47

Chapter 5
Measurement and modelling of rainfall interception
by a tropical upland mixed cropping system
Abstract: To improve the description of rainfall partitioning by a vegetation canopy that changes
in time a number of adaptations to the revised analytical model for rainfall interception by sparse
canopies (Gash et al., 1995) was proposed in Chapter 4. This chapter presents an application of
this adapted analytical model to simulate throughfall, stemflow and interception as measured in a
mixed agricultural cropping system involving cassava, maize and rice during two seasons of
growth and serial harvesting in upland West Java, Indonesia. Measured interception losses were 18
and 8% during the two measuring periods, while stemflow fractions were estimated at 2 and 4%,
respectively. The main reasons for these discrepancies were differences in vegetation density and
composition, as well as differences in the exposure of the two sites used in the two respective
years. Functions describing the development of the leaf area index of each of the component crops
in time were developed. Leaf area index (ranging between 0.7 and 3.8) was related to canopy cover
fraction (0.41-0.94). Using average values and time series of the respective parameters,
interception losses were modelled using both the revised analytical model and the presently adapted
version. The results indicate that the proposed model adaptations substantially improve the
performance of the analytical model and provide a more solid base for parameterisation of the
analytical model in vegetation of variable density.
Published as: Van Dijk, A.I.J.M., Bruijnzeel, L.A., 2001. Modelling rainfall interception by
vegetation of variable density using an adapted analytical model. 2: Model validation for a
tropical upland mixed cropping system. Journal of Hydrology 247: 239-262.
5.1. Introduction
The current generation of dynamic rainfall interception models, of which the Gash
(1979) and Rutter et al. (1971) models and their sparse canopy derivatives (Gash et al.,
1995; Valente et al., 1997) are the most widely used, assume canopy conditions to be
constant in time. More or less abrupt changes in canopy characteristics due to for example
leaf fall, storm damage or selective logging are usually addressed by simply separating the
data into events before and after the change (Dolman, 1987; Waterloo, 1994; Asdak et al.,
1998). However, there are many situations in which vegetation density changes more
gradually but still relatively rapidly, as with annual agricultural crops (Van Dijk, 1996) or
fast-growing tree plantations (Beadle, 1997). In such cases, a modelling approach which is
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 5 - MEASUREMENT AND MODELLING OF RAINFALL INTERCEPTION
48
capable of describing the changes in canopy characteristics as a continuous function of
time is clearly preferable to having to distinguish a series of consecutive periods, each with
their own constant canopy characteristics (Chapter 4; cf. Carlyle-Moses and Price,
1999).
In Chapter 4, various adaptations to the revised analytical model of Gash et al. (1995)
have been proposed to further improve the description of evaporation from wet vegetation
whose canopy characteristics vary in time. In this chapter the results of an application of
the adapted model are reported, to predict rainfall partitioning during two seasons of
growth and serial harvesting in a mixed agricultural cropping system involving cassava,
maize and rice in upland West Java, Indonesia.
5.2. Study area
The present study was conducted within the 1 km
2
upper catchment of the
Cikumutuk river, situated about 40 km East of Bandung, West Java, near the town of
Malangbong in the middle reaches of the Cimanuk basin at an altitude of 560 to 740 m
a.s.l. (7-03S, 108-04W; Fig. 3.1). The geology consists of Tertiary volcanic
agglomerates overlain by weathered Quarternary tuffs in which Oxisols have developed.
Slopes are generally fairly steep at about 15- and are usually bench terraced. Land use in
the catchment during the years of study was dominated by rainfed mixed crops (ca.62%),
fallow land and grass land on the hillslopes (15%), and paddy rice fields in the valleys
(12%), while the remaining land was occupied by settlements, home gardens and
plantation forest (11%). The area experiences a humid tropical climate with a drier season
(average monthly rainfall less than 60 mm) generally extending from July until September.
Mean annual rainfall is about 2650 mm. Rainfall data collected in the study catchment
between 1994 and 2001 are summarised in Fig. 3.5. Average daily minimum and
maximum temperatures during the wet season are 20.1-C and 27.8-C, respectively, and
Fig. 5.1. Experimental set-up for stemflow measurements in maize. On the left the vase method
used in 1995, on the right the collar method used in 1999.
CHAPTER 5 - MEASUREMENT AND MODELLING OF RAINFALL INTERCEPTION
49
the corresponding average daily minimum and maximum humidity values 56% and 96%.
Incident short-wave radiation during the wet season is around 16 MJm
-2
d
-1
, resulting in an
average daily Penman open water evaporation of 3.8 mm d
-1
. The annual Penman open
water evaporation total is estimated at about 1400 mm (Chapter 6).
5.3. Cropping system
The rainfed crops in the study area usually include cassava (Manihot esculenta
Crantz) and maize (Zea mays L.). Often these are intercropped, sometimes together with
a third crop which may be upland rice (Oryza sativa L.), groundnut (Arachis hipogea L.),
kidney bean (Phaseolus vulgaris L.) or cowpea (Vigna sinensis L.). The cassava and
maize (with or without a third crop) are sown after the rainy season has truly started
(usually in November), while the cassava is planted in relay two or three weeks later.
Sowing rows spaced 60-70 cm apart (30 cm if there is a third crop) are marked on the
terrace bed, orientated perpendicularly to the drain that runs along the foot of the upslope
terrace riser. Two or three maize grains are sown per hole, spaced at about 50 cm,
resulting in planting densities of 3 to 7 plants m
-2
. Where rice is sown as well, this results
in planting densities of 40 to 160 plants m
-2
. Cassava stem cuttings of a few decimetres are
trimmed from leftovers of the previous crop and are planted in alternating rows with the
maize, with typical planting densities ranging between 0.7 and 3 plants m
-2
. Maize will
flower after about 50 days weeks and the cobs start developing a week later. The plants
start wilting after 80 days while the cob continues to ripen. Some of the young cobs are
harvested fresh at this stage for private consumption; the rest is left to mature and
harvested dry after 110 to 120 days. Rice will form panicles after 100 days, which are
harvested a few weeks after the maize, after 120 to 130 days. Cassava starts to form tubers
after about three months. In a mixed cropping system, its growth stagnates somewhat at the
ripening stage of the other crops, which often coincides with infestation by the mosaic virus
(Lozano, 1978; Cock, 1978). The tubers may be harvested at any time, depending on market
prices. However, the bulk is harvested at the end of the dry season, normally in September or
October, just before the land is prepared again and a new crop of cassava is planted (Purwanto,
1999).
5.4. Methods
5.4.1. Gross rainfall, throughfall and stemflow
Measurements of rainfall, throughfall and stemflow were made during the 1994/95
and 1998/99 rainy seasons at two different locations (see Fig. 3.1). Daily gross rainfall
(Pg) was measured at well-exposed locations near the plots (<30 m) using 100 cm
2
orifice
rain gauges placed well above the crop canopy. Rainfall was also measured routinely at a
meteorological station situated on the southern divide of the catchment (630 m a.s.l.; code
CS in Fig. 3.1). Throughfall was measured on a daily basis, using gauges that were
randomly moved each day to minimise errors originating from spatial variability (cf. Lloyd
and Marques-Filho, 1988). The gauges had an orifice of 100 cm
2
placed at about 30 cm
above ground level. During the 1994/95 season, throughfall was measured within the
context of a lysimeter water budget study (Van Dijk, 1996). Throughfall was measured at
CHAPTER 5 - MEASUREMENT AND MODELLING OF RAINFALL INTERCEPTION
50
and around three 1x1 m free-draining lysimeters situated on a single terrace situated just
below the southern water divide, not far from the meteorological station (640 m a.s.l.;
code LS in Fig. 3.1). The lysimeters and their surroundings were planted with
intercropped maize, rice and cassava since 21 November 1994 and throughfall was
measured using six roving gauges from 8 January to 11 May 1995 (48 to 172 days after
sowing). In the 1998/99 rainy season, throughfall measurements were made within the
context of a nutrient budget study on a terrace considered representative of the situation
prevailing on intermediate slopes near the bottom of the catchment (580 m a.s.l.; code
EF in Fig. 3.1) and planted with maize and cassava since 17 November 1998.
Throughfall was measured daily from 2 January to 17 July 1999 (46 to 239 days after
sowing) using ten roving gauges. In neither season could throughfall be measured during
the first seven weeks after sowing, since a significant part of the crop canopy remained
below the level of the throughfall gauges at this initial stage of growth. The use of
ground-level gauges was precluded because of the high intensity of the rains which tends
to move substantial amounts of soil material via splash (Chapter 9).
Initial tests were carried out on cassava and maize to evaluate the importance of
stemflow. Stemflow appeared to be negligible on cassava, but was significant on maize
plants. Subsequently, two approaches were followed to quantify stemflow on maize. From
10 February to 24 March 1995 stemflow was measured on terraces at some tens of metres
distance from the lysimeters; the maize plants were cut at ground level in late morning and
put upright in a container, with the cut end bent and sealed in a plastic bag to prevent
water uptake. The plant and the container were placed back at the original position of the
plant, inserting three small twigs into the space between the stem and the container rim to
ensure that stemflow would not directly flow over the brim of the container (Fig. 5.1). If
there was enough rainfall to produce stemflow, stemflow volume and plant height were
determined the morning after. Fresh plants had to be used each day to avoid effects of
wilting. To avoid this destruction of plants, stemflow was measured in situ on six plants in
the vicinity of the throughfall gauges using small spiral collars from 27 December 1998 to
4 March 1999 (Fig. 5.1). The heights of the plants were monitored throughout the
measurement period. The spiral collars were made of rubber strips and iron wire attached
to the stem at 20 cm above ground level and draining into 1.6-litre containers that were
emptied daily. In both experiments, the capacity of the containers was insufficient to
accommodate storms exceeding 50 mm. Amounts of stemflow in those few cases were
calculated by extrapolation of a regression equation linking stemflow to throughfall (Eq.
[5.6], Section 5.5.2).
5.4.2. Vegetation leaf area index and canopy cover
A number of direct and indirect methods have been advanced to estimate leaf area
index and canopy cover, many of which are discussed in Chason et al. (1991). In the
present study, the development of leaf area index was estimated indirectly from data on
leaf biomass, plant height and planting density, which were being collected as part of a
nutrient cycling study. The methodology is described in detail by Van Dijk (1996), on
which the following summary is based. At different stages during the cropping cycle,
plants were sampled destructively after measuring their height. Height was measured as
the distance between the ground and the highest point of the plant. Next, the various plant
parts (including the leaves) were air-dried in a greenhouse for a few weeks and weighed to
CHAPTER 5 - MEASUREMENT AND MODELLING OF RAINFALL INTERCEPTION
51
the nearest 0.1 g, after which a sub-sample was dried to constant weight at 80-C at the
field laboratory for conversion to oven-dry weight. Separate relationships between plant
height and leaf biomass were derived for cassava, maize and rice. Values of the specific
leaf area (ASL in m
2
kg
-1
) of the respective species were determined on a representative
number of fresh leaves (typically 30 on each sample occasion). The surface area of the
leaves was calculated from measurements of length and width at regular intervals for
leaves of simple geometry (maize and rice) and through planimetry and digital analysis for
the compound cassava leaves. Leaves of known surface area were dried to constant
weight in an oven at 80-C and weighed to the nearest mg. Whilst it is acknowledged that
the specific leaf area of a leaf may change during the course of its life (Beadle, 1997), it
was assumed as a working hypothesis that this had a negligible influence on the estimation
of total leaf area index.
Next, the measurements of crop height (h), oven-dry leaf biomass (mL) and specific
leaf area (ASL) were used to establish relationships between individual crop height and leaf
area (AL) for each crop type. Finally, measurements of planting density (n) and average
crop height (h) were used to estimate leaf area index L at regular intervals throughout the
cropping cycle according to:
) (h f n nA L
L
[5.1]
where f(h) denotes the relationship found between individual plant height and leaf area for
each crop type. Weed growth was limited because of regular weeding and no attempts
were undertaken to include weed leaf area in total leaf area index estimates.
Canopy cover c was related to leaf area index L. For this purpose, photographs were
taken on a regular basis near the experimental site from above the canopy of crops of
known but slightly different height and planting density between November 1998 and May
1999 (N=10). The photographs were scanned and analysed using image processing
software to derive canopy cover fractions. L values of the vegetation estimated with Eq.
[5.1] were related to corresponding canopy cover values using the theory outlined in
Chapter 4 and summarised by Eq. [5.2]:
L
e c

1 [5.2]
where

is a dimensionless extinction coefficient (see Chapter 4 for details).

5.4.3. Vegetation storage capacity
An indication of the water retention capacity of the crops under consideration was
obtained through simple laboratory experiments comparable to those described by Liu
(1998). Single plants of cassava, maize and rice were cut at ground level and taken to the
field laboratory to be wetted in three different ways. Firstly, vertical rainfall was simulated
by spraying the plants from above using a vaporiser. Secondly, the total water retention
capacity was estimated (i) again using a vaporiser but spraying from all sides; and (ii) by
momentarily immersing the plant, keeping it upright when possible or else keeping the cut
end above the water to prevent water uptake and carefully turning it upright after
immersion. Water of rainfall quality (electrical conductivity <10 S cm
-1
) and ambient
temperature (ca. 20 -C) was used in all experiments. The total amount of water retained
CHAPTER 5 - MEASUREMENT AND MODELLING OF RAINFALL INTERCEPTION
52
by the plant was determined by weighing the plants (to the nearest 0.001 or 0.1 g,
depending on plant size) just before wetting and after drainage from the wetted plants had
just ceased. To estimate the water retention capacity per unit surface area, it was assumed
that the stems of the plants retained the same depth of water on their surface as the leaves.
Subsequently, leaf surface area was estimated by drying and weighing the leaves, while the
surface of the stems was estimated by measuring stem circumference at the base and the
top of the stem and assuming a tapering geometrical form. The volume of water retained
by the plant was divided by the total surface area of the leaves (approximated as two times
the one-sided leaf area) and stems. Of each species, between 8 and 21 individual plants
were tested.
5.5. Results
5.5.1. Leaf area index and canopy cover development
For each of the three crops, relationships between plant height and the mass of living
leaves were obtained. Values of specific leaf area ASL were 21.1, 24.7 and 22.9 m
2
kg
-1
for
maize, rice and cassava, respectively. The respective regression equations linking plant leaf
area to height are shown Fig. 5.2. For maize and rice, a simple exponential function described
the relation between height and leaf area best (Fig. 5.2a and b). However, it appeared from the
height and biomass data as well as from observations in the field that the cassava plants entered
a stage (coinciding with tuber formation) in which vertical growth hardly resulted in a further
increase in leaf area. The relationship between height and leaf area was consequently best
described by an S-shaped exponential growth curve (Fig. 5.2c).
Combining the above relations with height and planting density measurements made on
the experimental plots, curves describing the development of L in time were constructed. Leaf
area index started to decrease shortly after the maize and rice crops reached their maximum
height because of yield formation and subsequent wilting. The remaining fraction of living
leaves was estimated by comparing the numbers of living and wilted leaves. To improve
estimates of leaf area index, the number of (living) leaves was also monitored on 18 cassava
plants after the other crops were harvested.
To describe the development of leaf area index in maize and rice with time, sinoid curves
were fitted to the L data, having the following form (with the argument of the cosine expressed
in radians; Van Dijk, 1996):

1
]
1

,
_

1 1 cos
2
max
max
t
t L
L
t
[5.3]
where Lt is the leaf area index at t days after sowing, Lmax the maximum leaf area index,
reached at time tmax, and a fitting parameter. Different values of where necessary for
t<tmax and ttmax, denoted by 1 and 2, respectively (Table 5.1).
In cassava, on the other hand, the development of L showed an S-shaped increase
towards a maximum leaf area index, which was best described by a logistic growth curve
of the form:
[ ]
1
2 1 max
) exp( 1

+ t g g L L
t
[5.4]
CHAPTER 5 - MEASUREMENT AND MODELLING OF RAINFALL INTERCEPTION
53

0.00
0.02
0.04
0.06
0.08
0.0 0.5 1.0 1.5 2.0
h (m)
A
L

(
m
2
)
(a) Rice
A
L
= 0.0238 h
2.31
r
2
= 0.97
N = 34
0.0
0.2
0.4
0.6
0.8
1.0
0.0 0.5 1.0 1.5 2.0 2.5
h (m)
A
L

(
m
2
)
(b) Maize
A
L
= 0.224 h
1.57
r
2
= 0.90
N = 50
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
0.0 0.5 1.0 1.5 2.0
h (m)
A
L

(
m
2
)
(c) Cassava
A
L
= 0.666 [1-exp(-2.9 h )]
5.7
r
2
= 0.58
N = 68
Fig. 5.2. The relationships between plant height (h) and leaf area (AL) for (a) maize, (b) rice,
and (c) cassava in the study area.
CHAPTER 5 - MEASUREMENT AND MODELLING OF RAINFALL INTERCEPTION
54
where g1 and g2 are fitting parameters (see Table 5.1 for values). The accuracy of the
estimated leaf biomass values could be checked for maize and rice in 1995 by drying and
weighing the leaves of the maize and rice plants after harvesting. Eq. [5.3] turned out to
be quite accurate, the difference between measured and predicted values being less than
5% (Van Dijk, 1996). It remains uncertain how accurate the estimation of the leaf area
index for cassava is, because it was not harvested within the measuring period. In view of
the good results obtained for maize and rice, however, and because the number of cassava
leaves was also monitored as an additional control, Eq. [5.4] is expected to be sufficiently
reliable as well. Fig. 5.3 shows the development of component and cumulative leaf area
index before and during the two measuring campaigns in 1994/95 and 1998/99.
The surface area of cassava and maize stems was also estimated. From measurements
made on a selected number (N=26) of maize plants, a highly significant relationship (r
2
=0.99)
was found between plant height and stem surface area (as calculated from circumference
measurements at the top and base of the stems). Comparing the stem surface areas with
corresponding leaf surface areas suggested that stem surface area (per unit ground area) was
approximately 10% of the leaf area index. It should be noted that unlike the stem surface, leaf
area index involves only one side of the leaves and therefore represents only half of the total
leaf surface area. The stem surface fraction of total plant surface area (including leaves and
stem) is therefore ca. 5% for maize. Using a similar approach for rice and cassava, the
respective fractions were estimated to be 5% and 3%, respectively, including twigs and
branches.
The leaf area index versus canopy cover fraction data collected during the 1998/99 season
near the experimental site were used to derive a relationship between the two having the form
of Eq. [5.2] (Fig. 5.4). The best fit was achieved with an extinction coefficient

of 0.75
(r
2
=0.96). This value is intermediate to values reported for maize (0.64 for PAR; Sivakumar
and Virmani, 1984) and cassava (0.72 to 0.88; Veltkamp, 1986). The combined effect of the
two crops is illustrated further by the fact that points representing vegetation stages dominated
by maize invariably lie below the fitted line in Fig. 5.4, while those representing cassava lie
above it.
1994/95 1998/99
Maize Rice Cassava Maize Cassava
Density, n (m
-2
) 4.0 45 2.5 4.1 1.1
Maximum height, hmax (m) 1.5 1.1 1.2 2.2 1.5
Max. plant leaf surface, AL,max (m
2
) 0.43 0.033 0.56 0.73 0.64
Number of days until max. L (tmax) 80 110 n.a. 83 n.a.
1, g1 2.2 2.4 5.9 2.8 5.0
2, g2 8 200 0.09 5 0.09
Maximum leaf area index, Lmax 1.7 1.5 1.4 3.0 0.7
Table 5.1. Leaf area index model parameters used to characterise the vegetation of the study
plots.
CHAPTER 5 - MEASUREMENT AND MODELLING OF RAINFALL INTERCEPTION
55

0
1
2
3
4
0 30 60 90 120 150 180 210 240
DAS
L
cassava
rice
maize
(a)
0
1
2
3
4
0 30 60 90 120 150 180 210 240
DAS
L
cassava
maize
(b)
Fig. 5.3. Estimated leaf area index (L) development during the first 8 months of the cropping
cycle in (a) the 1994/95 season and (b) the 1998/99 season. The periods of throughfall
measurements are indicated on the time axis, in days after sowing (DAS).
0.0
0.2
0.4
0.6
0.8
1.0
0 1 2 3 4
L
c
c =1-e
- 0.75 L
r
2
=0.96; N=10
Fig. 5.4. Relationship between leaf area index (L) and canopy cover fraction (c) for a maize-
cassava cropping system in 1998/99.
CHAPTER 5 - MEASUREMENT AND MODELLING OF RAINFALL INTERCEPTION
56
5.5.2. Rainfall partitioning by mixed crops
Measured rainfall and throughfall totals for the two observation periods in 1995 and
1999 are summarised in Table 5.2. Whilst the overall rainfall totals for the two periods
were similar at 1597 and 1642 mm, respectively, the observation period in 1999 lasted
more than ten weeks longer than that in 1995. The average rainfall depth per event in
1995 (15.2 mm) was slightly more than in 1999 (14.4 mm), whereas the observed range
was also higher in 1995 (maximum rainfall depths of 110 and 54 mm, respectively). In
addition, the 1999 observation period had rain on 58% of the days versus 83% in 1995.
Rainfall intensity data were collected between 7 November 1998 and 11 April 1999 from
which an average rainfall intensity of 4.3 mm h
-1
was derived. Only scattered rainfall
intensity records were available for 1995. Therefore, the 1998/99 intensity-duration data
were regressed against corresponding rainfall depths per storm and daily totals measured
in 1995 inserted in the resulting equation (r
2
=0.61) to estimate average rainfall intensity
and (total) duration. Average rainfall intensity in 1995 derived in this way was slightly
higher than that observed in 1999 but average storm duration was the same in both
periods (3.3 h, cf. Table 5.2)
Interception (Ei) was calculated from daily measurements of incident rainfall (Pg),
throughfall (Tf) and stemflow (Sf) according to:
1995 1999
Experimental design
Location near divide near valley bottom
Number of gauges 6 10
Throughfall measuring period 8 Jan. 11 May, 1995 2 Jan. 17 July, 1999
Rainfall characteristics
Gross rainfall, Pg (mm) 1,577 1,642
Number of days 124 196
Number of storms (% of days) 103 (83%) 114 (58%)
Average storm size (mm) 15.2 14.4
Average storm duration (h) 3.3 3.3
Avg. rainfall intensity, R (mm h
-1
) 4.7 4.3
Partitioning (in mm and as % of Pg)
Throughfall, Tf 1,255 (79.6%) 1,446 (88.1%)
Stemflow, Sf 37 (2.4%) 64 (3.9%)
Interception loss, Ei 284 (18.0%) 132 (8.0%)
Vegetation characteristics
Crops maize, rice, cassava maize, cassava
Leaf area index L, avg. and range 2.0 (1.1-3.8) 1.3 (0.7-3.6)
Canopy cover c, avg. and range 0.72 (0.55-0.94) 0.55 (0.41-0.93)
Table 5.2. Summary of rainfall partitioning by mixed crops during the measurement periods in
1994/95 and 1998/99, along with basic rainfall and vegetation characteristics (see text for
explanation).
CHAPTER 5 - MEASUREMENT AND MODELLING OF RAINFALL INTERCEPTION
57
Ei = Pg - Tf - Sf [5.5]
Measurements of stemflow on maize were initiated within a month after throughfall
measurements started in 1995. Also, stemflow measurements were not always successful
because some of the collecting containers fell over or did not have enough capacity. Finally, a
fairly small number of plants was used for the measurements, which affected the reliability of
average stemflow amounts for individual storms. For these reasons it was decided to model
stemflow depths on a storm basis for both seasons by relating measured stemflow volumes of
individual plants to corresponding leaf areas and throughfall depth. Stemflow was divided
by plant leaf area (estimated in turn from height measurements; cf. Fig. 5.2) and the
resulting average value per storm was regressed against measured throughfall. Because
the data sets of 1995 and 1999 were not statistically different they were combined. The
resulting regression equation is shown in Fig. 5.5 and is given by:
Tf A V
L Sf
054 . 0 (r
2
=0.71, N=73) [5.6]
where VSf denotes stemflow volume (in litres) of a plant with leaf area AL (in m
2
). Using
the time series of modelled leaf area index values for the maize, Eq. [5.6] was used to
estimate stemflow depth, and ultimately interception (using Eq. [5.5]), on a storm to
storm basis. The resulting stemflow totals amounted to 2.4% and 3.9% of incident rainfall
in the 1995 and 1999 measuring periods, respectively (Table 5.2). Interception in 1995
was 18.0% of incident rainfall or an average 2.8 mm per storm. Conversely, in 1999 this
was only 8.0% or an average 1.2 mm per storm. The reasons for this marked difference
will be discussed in Section 5.7.
0.001
0.01
0.1
1
10
0.1 1 10 100
Tf (mm)
V
S
f

/

A
L

(
m
m
)
V
Sf
/ A
L
= 0.054 Tf
Fig. 5.5. Stemflow volume VSf (litres) per unit leaf area AL (m
2
) in maize as measured using the
vase method in 1995 (solid dots) and the collar method in 1999 (open circles) versus
corresponding amounts of throughfall Tf (mm).
CHAPTER 5 - MEASUREMENT AND MODELLING OF RAINFALL INTERCEPTION
58
5.5.3. Vegetation storage capacity
When determining the maximum volume of water that could be retained by a plant, it
did not matter whether it was wetted by spraying from all sides or by immersion.
However, spraying from above produced consistently lower values. Furthermore, it was
observed in the experiments (as well as in the field) that the junctions between the stem,
branches and leaves were important places for water storage. Expressed per unit total
plant area, i.e. double-sided leaf area plus stem area, the average depths of water retained
on maize plants (N=10) were 0.033 (0.015) mm and 0.044 (0.024) mm after vertical
spraying and total wetting, respectively. For cassava (N=8) these values were 0.033
(0.017) mm and 0.065 (0.020) mm and for rice 0.040 (0.009) and 0.044 (0.009) mm
(N=21), respectively. These results were used as an estimate of specific leaf storage
capacity SL, i.e. the water retention capacity per unit leaf area, as well as estimates of the
storage capacity per unit stem surface, Ss.
5.6. Modelling rainfall interception by mixed crops with variable canopy characteristics
5.6.1. Models used and their parameterisation
The daily interception values derived from throughfall measurements and stemflow
estimates were used to test the performance of three interception models of increasing
complexity:
1. a simple linear regression approach;
2. the revised analytical model for a sparse but constant canopy (Gash et al., 1995);
3. the adapted analytical model for vegetation of variable density (Chapter 4).
The linear regression approach (1) involved the derivation of a simple linear relation
between gross rainfall and interception. The reformulated analytical model (2) of Gash et
al. (1995) was applied in two ways: (i) using vegetation structure and evaporation
parameter values derived from conventional regression methods (Jackson, 1975; Gash and
Morton, 1978; Gash 1979), and (ii) using an estimate of canopy storage capacity (S)
derived from the laboratory wetting experiments (Section 5.5.3) and optimising the ratio
of average wet canopy evaporation rate over average rainfall intensity ( E / R ) by least
squares (Marquardt, 1963). The adapted analytical model (3) was tested in three ways: (i)
using parameters based on the regression methods described below; (ii) assuming that
E / R is proportional to canopy cover fraction c (cf. Gash et al., 1995) and again
optimising the value of ( E / R )c by least squares; and (iii) non-linear optimisation of both
the energy exchange parameter

(relating actual E / R to L) and the maximum ratio of

wet canopy evaporation rate over rainfall rate ( E / R )a (cf. Chapter 4). These exercises
resulted in a total of six model configurations. The performance of each model
configuration was tested for both data sets (1995 and 1999) using the method described
by Nash and Sutcliffe (1970), to give a value of the so-called model efficiency (ME)
which is defined in an analogous manner to the coefficient of determination of a linear
regression (r
2
) as:
2
0
2 2
0
F
F F
ME

[5.7]
CHAPTER 5 - MEASUREMENT AND MODELLING OF RAINFALL INTERCEPTION
59
where F0
2
represents the initial variance in the observed (or estimated in this case, since
interception losses were calculated using stemflow estimates) values and F
2
the residual
variance, i.e. the sum of squared differences between observed and modelled values.
It should be stressed that interception was estimated using a variable stemflow
fraction that was linked to the leaf area index of the maize (cf. Eq. [5.6]). Normally, this
kind of information will not be available when the simple linear regression technique or the
(revised) analytical model is adopted. However, to enable a better comparison of the three
approaches, the use of a variable rather than a constant stemflow fraction was preferred,
because the latter would have resulted in very different, and for certain periods even
negative, interception estimates. The values of all the parameters used in the modelling are
presented in Tables 5.3, 5.4 and 5.5 whereas the indicators of the performance of the
respective models are given in Table 5.4.
The simple linear regression approach, which links daily throughfall or interception to
incident rainfall, yields two parameters that are usually interpreted as indicators of,
respectively, the canopy storage capacity (S, Gash and Morton, 1978) and the relative wet
canopy evaporation rate ( E / R , Gash, 1979; Fig. 5.6a and b). The Gash and Morton (1978)
approach involves plotting rainfall data in excess of the amount of rainfall necessary to
saturate the canopy (Pg) against throughfall. A straight line with a slope of unity minus
the average stemflow fraction (1-pt) is drawn through the data in such a way that it passes
through those points that are assumed to represent conditions of minimum evaporation
(i.e. the highest points). The negative intercept with the y-axis then gives the value of S.
As shown by Fig. 5.6c and d, application of this method produced a positive intercept
(and hence a negative value of S), for both the 1995 and 1999 data.
Alternative regression techniques to evaluate S (Jackson, 1975; Rowe, 1983) suffered
Parameter Regression equation r
2
Fig. Storm size
class
N Value Model
Both seasons
ps Sf / L = 0.054 Tf 0.71 (5.5) All storms 73 0.054 L 1,2,3
1995
R E Ei = 0.193 Pg - 0.20 0.70 (5.6a) Pg 101 0.193 1,2
S Tf = 0.976 Pg + 0.25 n.a. (5.6c) Pg 101 -0.25 1,2
p Tf = 0.992 Pg 0 (5.6e) 0<Pg<1 mm 14 0.992 2
a
R E Ei / c = 0.234 Pg + 0.0 0.73 (5.6g) Pg 101 0.234 3
1999
R E Ei = 0.069 Pg + 0.16 0.18 (5.6b) Pg 114 0.069 1,2
S Tf = 0.961 Pg + 0.20 n.a. (5.6d) Pg 114 -0.20 1,2
p Tf = 0.886 Pg -0.06 (5.6f) 0<Pg<1 mm 16 0.886 2
a
R E Ei / c = 0.105 Pg + 0.68 0.11 (5.6h) Pg 114 0.105 3
Table 5.3. Regression equations used for the estimation of the vegetation structural parameters
p and S and the relative wet evaporation rate E / R in the respective versions of the analytical
model of interception (see text and model descriptions in Chapter 4 for explanation). Model
numbers refer to: (1) simple linear regression, (2) revised analytical model, and (3) adapted
analytical model, respectively.
CHAPTER 5 - MEASUREMENT AND MODELLING OF RAINFALL INTERCEPTION
60
from the same problem. Therefore, the averaged values of the specific canopy capacity
(SL) for the respective crops (0.077, 0.042 and 0.049 mm for maize, rice and cassava,
respectively; Section 5.5.3) were multiplied by the corresponding average values of L over
the measurement periods to yield overall estimates of S of 0.12 mm (1995) and 0.09 mm
(1999).
A second vegetation structural parameter, the free throughfall coefficient p may be
obtained from the slope of the regression between throughfall and rainfall less than Pg
(Jackson, 1975). The results obtained in this way for the crops under consideration are
presented in Fig. 5.6e and f. Storms with less than 1 mm were used because only two events
with Pg<Pg (0.17-0.19 mm) were available per data set. As shown in Fig. 5.6e and f, amounts
of throughfall were often higher than incident rainfall, with throughfall even being recorded
on a number of occasions in the absence of rain. The latter phenomenon almost certainly
reflects the deposition of dew. This resulted in unrealistically high values of p. For this reason,
p values were estimated instead from average values of canopy cover, c, for the respective
measuring periods, as p=1-c (see below).
The fraction of gross rainfall going to the stems after the canopy is saturated, pt (sensu
Gash et al., 1995) was assumed to equal the average stemflow fraction over the respective
periods (0.024 and 0.039 for 1995 and 1999, respectively). For the storage capacity of the
stems (St) the average of the values used in the adapted model was taken, amounting to 0.0044
and 0.0036 mm in the respective periods (see below).
Model (mode) S, cSc, Sv
R E E /
a c R ,
*

Model
efficiency
Eicum
ME mm (%)
1994/95
Linear regression (1) -0.20 0.193 - - 0.70 0 (0%)
Revised (2-i) 0.12 0.193 0.264 - 0.72 + 44 (15%)
Revised (2-ii) 0.12 0.183 0.250 - 0.72 + 27 (10%)
Adapted (3-i) 0.05-0.24 0.13-0.22 0.234 0.75 0.81 + 14 (5%)
Adapted (3-ii) 0.05-0.24 0.13-0.22 0.235 0.75 0.81 + 15 (5%)
Adapted (3-iii) 0.05-0.24 0.10-0.25 0.382 0.27 0.82 - 2 (-1%)
1998/99
Linear regression (1) 0.16 0.069 - - 0.18 0 (0%)
Revised (2-i) 0.09 0.069 0.125 - 0.31 + 2 (2%)
Revised (2-ii) 0.09 0.069 0.124 - 0.31 + 1 (1%)
Adapted (3-i) 0.04-0.27 0.04-0.10 0.105 0.75 0.34 - 5 (-4%)
Adapted (3-ii) 0.04-0.27 0.04-0.10 0.108 0.75 0.34 - 2 (-1%)
Adapted (3-iii) 0.04-0.27 0.05-0.09 0.089 1.32 0.34 + 2 (2%)
* Value denotes E /
c R in revised analytical model and E /
a R in adapted analytical model.
Table 5.4. Summary of main parameter values and performance statistics for the respective
model versions and fitting procedures. Parameter values derived by non-linear optimisation are
given in bold (see text for explanation).
CHAPTER 5 - MEASUREMENT AND MODELLING OF RAINFALL INTERCEPTION
61

(a) Ei = 0.193 P
g
- 0.20
r
2
= 0.70
-5
0
5
10
15
20
25
30
0 20 40 60 80 100 120
P
g
(mm)
E
i

(
m
m
)
(b) Ei = 0.069 P
g + 0.16
r
2
= 0.18
-5
0
5
10
15
20
25
30
0 20 40 60 80 100 120
P
g
(mm)
E
i

(
m
m
)
0
2
4
6
8
10
0 2 4 6 8 10
P
g
- P
g
' (mm)
T
f

(
m
m
)
(c) Tf = 0.976 P
g
+ 0.25
0
2
4
6
8
10
0 2 4 6 8 10
Pg - Pg ' (mm)
T
f

(
m
m
)
(d) Tf = 0.961 Pg + 0.20
(e) Tf = 0.992 P
g
r
2
= 0
0.0
0.2
0.4
0.6
0.8
1.0
1.2
0 0.2 0.4 0.6 0.8 1
P
g
(mm)
T
f

(
m
m
)
(f) Tf = 0.886 P
g
r
2
= -0.06
0.0
0.2
0.4
0.6
0.8
1.0
1.2
0.0 0.2 0.4 0.6 0.8 1.0
P
g
(mm)
T
f

(
m
m
)
(g) Ei / c = 0.234 P
g + 0.00
r
2
= 0.73
-5
0
5
10
15
20
25
30
0 20 40 60 80 100 120
P
g
(mm)
E
i

/

c

(
m
m
)
(h) Ei / c = 0.105 P
g + 0.68
r
2
= 0.11
-5
0
5
10
15
20
25
30
0 20 40 60 80 100 120
P
g
(mm)
E
i

/

c

(
m
m
)
Fig. 5.6. Determination of the various parameters for mixed crops used in the three tested
models for the 1995 and 1999 periods. Determination of: (a,b) relative average evaporation rate
according to the method of Gash (1979) for the revised model; (c,d) canopy capacity S
according to the method of Gash and Morton (1978); (e,f) the free throughfall coefficient p
according to the method of Jackson (1975); and (g,h) the maximum relative wet canopy
evaporation rate for the adapted model analogous to the method of Gash (1979).
CHAPTER 5 - MEASUREMENT AND MODELLING OF RAINFALL INTERCEPTION
62
The relative evaporation rate, E / R , was determined using the method of Gash (1979),
i.e. by regressing estimated interception values against gross rainfall (Table 5.4, Fig. 5.6a
and b). Values of 0.193 and 0.069 were obtained for the 1995 and 1998 data sets,
respectively.
Parameters needed for the revised analytical model (Gash et al., 1995) include
canopy cover (c), the storage capacity per unit area of canopy cover (Sc ), and the relative
evaporation rate per unit canopy cover (
c E / R ). Normally, the latter two are obtained by
simply dividing the values of S and E / R , as derived via the regression methods described
earlier, by canopy cover fraction c. However, because of the previously mentioned
problems with the derivation of S from throughfall measurements, the alternative estimates
of S (derived in turn from measurements of SL and L) were used instead. Average c-values
of 0.73 and 0.55 were derived from canopy cover time series for the 1995 and 1999
measuring periods, respectively (see below), and used for all storms, regardless of the
growth phase of the crops. This yielded a value of 0.16 mm for Sc in both periods. Values of
c E / R determined by dividing the above-mentioned values of E / R by canopy cover c
amounted to 0.264 and 0.125, respectively.
The adapted analytical model requires a continuous representation of leaf area index,
L, as well as values for the extinction coefficient (

), the specific canopy storage capacity

(SL) and the parameter representing maximum relative evaporation rate from a variable
canopy (
E / R ) together with the energy exchange coefficient

(see Chapter 4 for

details). Time series of L were constructed for both periods using the models presented in
Section 5.5.1 (Eqs. [5.3] and [5.4]). This resulted in L values between 1.1 and 3.8 (2.1 on
average) in the 1995 period and between 0.7 and 3.6 (1.3 on average) in the 1999 period.
Canopy cover was estimated from these data, using the relationship between L and c that was
established for this type of vegetation (Eq. [5.2], with

=0.75). Resulting canopy cover

fractions varied between 0.55 and 0.94 (0.73 on average) in the 1995 period, and between
0.41 and 0.93 (0.55 on average) in the 1999 period. Vegetation storage capacity Sv was
derived by summation of canopy capacity S and stem storage capacity Ss. Canopy capacity was
calculated as the product of leaf area index L and specific canopy storage capacity SL of the
component crops (the latter being 0.077, 0.042 and 0.049 mm for maize, rice and cassava,
respectively). Similarly, storage capacity of the stems St was estimated from time series of L,
stem surface area fractions (0.05, 0.05 and 0.03, for maize, rice and cassava, respectively;
Section 5.5.1) and the SL values of the different crops (cf. Section 5.5.3).
With regard to the representation of wet evaporation from the changing canopy, the
model was run in three modes. In the first, it was assumed that the exchange coefficient

was identical to the extinction coefficient

. As shown in Chapter 4, the relative

evaporation rate is directly proportional to canopy cover (cf. Gash et al., 1995) in this
special case. Therefore, the parameter
a E / R essentially represents
c E / R in the revised
analytical model of Gash et al. (1995). Consequently, it can be obtained using the method
described earlier for deriving
c E / R , i.e. by regression of estimated interception values
against gross rainfall. The difference, however, is that daily interception values are no
longer divided by an overall average (i.e. constant) value of c, but rather by variable
canopy cover depending on vegetation density. Values of
a E / R determined in this manner
are 0.234 and 0.105 for 1995 and 1999, respectively (Table 5.3, Fig. 5.6g and h). In a
second mode, the value of
a E / R was optimised by the method of least squares
(Marquardt, 1963). This resulted in slightly higher optimum values of 0.235 and 0.108,
respectively. Finally, the values of both the exchange coefficient

and the parameter

a E / R
were optimised by iteration. For the 1995 period, this resulted in values for

and
a E / R of
CHAPTER 5 - MEASUREMENT AND MODELLING OF RAINFALL INTERCEPTION
63
0.27 and 0.382, whereas for the 1999 period the best fitting parameter values were

=
1.32 and
a E / R = 0.089 (Table 5.4).
5.6.2. Model performance and parameter sensitivity
The results of the modelling exercises are summarised in Table 5.4. Optimising the value
of
c E / R in both the revised and the adapted analytical interception models, rather than
using the value derived from the regression between gross rainfall and interception, in
most cases resulted in a slightly better model performance in terms of cumulative
modelled interception, but had negligible effect on model efficiency. Optimising the values
of both
a E / R and

somewhat improved the performance of the adapted model for the

1995 data set but had very little effect for the 1999 data set. Furthermore, the resulting
1995 1999
Revised Adapted Revised Adapted
Vegetation parameters
Leaf area index L 2.1 1.1-3.8 1.3 0.7-3.6
Canopy cover c 0.73 0.55-0.94 0.55 0.41-0.93
Free throughfall coefficient p 0.27 0.06-0.45 0.45 0.07-0.59
Stemflow fraction (pt)
*
0.024 0-0.15 0.039 0-0.15
Canopy capacity S (mm) 0.12 0.05-0.24 0.09 0.04-0.27
Stem storage capacity St (mm) 0.0044 0.002-0.0096 0.0036 0.001-0.012

Meteorological parameters
E (mm h
-1
)
0.86 0.60-1.04 0.30 0.19-0.44
R (mm h
-1
)
4.7 4.7 4.3 4.3
Interception components (mm)
For storms Pg<Pg :
Evaporation from canopy 0.1 (0.03%) 0.2 ( 0.07%) 0.1 (0.07%) 0.4 (0.3%)
For storms PgPg :
Wetting up of canopy 14.3 (4.6%) 13.3 (4.4%) 10.7 (8.1 %) 12.2 (9.6%)
Ei from saturated canopy 284.2 (91.2%) 263.6 (8.2%) 111.5 (84.0%) 100.9 (77.6%)
Ei from canopy after storm 12.4 (4.0%) 12.0 4.0 %) 10.1 (7.6%) 11.5 (8.8%)
Evaporation from stems 0.5 0.1%) 10.3 3.5%) 0.4 (0.3%) 5.1 (3.9%)
Total modelled interception loss 311.6 299.1 132.7 130.1
Total estimated interception loss 284.2 284.2 131.9 131.9
Deviation (%) 9.6% 5.2% 0.6% -1.4%
Total modelled stemflow 37.4 36.8 63.6 64.2
Total estimated stemflow 37.2 37.2 64.1 64.1
Deviation (%) 0.5% -1.1% -0.8% 0.2%
* In the case of the adapted model, stemflow fraction (pt) is expressed here as the range of estimates of
stemflow fraction (as Sf/Pg) for separate events, based on the leaf area index (L) time series for maize.
Table 5.5. Components of interception loss for mixed cropping in 1995 and 1999, according to
the revised (Gash et al., 1995) and the adapted (Chapter 4) analytical models with only the
relative wet evaporation rate optimised.
CHAPTER 5 - MEASUREMENT AND MODELLING OF RAINFALL INTERCEPTION
64
values for

differed strongly between the two applications at 0.27 and 1.32 for the 1995
and 1999 data, respectively. This discrepancy and the minor effect on model efficiency
(Table 5.4) provide little argument to deviate from the assumption of a direct relationship
between canopy cover and relative wet canopy evaporation rate, as postulated by Gash et
al. (1995). Therefore, the corresponding mode of model application was preferred and

set at 0.75.
In terms of model efficiency, the best performing model was the adapted Gash model in
both cases (Table 5.4). Based on the close agreement between total estimated and predicted
interception losses, it would seem that the revised analytical model of Gash et al. (1995)
performed slightly better on the 1999 data. However, using only the agreement between
cumulative values as an indicator of model performance can be misleading as becomes evident
from Fig. 5.7. Thus, while estimated and modelled total interception over the 1999 period
agree well for both the (optimised) revised and adapted analytical model (+1% versus +2%;
Table 5.4), the adapted model describes the seasonal cumulative pattern much better than the
revised model. The same is true for the 1995 period (Figs. 5.7c and d).
The parameters of the revised and adapted models are presented again in Table 5.5,
together with the calculated components of interception loss. It appears that the magnitude of
the interception components is similar for both models, with the exception of evaporation from
the stems, which is calculated as 3.5-3.9% of total interception losses in the adapted model,
versus 0.1-0.3% in the revised analytical model of Gash et al. (1995). This is not surprising,
because the evaporation rates per unit surface area were assumed equal for both the leaves and
the stems in the adapted model. Consequently, the interception losses from the leaves and
stems are directly proportional to their respective surface areas (Chapter 4). Conversely, the
0
100
200
300
30 60 90 120 150 180 210 240
DAS
C
u
m
u
a
l
t
i
v
e

E
i

(
m
m
)
(a)
0
100
200
300
30 60 90 120 150 180 210 240
DAS
C
u
m
u
l
a
t
i
v
e

E
i

(
m
m
)
(b)
0
100
200
30 60 90 120 150 180 210 240
DAS
C
u
m
u
l
a
t
i
v
e

E
i

(
m
m
)
(c)
0
100
200
30 60 90 120 150 180 210 240
DAS
C
u
m
u
l
a
t
i
v
e

E
i

(
m
m
)
(d)
Fig. 5.7. Cumulative estimated (thick dotted) and modelled (thin line) interception according to
(a, c) the revised analytical model (Gash et al., 1995) and (b, d) the adapted version (Chapter 4)
during the 1995 and 1999 measuring periods (DAS=days after sowing).
CHAPTER 5 - MEASUREMENT AND MODELLING OF RAINFALL INTERCEPTION
65
revised analytical model assumes that evaporation from the stems only takes place after the
storm, resulting in much lower losses. Arguably, in the present case of a well ventilated and
relatively open agricultural vegetation, the former assumption comes closer to reality, also in
view of the observation that during rainfall a significant part of the water stored on the
vegetation was retained in the junctions between the stem, branches and leaves (Section 5.5.3).
On the basis of the foregoing observations, the adapted Gash model with only the value of
a E / R optimised was considered to describe the measurements best (Table 5.4, Fig. 5.7). It
was subsequently used to test the sensitivity of the various model parameters, the results
of which are given in Table 5.6. The relative wet canopy evaporation rate
a E / R can be
identified as the most sensitive parameter (cf. Gash, 1979; Gash et al., 1980). Leaf area
index (L) is important as well, because of its influence on canopy cover (c) and vegetation
storage capacity (Sv). Leaf area index also exhibits a linear relationship with stemflow
depth, Sf (cf. Eq. [5.6]) and therefore affects estimated interception in this way too.
Although stemflow does not influence predicted interception (because stemflow was
expressed as a percentage of throughfall rather than gross rainfall), by changing estimated
interception it of course affects the agreement between estimated and modelled
interception values. Finally, it appears that, within the parameter ranges tested in this case,
the extinction coefficient (

), the specific vegetation storage capacity (SL), and the energy

exchange coefficient (

) are all less sensitive parameters of the model (Table 5.6).

5.7. Discussion
5.7.1. Rainfall partitioning by mixed crops
Using the vegetation structural parameters and (optimised) relative evaporation rates
listed in Table 5.5 in combination with leaf area index time series to cover the full annual
cropping cycle (including fallow conditions at the end of the dry season) and daily rainfall data
for the 1994/95 and 1998/99 agronomic years (starting on the first of November), the adapted
analytical model was run to estimate annual totals of throughfall, stemflow and interception
loss. Because of the difference in cropping mixture (maize, cassava and rice versus maize and
cassava only), average leaf area index (1.2 versus 0.8) and average (optimised) maximum
Parameter 1995 1999
-10% +10% -10% +10%
a E / R - 9.2 + 9.2 + 7.9 + 7.9
L - 4.4 + 3.9 - 6.2 + 5.8
Sf
*
+ 1.2 - 1.2 + 4.8 - 4.8
SL - 0.75 + 0.75 - 2.1 + 2.1

+ 1.7 - 1.5 + 0.38 - 0.29

$ - 0.03 + 0.02 - 0.03 + 0.03
* change in estimated interception.
Table 5.6. Sensitivity analysis for parameters used in the adapted Gash model showing the
percentage change in modelled interception after a 10% decrease or increase in the value of the
tested parameter.
CHAPTER 5 - MEASUREMENT AND MODELLING OF RAINFALL INTERCEPTION
66
relative wet evaporation rate (0.108 versus 0.235) between the two years, all possible
combinations of rainfall, L time series and
a E / R values were used in the calculations to
assess their respective influence (Table 5.7)
Rainfall in 1994/95 was only slightly above the long-term average of 2650 mm yr
-1
but distinctly higher in 1998/99. On the other hand, the average leaf area index was lower
in 1998/99, mainly because of the lower planting density of cassava (cf. Table 5.1).
Depending on the combination of L and
a E / R , computed annual interception loss for the
mixed crops ranged from 5.0% to 14.4%, with the highest values obtained for high values
of L and, particularly,
a E / R (Table 5.7). Not surprisingly, the development of the leaf
area of the maize was the main factor determining the average stemflow fraction (cf. Eq.
[5.6]). Annual stemflow ranged between 35 and 106 mm (1.2-4.0% of gross rainfall), but
on an individual storm basis it was occasionally estimated at more than 15% of gross
rainfall when the maize had reached its full height (Table 5.5). This underlines the
importance of taking stemflow into account in this type of agricultural vegetation. High
stemflow fractions have been reported for many tropical forest types as well. Some of the
more pertinent studies are listed in Table 5.8. Although some of the values may have been
affected by measurement or interpretation errors, it serves to illustrate the point that
stemflow cannot always be assumed negligible as is often done (cf. Bruijnzeel, 1990).
With regard to overall annual interception losses, it appears from Table 5.7 that the
relative evaporation rate (
a E / R ) has the greatest influence on model output. Using an
a E / R value of 0.108 (as derived from the 1999 data) gave annual interception losses of
142-194 mm (5-7% of gross rainfall), depending on the time series of L and Pg that were
used. Conversely, a value of
a E / R of 0.235, as found for the 1995 data, resulted in
substantially higher interception totals ranging between 278 and 387 mm (10-14% of
gross rainfall). The contrast in annual interception fraction derived for the two years is
striking and the reasons for this discrepancy will be discussed further below. However,
Year Lavg a E / R Pg Tf Sf Ei Tf Sf Ei
mm yr
-1
mm yr
-1
mm yr
-1
mm yr
-1
% % %
1994/95 1.2 0.108 2680 2426 59 194 90.5 2.2 7.2
0.8 0.108 2680 2408 106 166 89.8 4.0 6.2
1.2 0.235 2680 2241 52 387 83.6 2.0 14.4
0.8 0.235 2680 2258 94 328 84.3 3.5 12.2
1998/99 1.2 0.108 2824 2612 40 172 92.5 1.4 6.1
0.8 0.108 2824 2609 73 142 92.4 2.6 5.0
1.2 0.235 2824 2449 35 340 86.7 1.2 12.0
0.8 0.235 2824 2482 65 278 87.9 2.3 9.8
Table 5.7. Rainfall partitioning for two agronomic years as predicted by the adapted analytical
model for mixed crops of variable density, using measured rainfall records, estimated leaf area
index time series and corresponding values for
a E / R , in all combinations. Results based on
actual measured combinations are given in bold. Note: annual totals differ from the totals for
the measuring periods (cf. Table 5.2); the agronomic year started on 1 November.
CHAPTER 5 - MEASUREMENT AND MODELLING OF RAINFALL INTERCEPTION
67
relatively high interception losses (13-15%) have also been observed for cassava
monocropping in Thailand (Witthawatchutikul and Tangtham, 1986) and Cameroon
(Waterloo et al., 1997). Similar values (9-11%) have been reported for young
Austroeupatorium inulifolium scrub elsewhere in Java (Wasser, 1987; Bruijnzeel, 1988), as
well as in young Acacia auriculiformis plantation forest in West Java (Bruijnzeel and
Wiersum, 1987) and selectively logged rainforest in Kalimantan (Asdak et al., 1998).
Conversely, typical values of mature natural and plantation forest in the region are in the order
of 18-22% (e.g. Calder et al., 1986; Bruijnzeel, 1988; Dykes, 1997; Burghouts et al., 1998).
5.7.2. Vegetation storage capacity
The specific storage values determined in this study were much lower than values
inferred from simulated rainfall experiments for heather (Calluna vulgaris; 0.49-0.94 mm;
Leyton et al., 1967; Hall, 1985) and bracken (Pteridium aquilinum; 0.160.78 mm;
Leyton et al., 1967; Pitman, 1989; Table 5.9). This may well be related to the rougher
surface of the latter compared with the crops of the present study and to the fact that the
surface area of plant parts other than leaves was not included in these studies, which tends
to inflate the values of SL. The values determined in the present study are within the range
derived for Eucalypts (0.037-0.19 mm), but generally somewhat lower than found for
other broadleafs (0.074-0.37 mm) and conifers (0.077-0.6 mm; Table 5.9). Again,
however, such calculations do not usually include the surface area of branches and stems.
Vegetation type Location Sf (%) Reference
(Sub-)tropical plantations
Acacia auriculiformis West Java 7-8 Bruijnzeel and Wiersum (1987)
Eucalyptus tereticornis South India 8 George (1978)
Elaeis guineensis (oil palm) Malaysia 8 Maene et al. (1978)
Peronema canescens South Kalimantan 6 Ruslan (1983)
Schima wallichii West Java 5 Solo (1980)
Tectona grandis (teak) Northern India 6 Dabral and Subba Rao (1968)
(Sub-)tropical forest
Secondary lowland rain forest South Kalimantan 5 Ruslan (1983)
Secondary Shorea robusta forest Northern India 7-8 Dabral and Subba Rao (1968)
Lowland rainforest Venezuela 8 Jordan and Heuveldop (1981)
Primary laurel forest Canary Islands 7 Aboal et al. (1999)
Pinus kesiya forest Philippines 6 Florido and Saplaco (1981)
Lower montane rain forest Puerto Rico 8 Clements and Colon (1975)
Upper montane rainforest Jamaica 12-18 Hafkenscheid (2000)
Upper montane rainforest Philippines 12 Mamanteo and Veracion (1985)
Elfin cloud forest Puerto Rico 5-10 Weaver (1972)
Table 5.8. Selected examples of high stemflow fractions (Sf) in (sub-)tropical forest and
plantation ecosystems reported in the literature.
CHAPTER 5 - MEASUREMENT AND MODELLING OF RAINFALL INTERCEPTION
68

Vegetation type L S SL
Method Reference
mm mm
Shrubs and crops
heather (Calluna vulgaris) 1.8 1.2 0.67 A Hall (1985)
heather (Calluna vulgaris) 1.6-2.9 1.4-1.7 0.49-0.94 A Leyton et al. (1967)
bracken (Pteridium aquilinum) 0.4-5.9 0.2-2.7 0.47
#
A Pitman (1989)
bracken (Pteridium aquilinum) 0.4-5.9 0.2-2.7 0.16 A Pitman (1989)
bracken (Pteridium aquilinum) 1.3-4.4 1.0-2.1 0.46-0.78 A Leyton et al. (1967)
maize (Zea mays) 0-3.0 <0.26 0.066-0.088 W this study
cassava (Manihot esculenta) 0-1.4 <0.091 0.033-0.065 W this study
rice (Oryza sativa) 0-1.5 <0.066 0.039-0.044 W this study
Eucalypts
Eucalyptus cinerea 3.6* 0.4 0.11 A Aston (1979)
E. dives 4.1* 0.3 0.073 A Aston (1979)
E. globulus (7 yrs) 3.2 0.2 0.066 R Valente et al. (1997)
E. maculata 13.4* 0.5 0.037 A Aston (1979)
E. maculata / E. globoidea 3.0 0.4 0.12 L Dunin et al. (1988)
E. mannifera 2.9* 0.3 0.10 A Aston (1979)
E. pauciflora 4.2* 0.8 0.19 A Aston (1979)
E. spp. 1.2 0.1 0.10 W Crockford and Richardson (1990)
E. viminalis 4.1* 0.2 0.049 A Aston (1979)
Other broadleaved species
Acacia auriculiformis 6.8-8.1 0.5-0.6 0.074 R Bruijnzeel and Wiersum (1987)
Acacia longifolia 7.6* 0.6 0.079 A Aston (1979)
Nyssa sylvatica 0.082 W Liu (1998)
Paraserianthes falcataria 0.091 W Van Dijk (unpublished)
Evergreen laurel forest 7.8 2.5 0.32 R Aboal et al. (1999)
Regenerating rain forest 5.9 1.1 0.19-0.20 R Schellekens et al. (1999)
various tropical forest trees 2.2-8.3 0.11-0.16 W Herwitz (1985)
lowland rain forest 4.4-6.7 1.6-2.2 0.23-0.37 R Tobn Marin (1999)
lowland rain forest 4.3 0.74 0.17 R Lloyd et al. (1988);
McWilliam et al. (1993)
Conifers
Pinus caribaea 3.7-4.0 0.8-1.4 0.16-0.20 R Waterloo (1994)
P. caribaea (post cyclone) 1.5-3.1 0.3-0.6 0.22-0.35 R Waterloo (1994)
P. pinaster (15 yrs) 3.0 0.6-0.7 0.19-0.22 R Lousteau et al. (1992)
P. pinaster (60 yrs) 2.7 0.4 0.15 R Valente et al. (1997)
P. radiata 1.7 0.4 0.24 R Kelliher et al. (1992)
P. radiata 13.1* 1.0 0.08 A Aston (1979)
P. radiata 8.8 2.0 0.23 W Crockford and Richardson (1990)
P. radiata 0.12 ? Whitehead and Kelliher (1991)
P. elliotti (slash pine) 0.077 W Liu (1998)
Pine-spruce forest 4-5 1.5-1.7 0.3-0.4 R Lankreijer et al. (1999)
Pine-spruce forest 4-5 2.4 0.5-0.6 M Lankreijer et al. (1999)
Taxodium ascendens 0.093 W Liu (1998)
Pseudotsuga menziesii 19 1.5 0.079 ? Massman (1983)
Pseudotsuga menziesii 9-13 2.4 0.18-0.27 M Klaassen et al. (1998)
A = artificial rainfall; W = wetting or immersion; R = regression on throughfall data; L=lysimeter study;
M= microwave attenuation;
*
leaf area per unit crown cover;
#
maximum value during rainfall.
Table 5.9. Review of studies that have determined canopy storage capacity (S) along with leaf
area index (L).
CHAPTER 5 - MEASUREMENT AND MODELLING OF RAINFALL INTERCEPTION
69
As indicated earlier, it was observed during the wetting experiments, as well as in the field
during storms, that much of the water retained by the vegetation was stored in the
junctions between stems and branches or leaves. The same was observed on heather and
spruce by Hall (1985) using special photographic techniques.
Multiplying specific canopy storage values times leaf area index values over time (cf.
Section 5.6.1) yielded canopy storage values that ranged between 0.07 and 0.25 mm in 1995
versus 0.04 and 0.30 mm in 1999. Canopy storage capacity values of broad-leaved forest cited
in the literature range between 0.2 mm (Valente et al., 1997 for a Eucalypt forest in Portugal)
and 1.3 mm (Rowe, 1983 for Nothofagus forest in New Zealand). Compared with these
values, the storage capacity values estimated in the present study are on the low side, which
can be explained by the low leaf area index and the absence of woody parts in the agricultural
crops under consideration.
5.7.3. Wet canopy evaporation rates
In the absence of above-canopy climatic data, average wet canopy evaporation rate, E , is
often calculated from E / R values as determined from the slope of the regression equation
linking Pg and daily interception loss (Gash, 1979), and average rainfall rates. Combining the
presently obtained average rainfall intensities of 4.7 mm h
-1
and 4.3 mm h
-1
for the 1995 and
1999 periods, respectively, with the optimised values for
a E / R in the adapted Gash model
yielded wet evaporation rates of 0.60-1.04 mm h
-1
and 0.19-0.44 mm h
-1
for the two periods
(Table 5.5). Likewise, wet canopy evaporation rates per unit canopy cover (
c E ) derived for
the revised and adapted analytical models were 1.1-1.2 mm h
-1
and 0.48-0.55 mm h
-1
for 1995
and 1999, respectively. A very similar contrast in evaporation rates between rainy seasons
has been reported by Bruijnzeel and Wiersum (1987), who measured rainfall interception
in a young Acacia auriculiformis plantation elsewhere in West Java. They ascribed the
discrepancy between years largely to the failure of a fixed throughfall gauge arrangement
to adequately sample the spatially highly variable canopy. However, the present study was
conducted in a more homogeneous vegetation and used roving instead of fixed gauges.
Although the number of gauges was limited (six in 1995, ten in 1999), statistical analysis
of the standard deviations from the mean for individual throughfall gauges showed that
this could not explain the large difference in wet evaporation rate (and hence interception
loss) inferred from the two data sets. Rather, the presently observed contrast in overall
interception loss between the 1995 and 1999 observation periods may, at least partly, be
explained in terms of the difference in exposure to wind of the two sites that were used. In
the 1995 season, measurements were made near the top of a ridge, whereas the 1999 data
were collected on a more sheltered part of a slope near the bottom of a valley (cf. Fig. 3.1).
Apart from differences in ventilation, the overall leaf surface area and canopy cover fraction
were smaller in 1999 as well (by 38% and 25%, respectively; cf. Table 5.5). This does not only
have a bearing on the fraction of free throughfall and on canopy capacity (on average 0.12 mm
in 1995 versus 0.09 mm in 1999), but also on the computed stemflow fraction (cf. Eq. [5.6])
and, ultimately, on interception loss. Furthermore, the development of the maize plants,
planting density and the number of seeds per hole differed somewhat between the two years.
Although Eq. [5.6] was designed to take such differences into account, a residual error may
have remained. A 25% error in the estimation of the stemflow fraction would result in a change
in estimated total interception of 9 mm (1995) and 16 mm (1999), or 3% and 12% respectively
(cf. Table 5.6).
CHAPTER 5 - MEASUREMENT AND MODELLING OF RAINFALL INTERCEPTION
70
The optimised wet canopy evaporation rates per unit canopy cover are well above the 0.1-
0.2 mm h
-1
normally computed for E under humid tropical conditions using the Penman-
Monteith equation and above-canopy climatic data (cf. Bruijnzeel and Wiersum, 1987;
Lloyd et al., 1988; Schellekens et al., 1999). There is a steadily growing number of studies
that report wet canopy evaporation rates inferred from throughfall measurements to be
much higher than suggested by Penman-Monteith theory, particularly under wet maritime
climatic conditions. Pertinent examples from tropical sites include Brunei (Dykes, 1997),
West Java (Bruijnzeel and Wiersum, 1987), Fiji (Waterloo et al., 1999) and Puerto Rico
(Schellekens et al., 2000), and in the temperate zone the U.K. (Shuttleworth and Calder,
1979) and New Zealand (Pearce et al., 1980). In view of the fact that the phenomenon is
most pronounced for near-coastal and island locations (see discussion in Shuttleworth and
Calder (1979) and Schellekens et al. (2000)), advected energy is likely to play an
important role in maintaining such high rates of evaporation.
5.8. Model performance and outlook for further application
The adapted analytical model was able to predict the measured interception values well
within the accuracy of measurement. The introduction of leaf area index L as an additional
parameter substantially improved the performance of the model compared with the revised
version of Gash et al. (1995) (cf. Fig. 5.7, Table 5.4). It offers a good physical representation
of the changes in vegetation characteristics that occurred during the periods of measurement.
The inclusion of L as a model parameter also provides an opportunity to estimate canopy
storage capacity S independently, e.g. from wetting experiments (cf. Aston, 1979; Liu, 1998).
At the very least, it allows a useful check as to how realistic canopy storage values estimated
from throughfall analysis are likely to be (cf. Rowe, 1983).
Fitting the predictions of both the sparse canopy and the adapted analytical model by
optimisation of the corresponding relative wet canopy evaporation rates gave marginally
better results than using values derived from regression methods (Table 5.4). Because
both the conventional regression methods and the presently used optimisation approach
were based on the criterion of least squared differences, and because the computational
efforts are no greater using modern computers, there seems to be no particular argument
anymore in favour of the traditional regression analysis.
Somewhat surprisingly in view of the smaller number of gauges used in 1995 (six,
versus ten in 1999), overall model efficiency was notably higher for the 1995 period
(Table 5.4). Indeed, when comparing the same model, the residual variance (F
2
) of the
1995 data was greater than that of the 1999 data (3.6 versus 2.3 mm
2
). However, the
initial variance (F0
2
) of the 1995 data was more than five times greater (18.3 versus 3.5
mm
2
), thereby resulting in an apparently greater model efficiency. In other words, the
absolute error of the interception estimates was probably higher in 1995, but because of
the much larger variation between individual interception estimates (caused by a larger
number of large storms and a generally higher magnitude of interception in 1995; Table
5.2), agreement between modelled and estimated interception was still better. This point is
illustrated further by a comparison of the residual differences between estimated and
modelled interception values with the variation in individual throughfall gauge catch on a
storm-to-storm basis. It appeared that the residual difference between modelled and
estimated interception was within one standard deviation from the mean catch for 82% of
all 103 storms in 1995 and for 88% of the 116 storms in 1999. Of the remaining 19 storms
CHAPTER 5 - MEASUREMENT AND MODELLING OF RAINFALL INTERCEPTION
71
in the 1995 data set only five, and of the remaining 14 storms in the 1999 data set only
three, were larger than 2 mm. Thus, most of the storms for which interception was poorly
predicted were very small. As suggested earlier (Section 5.6.1), throughfall amounts for
such small storms may well have been exaggerated by dew deposition.
The two data sets suggested strongly different relationships between leaf area index L on
the one hand and the magnitude of the relative evaporation rate E / R on the other. The 1995
data resulted in a value for the energy exchange coefficient

of 0.27, which was less than the

value of the extinction coefficient

(relating L to canopy cover; cf. Eq. [5.2]). During the early

stages of crop growth, this would suggest evaporation rates that are lower than expected on
the basis of the revised analytical model of Gash et al. (1995), using appropriate values for
canopy cover fraction. Conversely, the 1999 data suggested a much higher value of 1.32 for

,
implying that wet canopy evaporation rate reached high values even at low vegetation density.
This discrepancy may be related to the fact that the two sites experienced very different micro-
meteorological conditions (cf. Section 5.7.3), but it may also reflect sampling errors originating
from the use of a limited number of throughfall gauges (apart from the fact that in reality R E
values are not equal for individual storms, as assumed by the analytical model; Gash, 1979).
Therefore, the present results do not provide a strong argument to refute the assumption
expressed in the revised Gash model (Gash et al., 1995), that the wet canopy evaporation rate
is directly related to canopy cover fraction. However, the introduction of the energy exchange
coefficient

in the adapted model constitutes a flexible and powerful tool to test hypotheses
regarding this assumption in future studies of interception in vegetation of contrasting or
changing density.
Finally, the additional adaptation introduced in the adapted analytical model of also
allowing evaporation from wetted stems during storms can have a significant impact on the
relative magnitude of the components of interception loss. In the present application of the
adapted model, it was assumed that evaporation rates from wetted stems equal those from the
wet canopy. In open or low vegetation types where within-stand ventilation will be
pronounced, this is conceptually justified. However, the same assumption will clearly not hold
in dense forests. For example, Rutter and Morton (1977) suggested the evaporation rate from
stems (on a projected area basis) to be only 2% of wet canopy evaporation rate. Therefore,
alternative formulations of the relative magnitude of the evaporation rate from wetted stems
compared with that from the wet canopy will be needed in such cases (cf. Chapter 4).