0
'
' ) (

[8.9]
For comparative purposes it will prove useful to express q(w) as a fraction F
w
of the
infinite splash transport rate (q), by isolating the latter in Eq. [8.8] and substitution into
Eq. [8.9], after which the resulting equation can be simplified (reversing the order of
integration and rewriting; cf. Eq. [11.3.27] in Abramowitz and Stegun, 1965) to:
( ) ( )

'


,
_


0 1
0 0
0 2
1
'
'
1 ) (
K d w K
w x x
K dx dx
q
w q
F
w
w
w
[8.10]
where the integration variable > is equal to (x/). F
w
may be regarded as a geometrical
correction factor that may be used to correct measurements of splash from a tray of
limited width. Equation 12 is still not readily solved analytically but it can be solved
numerically (cf. Abramowitz and Stegun, 1965; Eqs. 11.1.9 and 11.1.18). The resulting
solution is not an exponential function itself, but F
w
can be approximated very well by
the following exponential function:

,
_


w
q
w q
F
w
30 . 1 exp 1
) (
[8.11]
This expression agrees within 1% with values calculated using Eq. [8.10] for
1<w/<100. For 0.2w/1, the difference is still less than 10%. In principle, the
detachment rate can be calculated from the results of soil tray experiments using the
geometrical correction factor given by Eq. [8.11], but the average splash length () will
need to be known. Its value may be derived from measurements of splash over a range of
distances, as will be described below. Of course Eq. [8.11] still assumes an infinitely
long tray. This problem will be approached with a numerical model further on.
A situation analogous to the one described above concerns splash transport from an
infinitely large source area on one side of a boundary, onto the area between distance 0
and X from it, denoted here by q(<X). Because of the mathematical reciprocity of this
situation compared with that of a tray of limited width, Eqs. [8.8] and [8.11] can also be
used here, substituting X for w in the latter. In a similar manner equations can be derived
for splash beyond a distance X and splash between distances X
1
and X
2
, denoted by q(>X)
and q(X), respectively. The resulting equations read:
1
]
1

,
_

<
X
X q 30 . 1 exp 1 ) (

[8.12a]

,
_

>
X
X q 30 . 1 exp ) (

[8.12b]
1
]
1

,
_


,
_


2 1
30 . 1 exp 30 . 1 exp ) (
X X
X q

[8.12c]
CHAPTER 8 - EXPONENTIAL SPLASH DISTRIBUTION THEORY
140
Clearly, these equations too are based on an approximation of the actual geometrical
correction factor (Eq. [8.11] versus Eq. [8.10]) and therefore should not be used at very
small distances from the boundary. The practical application of the equations above is
that through a plot of distance on the x-axis versus the amount of material splashed
beyond or before X on the y-axis, an exponential function can be drawn, the exponent of
which yields the average splash length (; Eq. [8.12a] or [8.12b]). Once is known, the
detachment rate () can be calculated from the coefficient of the fitted function.
Examples of experimental results that can be interpreted with Eq. [8.12c] include
those of Savat and Poesen (1981), who measured splash from a soil-filled tray (100 x 20
cm) onto a series of 10 cm wide strips next to the long side of the tray (Poesen & Savat,
1981). The results are shown in Fig. 8.3; the average splash lengths () derived from the
exponential curves that were fitted to the data were between 0.11 and 0.15 m. However,
strictly speaking, Eq. [8.12c] could not be used in this case because the tray width was
limited to 20 cm and numerical calculation is therefore needed. Cursory analysis using
Eq. [8.11] in combination with the determined values suggests that total splash
transport from the experimental tray would have been ca. 8-18% less than that from an
infinitely large area. Similarly, the results of soil tray experiments (size 200 by 50 cm) by
Torri et al. (1987) were described very well by exponential functions, the application of
which yielded average splash lengths of 0.10-0.12 m. In this case, application of Eq.
[8.11] suggested an under-estimation of true transport rate (q) of less than 0.5%. The fact
that simple exponential functions fitted the measurements so well in both cases offers
further support to the theory outlined above, although of course the distribution of splash
very close to the source remained unknown in these experiments.
0
5
10
15
20
0 10 20 30 40 50 60
Distance from the tray (cm)
q
(

X
)

(
g

m
-
1
)
sand (600 mu)
sand (300 mu)
sand (125 mu)
silt (30 mu)
Fig. 8.3. Rates of splash transport [q(x)] onto 10-cm wide strips, placed at increasing
distances from a soil tray subjected to artificial rainfall and filled with different soils
(approximate median grain size indicated; data from Savat and Poesen, 1981) with Eq. [8.12c]
fitted to the data.
CHAPTER 8 - EXPONENTIAL SPLASH DISTRIBUTION THEORY
141
8.4. Splash from or onto a confined rectangular area
A more complicated situation arises when an area of both limited width and limited
length is considered. The question now is how the observed transport rates compare with
true transport rates (q). Dimension analysis shows that the three variables involved
include the average splash length () and the sides of the tray, denoted by A and B (see
Fig. 8.4a). Furthermore, two dimensionless groups can be formed from these variables by
considering the ratios A/ and B/ or, more conveniently, A/ and A/B.
The problem was approached using a simple matrix model involving the density
function represented by Eq. [8.1]. For the calculations, input values of , A/ and A/B
were needed. Detachment rate () was set at unity. A matrix was constructed with
elements of unit area and of such a size that the edge of the matrix was at a distance of
five times from the central tray on all sides. Application of Eq. [8.12b] suggests that
under these conditions less than 0.15% of the amount of material splashed from the tray
would be unaccounted for. All elements were given (x,y) co-ordinates and the elements
representing the soil in the tray were labelled (Fig. 8.4a). For each target matrix element
(x
1
,y
1
), the amount of material (per unit area) received from any source matrix element
(x
2
,y
2
) was calculated using Eq. [8.1] (provided of course that the source element was
part of the tray area). As will be apparent from Eq. [8.1], calculations cannot be made if
the source and target area are one, but this will not affect the final result since the source
(tray) and target (receiving) areas are always separate. However, to allow a check on
numerical errors in the model, an accurate substitute value for this special case was
determined (details about the followed procedure are available from the authors upon
request). The output of the model consists of a matrix in which the total amount of
material deposited on each element, expressed as a fraction of (unit) detachment rate, was
Fig. 8.4. Illustration of (a) the matrix used to model the distribution of particles splashed from
a rectangular area (the meaning of the areas labelled SCREEN and CORNER is explained in
the text); (b) a soil tray bordered by a very wide collecting screen having the same length ; (c)
a soil tray bordered by an infinitely large horizontal collecting screen; and (d) a soil tray
bordered by a collector with vertical splash guards on three sides.
CHAPTER 8 - EXPONENTIAL SPLASH DISTRIBUTION THEORY
142
assigned to that element. As a control on numerical errors the total amount of material for
the entire target matrix was calculated and divided by the number of source elements; the
resulting value should be close to unity. If the latter deviated by more than 1%, the values
of , A and B were doubled to reduce numerical errors and the model was run again.
The matrix model was used to find relationships between A/% and A/B on the one
hand, and transport rate on the other, depending on the boundary conditions of the
problem. For example, consider a tray of soil with a horizontal collecting screen on one
side, having the same length as the tray and a width of five times the average splash
length in the direction of the tray (numerically, this will be virtually equal to a screen that
effectively extends infinitely in the direction of the tray; see Fig. 8.4b). The total amount
of material splashed onto this screen (as a ratio to detachment rate) is obtained by
summing the amounts for all elements in the area labelled SCREEN in Fig. 8.4a. This
total was calculated for a range of A/ and A/B values. The results were then compared
with transport rates (q) from an infinitely large area of soil (Eq. [8.8], again using unit
detachment rate). The ratios of the two values are listed in Table 8.1a for various
combinations of A/ and A/B are shown in Fig. 8.5a. These ratios can be used to correct
the results of experiments involving splash trays and screens for their geometry.
Similarly, the total amount of material splashed from the tray onto a larger screen
extending to either side of the area labelled SCREEN is obtained by summation of all
elements in the areas denoted SCREEN and CORNER (Fig. 8.4a, c). Again, these
values were calculated for a range of A/ and A/B values and the corresponding ratios to
actual transport rate (q) are listed in Table 8.1b and plotted in Fig. 8.6b. A third situation
concerns a tray having (on one side) a splash collector with vertical guard screens on all
other sides (Fig. 8.4d). The guards intercept material that would otherwise be deposited
beyond or next to it (cf. Wan et al., 1996). In the case of a square source tray with
collectors on all sides, it is not difficult to see that the amount of material intercepted by
(a) 'Finite screen' (b) 'Infinite screen'
A/B A/B
A/ 1/2 2/3 1/1 3/2 2/1 1/2 2/3 1/1 3/2 2/1 A/
0.1 0.02 0.01 0.01 0.01 0.01 0.24 0.18 0.14 0.09 0.07 0.1
0.2 0.06 0.05 0.04 0.03 0.02 0.42 0.33 0.25 0.16 0.13 0.2
0.5 0.17 0.16 0.13 0.10 0.09 0.72 0.61 0.50 0.34 0.28 0.5
1 0.35 0.33 0.30 0.24 0.21 0.91 0.83 0.73 0.55 0.47 1
2 0.56 0.56 0.53 0.47 0.42 0.99 0.97 0.91 0.79 0.71 2
5 0.80 0.80 0.80 0.78 0.77 1.00 1.00 1.00 0.97 0.95 5
10 0.90 0.90 0.90 0.90 0.90 1.00 1.00 1.00 1.00 1.00 10
25 0.97 0.97 0.97 0.97 0.97 1.00 1.00 1.00 1.00 1.00 25
50 0.99 0.99 0.99 0.99 0.99 1.00 1.00 1.00 1.00 1.00 50
100 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 100
Table 8.1. Geometric correction factors for splash transport measured in tray experiments, for
a range of tray dimensions (A and B) and average splash lengths (). The length A of the
considered side of the tray is made dimensionless through division by average splash length
(), while tray dimensions are expressed in the aspect (A/B). The geometric correction factor
represents measured transport as a fraction of transport from an infinitely large area. Factors
are calculated for the situation of a tray equipped with (a) a very wide screen having the same
length as the side A of the tray and (b) an infinitely large screen.
CHAPTER 8 - EXPONENTIAL SPLASH DISTRIBUTION THEORY
143
one splash collector and its guards should be equal to the sum of elements in the area
denoted SCREEN plus half the elements in the two areas denoted CORNER in Fig.
8.4a. Clearly, this will not change if the other collectors are removed. However, if the
tray is rectangular instead of square the solution becomes less straightforward.
Nevertheless, as long as A/ and B/ are sufficiently large, the rate of transport can be
calculated in the same manner without introducing large errors. Consequently, the
correction factor needed for this type of experiment can be approximated by averaging
(a) Finite screen
0.00
0.25
0.50
0.75
1.00
0.1 1 10 100
A/ %
D
i
m
e
n
s
i
o
n
l
e
s
s

s
p
l
a
s
h

t
r
a
n
s
p
o
r
t
From top to bottom:
A/B=
1/2
2/3
1/1
3/2
2/1
(b) Infinite screen
0.00
0.25
0.50
0.75
1.00
0.1 1 10 100
A/ %
D
i
m
e
n
s
i
o
n
l
e
s
s

s
p
l
a
s
h

t
r
a
n
s
p
o
r
t
From top to bottom:
A/B=
1/2
2/3
1/1
3/2
2/1
Fig. 8.5. Splash transport onto (a) a wide screen having the same length as a side A, and (b) an
infinitely large screen, both placed along a side A of a tray with dimensions A and B, as a
function of the dimensionless length of that side, A/, and for a number of different values of
the aspect (A/B) of the tray. Splash transport was made dimensionless by expressing it as a
fraction of the transport from an infinitely large area of soil.
CHAPTER 8 - EXPONENTIAL SPLASH DISTRIBUTION THEORY
144
the correction factors for the appropriate combination of A/B and A/ values listed in
Tables 8.1a and 8.1b, respectively.
8.5. Splash from or into a cup
In studies with splash cups, it is important to know the relationship between the cup
splash rate m
R
(i.e. the amount of material splashed into or out of a cup of radius R per
unit area of cup, in g m
-2
) and the detachment rate (). In principle, the mass [M(>r) in g
m
-2
] that is splashed beyond a radial distance r of a small unit source area is easily
obtained by integration of Eq. [8.2]:

,
_


,
_

>

r r
r d r M
r
exp exp ) (

[8.13]
However, a circular source area represents a population of individual sources whose
distance to the perimeter is a function of direction (except for the very centre point) and
this complicates the calculation of the distribution of splashed material. Torri and Poesen
(1988) attempted a combined analytical and numerical solution to this problem and
presented a table listing the factors that relate detachment rate to cup diameter and
average splash length. However, similar to the approach followed for the case of a
rectangular soil tray, it may be seen that cup radius R and average splash length () can
be combined into one dimensionless variable R/. Furthermore, the cup splash rate (m
R
)
can be expressed as a function of R/ and detachment rate ():
) / ( R f
m
F
R
R

[8.14]
Because of mathematical reciprocity, Eq. [8.14] is valid for both ejecting and receiving
splash cups. It proved very difficult to evaluate Eq. [8.14] in an analytical manner and
therefore again a numerical model was constructed, similar to the one described for a
rectangular source area in earlier. The representation of a circular area in a matrix
introduced additional numerical errors, but these can be reduced by increasing the radius
of the source area. As a control on these errors, the number of source elements
representing the cup area was compared with the theoretical area (*R
2
) for each model
run. If the difference was less than 0.5% the result was accepted, otherwise the values of
both R and were doubled and the model was run again. The summed amounts of
material deposited outside the circular source area were calculated and divided by the
sum for all elements to determine F
R
.
When plotted on a semi-logarithmic scale, the results indicate an S-shaped function
(Fig. 8.6). For low values of R/, the results approach those of Eq. [8.13] and,
consequently, F
R
approaches unity. For high values of R/ on the other hand, the fact
that the area is not infinitely large becomes progressively less important and eventually
the situation approaches that of splash across the boundary of an (effectively) infinitely
large circular area A
R
. The total length of the boundary equals the circumference of this
circle while transport rate (q) across the boundary is defined by Eq. [8.8]. In that case,
cup splash rate (m
R
) is given by:
CHAPTER 8 - EXPONENTIAL SPLASH DISTRIBUTION THEORY
145
R R
R
R
R q
A
m
F
R
R
R

2
2
1
2 ) (
) (
2 2
[8.15]
Again, this equation only depends on the ratio of R to and not on their absolute values.
Using the boundary conditions stated above, it was found empirically that the model
results were well described by:
R
R m
F
R
R

1
]
1

,
_

2
2
exp 1 [8.16]
Predictions by Eq. [8.16] agreed with the numerical model results within 4% (Fig. 8.6)
and, probably, even a significant part of this difference relates to numerical errors within
the model itself. Predictions of splash transport based on Eq. [8.16] were compared with
the measurements of Poesen and Torri (1988) who used receiving splash cups. As shown
in Fig. 8.7, Eq. [8.16] fits the experimental data slightly better than the exponential
function proposed by the authors themselves. Whereas they inferred a detachment rate of
790 g m
-2
and an average splash length of 0.060 m (Poesen and Torri, 1988; Torri and
Poesen, 1988), Eq. [8.16] yielded values of 950 g m
-2
and 0.035 m, respectively. It
should be noted, however, that the exponential function derived by Torri and Poesen
(1988) is not compatible with the theoretical requirement that m
R
is proportional to 1/R
for large cup sizes.
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
0.1 1 10 100
R/ %
F
R
Fig. 8.6. Relationship between dimensionless cup radius R/ and the amount of soil splashed
out of the cup, expressed as a fraction F
R
of the total amount of detached sediment.
CHAPTER 8 - EXPONENTIAL SPLASH DISTRIBUTION THEORY
146
8.6. Interpretation of splash measurements on a sloping surface
A number of problems arise when the theory developed above is applied to splash on
a slope. For ballistic reasons alone, a particle that is splashed downslope will travel a
longer horizontal distance before meeting the soil surface again, compared with a particle
that is splashed upslope. In principle, the resulting anisotropic distribution of particles
can be calculated using conventional ballistic equations, but this requires knowledge of
the distribution of vertical angles and initial velocities of all particles ejected from the
point of impact, or assumptions about these. Moreover, it needs to be assumed that these
distributions and the amounts of particles splashed do not vary in the different horizontal
directions. These assumptions fail to describe the actual characteristics of splash on a
slope. Laboratory experiments by Ghadiri and Payne (1988) demonstrated that, compared
with splash on a horizontal surface, splash on sloping surfaces:
(i) produced droplets with a lower average vertical splash angle in the downward
direction, and a higher angle in the upslope direction;
(ii) produced droplets with comparable velocity distributions in all directions;
(iii) detached similar amounts of particles as drops falling on a horizontal slope; yet,
(iv) produced more large droplets splashing in the downslope direction, carrying more
detached particles than droplets splashed upslope.
Clearly, extremely detailed information on the distribution of particles in all vertical and
horizontal directions as well as the corresponding initial velocities would be needed to
develop a mathematical framework for splash on a slope. To make matters worse, the
experiments of Ghadiri and Payne (1988) also demonstrated that the effect of slope on
the splash process varies, depending on the material involved. This suggests a wide range
of possible slope - (net) splash transport relationships, and indeed a considerable number
of such relationships has been reported for different soil types (e.g. Foster and Martin,
m
R
= 0.077e
-0.109R
(r
2
=0.83; after Poesen and Torri, 1988)
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0 5 10 15
R (cm)
m
R

(
g

c
m
-
2
)
Eq. [8.16] (r
2
=0.86)
Fig. 8.7. Application of the theory for interpreting splash cup data (Eq. [8.16]) to data
collected by Poesen and Torri (1988). The exponential equation proposed by Poesen and Torri
(1988) is shown for comparison. m
R
=cup splash rate, R=cup radius.
CHAPTER 8 - EXPONENTIAL SPLASH DISTRIBUTION THEORY
147
1969; De Ploey, 1969; Mosley, 1973; Moeyersons and de Ploey, 1976; Quansah, 1981;
Savat, 1981).
An alternative, more pragmatic approach is to use the equations derived earlier for a
horizontal surface to interpret measurements of splash on a slope and to try and obtain an
estimate of the error involved. For example, consider measurements made using a series
of long compartments, placed perpendicularly to the slope gradient, to measure
downslope or upslope splash transport, or along the slope gradient to measure lateral
splash transport. By interpreting these measurements using one-dimensional theory (Eqs.
[8.12a-c]) different apparent average splash lengths will be obtained for each direction in
which transport was measured. This apparent directional splash length represents a mass-
weighted average of the respective average directional splash lengths of particles
splashed in the various horizontal directions. An error is likely to be introduced if these
apparent splash lengths and observed transport rates are used to estimate detachment rate
using Eq. [8.8]. Arguably, the most important source of error lies in the preferential
splash of particles downslope. The order of magnitude of this error can be investigated
analytically. For the special case that all detached particles are splashed downslope, with
equal amounts of particles still being splashed in all downslope directions, it is readily
shown that detachment rate estimated with Eq. [8.8], using measurements of downslope
splash, will result in an over-estimate of twice the actual value. It can also be shown that
detachment rate determined from lateral splash measurements will still be correct. If the
radial distribution of particles is even more asymmetrical than this, the associated error
will be greater. The extreme situation is represented by a situation in which all particles
are splashed at very small angles to the slope gradient: in that case lateral and upslope
transport are virtually and absolutely absent, respectively. Furthermore, it may be
assumed that the distribution function for splash from a narrow strip perpendicular to the
slope gradient, becomes the one-sided function
( )

,
_


x
x M
down
exp

(x >0) [8.17]
where

is the weighted-average downslope splash length, x the distance downslope


from the source area and M
down
(x) (in g m
-3
) the splash mass per unit of distance per unit
source area. Following an approach that is more or less similar to the one described with
regard to the FSDF (but less complex), the relationship between transport rate (q) and
detachment rate () can be evaluated. Eventually, this relationship is given by:

,
_

0 0
'
exp '
x x
dx dx q [8.18]
Results obtained with Eq. [8.18] can be compared with those of Eq. [8.8], but it should be
noted that if splash has indeed become entirely one-dimensional as described by Eq.
[8.17], then interpreting the splash measurements with radial theory will yield an
apparent splash length that differs from the actual average downslope splash length (

)
by a factor of ca. 1.30 (cf. Eqs. [8.14a-c]). Eventually, the maximum relative difference
between apparent detachment rate (
a
) found using Eqs. [8.12a-c] and true detachment
rate () based on Eq. [8.18] is given by:
CHAPTER 8 - EXPONENTIAL SPLASH DISTRIBUTION THEORY
148
42 . 2
30 . 1

q
q
a

[8.19]
It follows that detachment rate estimated from measurements of downslope splash using
the radially isotropic theory for splash on a horizontal plane can be up to 2.42 times
higher than the actual detachment rate (in contrast, measurements of upslope and lateral
splash will yield an under-estimate that in theory can become infinitely large).
Uncertainties from these considerations are less than variations in the detachability of
different soils (defined as the detachment per unit erosive energy input), which cover
several orders of magnitude (Poesen and Savat, 1981; Poesen and Torri, 1988).
Therefore, despite the error involved, measurements of downslope, upslope and lateral
splash transport and the associated apparent splash lengths on a sloping surface may still
be interpreted with the theory developed in this paper, and will result in useful (albeit
initial) estimates of detachment rate.
8.7. Concluding remarks
The present theoretical study shows that experiments to determine detachment and
transport by rain splash must take into account the spatial distribution of the splashed
particles. There is considerable empirical evidence to suggest that an exponential FSDF
adequately describes reality. On the basis of the FSDF, a theory was developed that
facilitates the interpretation of various kinds of splash measurements. Several theoretical
applications of the FSDF compared very well with results of experiments involving
splash cups and splash trays. It is evident from the theory that the rate of splash
detachment () and observed rates of splash transport (q) constitute two different entities.
However, the relationship between the two proved surprisingly simple and only requires
knowledge of the average splash length (). A number of equations and correction
factors are presented that can be used to interpret the results of splash experiments on a
horizontal surface. The distribution of splashed particles is anisotropic on a slope and,
consequently, splash experiments become more difficult to interpret. It was shown that
one-dimensional (i.e. downslope, upslope or lateral) measurements of splash transport
may still be interpreted using the proposed theory, but the magnitude of the uncertainty
involved increases with slope.
The geometry of splash experiments is shown to have a significant effect on the
results. The use of soil trays introduces edge effects and correction factors for these were
proposed. However, the use of these correction factors for tilted trays leads to errors that
increase with slope gradient. In such cases the tray should be sufficiently large to
minimise edge effects. Furthermore, the distribution of splash distances needs to be
known, e.g. through the use of segmented splash collectors, boards or screens (Savat and
Poesen, 1981). Similarly, splash transport rates measurements using receiving splash
cups can only be interpreted correctly if cups of different sizes are used (Poesen and
Torri, 1988). In the case of ejecting cups, the distribution of material splashed from the
cups also needs to be measured, e.g. through the use of a series of concentric ring
collectors (Riezebos and Epema, 1985).
The practical application of the presently proposed theory to field measurements
using splash cups on sub-horizontal terrace beds and splash boards on steep, bare terrace
CHAPTER 8 - EXPONENTIAL SPLASH DISTRIBUTION THEORY
149
risers in Java, Indonesia is reported in Chapter 9, whereas experiments involving soil
trays similar to those used by Wan et al. (1996) are dealt with in Chapter 10.
Appendix 8.A. List of symbols used
A length of side A of a soil tray (m)
A
R
surface area of a cup with radius R (m
2
)
a splash source area for an individual drop impact (m
2
)
B length of side B of a soil tray (m)
F
R
geometrical correction factor for splash from or into a cup of radius R
F
w
geometrical correction factor for an infinitely long tray of width w
K
0
(x) zero-order modified Bessel function
M splash mass (g)
M(>r) splash mass beyond a distance r from point source, per unit source area (g m
-2
)
M
ring
(r) splash mass onto ring at distance r from point source, per unit of distance per unit
source area (g m
-3
)
M
down
(x) splash mass onto a point at distance x from point source, per unit of distance per
unit of source area (g m
-3
)
m
point
(r) splash mass onto a point at distance r from point source, per unit of source area
per unit of target area (g m
-4
)
m
R
cup splash rate out of or into a cup of radius R (g m
-2
)
q splash transport over an imaginary or existing boundary (g m
-1
)
R splash cup radius (m)
r radial distance from splash source (m)
S slope gradient
v
strip
(x) splash mass out of a source strip onto a point at distance x, per unit of source
width per unit of target area (g m
-3
)
v
beyond
(x) splash mass out of a source strip beyond a boundary line at distance x, per unit of
source width per unit of boundary length (g m
-2
)
w width of an infinitely long splash tray (m)
X, x perpendicular distance to a line (m)
(weighted-) average splash length (m)

downslope projected average splash length based on a one-dimensional


distribution (m)
splash detachment rate (g m
-2
)

a
apparent splash detachment rate (g m
-2
)

You might also like