N.A. Jackson Agricultural and Forest Meteorology 100 2000 323–336 329
that the value of g
a
did also. However, as canopy cover was less than 2 at this stage, any error in the value
of g
a
would account for a negligible percentage of the overall interception.
As the canopy storage capacity S
c
was known to be affected by canopy architecture Asdak et al., 1998b
it was expected to vary, depending on the height and shape of the trees as they grew, and on the degree
of canopy cover resulting from the pruning regime adopted. Canopy storage capacity was determined us-
ing the envelope method of Leyton et al. 1967, from the negative intercepts of linear regressions between
throughfall and gross rainfall with a pre-defined slope set to 1 − p
t
. There is an inevitable degree of subjectivism in-
volved in the envelope approach Rowe, 1983, who noted that seasonal variations, rainfall intensities and
wind speeds had a marked effect on determining S
c
using the Leyton et al. 1967 method. Data from 172 rainfall events recorded at various stages dur-
ing the experiment were used to derive S
c
values of between 0.71 and 0.93 mm. In general, as the tree
canopies expanded, S
c
increased. Pruning reduced S
c
, but to varying degrees, as might be expected from removing either the lowermost branches or
foliage higher up the tree. Variation in S
c
also re- flected monthly and seasonal variations in the rainfall
intensity.
Fig. 3. Changes in the fractional cover, c, from the G. robusta canopies in the sole tree thick line and intercropped thin line treatments, estimated from the formula from Fig. 2. Arrows denote times at which the canopies were pruned. Bars show weekly rainfall during the
long February–July and short October–January rainy seasons.
Trunk storage capacities for smooth barked Aus- tralian rainforest tree species like Grevillea were found
to be much lower than rough barked equivalents Her- witz, 1985, suggesting that stemflow from the Grevil-
lea in our experiment might be significant. Due to the observed variability in stemflow from trees of different
sizes, S
t
and p
t
were derived following the procedure of Lloyd et al. 1988. Separate linear regressions of
stemflow against gross rainfall were determined for 42 trees. The averages of the intercepts 0.185 ±0.03 mm
and of the slopes 0.026 ± 0.007 were taken as esti- mates of S
t
and p
t
, respectively. Monthly values of the canopy cover, c, were calculated using values of A
p
, as shown in Fig. 3.
3. Results and discussion
Fig. 3 shows the monthly changes in fractional tree canopy cover, c, in the T
d
and CT
d
plots during the ex- periment. After 2
1 4
years of growth after planting, tree canopy cover, c, in the T
d
and CT
d
plots in January 1994 was 0.27 and 0.18 respectively. The values of c
in the CT
d
plots reflect the smaller tree size resulting from earlier resource competition between trees and
crops after establishment Lott et al., 1997. Remov- ing the lowermost 1 m of canopy at this point reduced
c to 0.16 and 0.11 in the T
d
and CT
d
treatments, re- spectively.
330 N.A. Jackson Agricultural and Forest Meteorology 100 2000 323–336
Table 3 Comparison of gross rainfall, P
g
, canopy throughfall T
f
, and stemflow, S
f
, and rainfall interception, I, in the agroforestry CT
d
and sole tree T
d
plots, for 126 rainfall events between November 1994 and June 1997 Parameters
Units Sole tree plots
Agroforestry tree + crop plots Area average
a
By base of trees
b
Between trees
c
Area average By base of trees
Between trees P
g
mm 1583
1583 1583
1583 1583
1583 T
f
mm 1400
1338 1444
1409 1379
1452 88.4
84.5 91.2
89.0 87.1
91.7 S
f
mm 11
22 –
10 19
– 0.7
1.4 –
0.6 1.2
– I
d
mm 172
223 139
164 185
131 10.9
14.1 8.9
10.4 11.7
8.3
a
T
f
, S
f
and I expressed per 12 m
2
ground area occupied per tree. Values of throughfall are means of interception gauges weighted according to ground areas they represent.
b
Means of raingauges closest to the base of the trees 0.3 m in each treatment.
c
Means of raingauges furthest from the trees 2.5 m in each treatment.
d
Interception, I, defined as P
g
− T
f
− S
f
.
Canopy cover increased steadily following prun- ing, throughout the dry season, before increasing
rapidly in both treatments following the onset of the 1994–95 short rains. Both treatments were pruned
again, between the rainy seasons, reducing c from 0.45 to 0.29 and from 0.40 to 0.24 in the T
d
and CT
d
treatments, respectively. The final pruning, in Novem- ber 1996, reduced the canopy size by approximately
85, and hence reduced the canopy cover, c, from the largest values observed during the study T
d
= 0.54;
CT
d
= 0.46 to values similar to those observed dur-
ing establishment 0.02 for both treatments. Trees in both treatments recovered quickly, exhibiting values
of c of ∼0.21 by the end of the experiment. Table 3 summarises measurements of rainfall in-
terception, I, made between November 1994 and the end of the experiment in June 1997. During this
period, variation in both storm intensity and dura- tion, as well as in the degree of canopy cover led
to a wide range of interception values for individual rainfall events. At just over 10, rainfall interception
was similar to values of 11 reported for a sparse Mediterranean Eucalyptus plantation Valente et al.,
1997, montane stands of P. kesiya of varying densi- ties from 8 to 13, Veracion and Lopez, 1976, and
montane stands of bamboo and cypress in Kenya be- tween 10 and 15, Pereira, 1973. However, rainfall
interception in the Grevilleamaize system was much lower than values reported for both Ugandan mon-
tane forest ∼35, Hopkins, 1960, or multi-storey agroforestry systems in Costa Rica 30, Imbach et
al., 1989, where the vegetation in these other studies was considerably more dense.
Significant spatial variation in throughfall was ob- served below the canopy, e.g. in the 1994–95 short
rains shown in Fig. 4, as was reported by previous au- thors Lloyd et al., 1988; Asdak et al., 1998a. Greater
interception was observed in both treatments, at po- sitions closest to the trees 0.3 m, as compared with
both the area average and the value at 2.5 m from the trees i.e. midway between four trees, see Fig. 1,
confirming the strong dependence of throughfall on the degree of canopy cover as well as the position
of the gauge with respect to the tree trunk Eschner, 1967.
Stemflow volumes were influenced by the individ- ual tree size, i.e. canopy cover and trunk diameter, and
led to large variation in S
f
between trees. The areal average contribution to interception by stemflow was
small, at ∼0.7, as is generally the case Jackson, 1971; Dykes, 1997. However, it is worth noting that
partitioning of rainfall between throughfall, canopy storage and stemflow can lead to significant spatial
heterogeneity in water distribution below the canopy. Although for the purposes of this paper, stemflow was
calculated on the basis of the 12 m
2
area occupied by each tree, field observations confirmed that stemflow
concentrated water into a small area around the base of a tree, a process described as rainwater funnelling
Herwitz, 1986, 1987 and reported in other studies Jordan, 1978; Prebble and Stirk, 1980; Bellot and
Escarre, 1998.
N.A. Jackson Agricultural and Forest Meteorology 100 2000 323–336 331
Fig. 4. An example of the spatial variation in throughfall below the tree canopy, from the 1994–95 short rains. Gross rainfall P
g
is compared with: the areal average throughfall over the 12 m
2
occupied by each tree see Fig. 1, and throughfall recorded by the base of the tree 0.3 m and between the trees 2.5 m. Also shown are the monthly increases in fractional cover, c, and daily rainfall.
3.1. Comparisons between measured and modelled interception
In order to compare measured rainfall interception loss under continually varying canopy cover with that
estimated using the revised Gash model, 4 periods dur- ing the experiment were chosen, and examined sepa-
rately. The first period started soon after the start of the 1994–95 short rains see Fig. 3, on 1 November 1994
and ended just before the tree canopies were pruned on 10 March 1995. The second period started just af-
ter pruning, on 15 March 1995 and finished at the end of November, 1995. The fourth interval covered the
end of the 1996 long rains, May and June 1996 a period when tree canopy cover was at a maximum, and
was used to compare interception with that recorded during the final period, which started just after the
tree canopies had been severely pruned 4 November 1996 and lasted until the end of the experiment 2
June 1997. The comparisons between monthly observed and
modelled interception losses are shown in Table 4. Initial estimates from the model where the long term
1991–1993 value of ¯ R was used, underestimated the
overall measured interception by 33 mm 20, with more than 50 of this discrepancy occurring during
the first period studied. It is possible that as the net- work of interception gauges was not augmented until
early 1995, part of this discrepancy may be accounted for, by the errors in the measured interception occur-
ring between December 1994 and February 1995.
Moderate pruning of the tree canopy cover halfway through March 1995, by removing the lowermost 1 m
of foliage, caused a slight reduction in both measured
332 N.A. Jackson Agricultural and Forest Meteorology 100 2000 323–336
Table 4 Comparison of gross rainfall, P
g
, with measured and modelled interception by the Grevillea canopies in the sole tree T
d
plots over four periods throughout the experiment
P
g
Measured Interception estimated by Gash model
interception
a
Using long-term mean value of ¯ R
b
Using monthly mean ¯ R
m
values mm
mm mm
mm Period 1 — before moderate pruning
November 94 317
22 6.9
21 6.6
27 8.5
December 94 144
20 13.9
14 9.7
17 11.8
January 95 38
7 18.4
4 10.5
4 10.5
February 95 77
14 18.2
10 13.0
9 11.7
March 95
c
81 11
13.6 8
9.9 8
9.9 657
74 11.3
57 8.7
65 9.9
Period 2 — following moderate pruning March 95
56 5
8.9 6
10.7 6
10.7 April 95
103 10
9.7 9
8.7 9
8.7 May 95
32 4
12.5 4
12.5 5
15.6 August 95
13 0.0
2 15.4
3 23.1
September 95 6
0.0 1
16.7 1
16.7 October 95
48 9
18.8 4
8.3 4
8.3 November 95
82 17
20.7 9
11.0 13
15.9 340
45 13.2
35 10.3
41 12.1
Period 3 — before severe pruning May 96
57 9
15.8 10
17.5 12
21.1 June 96
017 7
41.2 4
23.5 6
35.3 74
16 21.6
14 18.9
18 24.3
Period 4 — following severe pruning November 96
156 0.0
2 1.3
3 1.9
January 97 3
0.0 0.0
0.0 March 97
58 3
6.9 1
1.7 2
3.4 April 97
229 14
6.1 13
5.7 16
7.0 May 97
66 9
13.6 6
9.1 9
13.6 512
26 5.1
22 4.3
30 5.9
Overall totals 1583 161
10.2 128
8.1 154
9.7
a
Mean interception per 12 m
2
ground area occupied per tree, including stemflow values.
b
R is the mean rainfall rate 2.28 mm h
− 1
recorded between 1991 and 1993, before measurements commenced. Monthly rainfall rates, ¯
R
m
were calculated for each month during the experiment in which measurements were made see text for details.
c
As pruning occurred mid-way through March, two rainfall totals are provided for this month: the first from 1 March to10 March inc., and the second from 15 March to 31 March inc.
and modelled interception, but by the end of Period 2 Nov 95 the tree canopies had re-grown and in-
terception losses were larger. More severe pruning in November 1996, where 85 of the canopy was re-
moved, showed a more marked effect on interception. High interception losses observed before pruning in
Period 3, were reduced to zero measured or minimal modelled interception following pruning.
On a month-to-month basis, the model tended to under- or over-estimate interception between 1 and
8 mm, and as several of these months showed mean rainfall rates ¯
R
m
that varied from the long term average used 2.28 mm h
− 1
, the model was re-run entering ¯
R
m
in each case. The results of this estima- tion are also shown in Table 4. This second modelling
run underestimated the total observed interception by only 7 mm ∼4. Although initially Nov 94, mod-
elled interception was greater than that recorded, the model adequately estimated interception, on a monthly
timescale, to within 3–5 mm.
N.A. Jackson Agricultural and Forest Meteorology 100 2000 323–336 333
The fact that the model still underestimated inter- ception losses might be due in part to considering
rainfall as one storm event per day, when in fact the monthly mean number of ‘discrete’ rainfall events per
day varied between 1.0 and 2.2 over the period studied. One might expect a higher interception loss from sev-
eral small storms in a day than from one larger event Jackson, 1971; Bruijnzeel and Wiersum, 1987, and
that isolated short periods of rain 1–3 h occurring within a ‘single’ prolonged event might be expected
to result in significant errors in the interception esti- mated by the Gash analytical model Mulder, 1985;
Dykes, 1997.
The calculations showed that relatively small amounts of water were lost through evaporation either
the canopy wetted up 4 or during storms when it is too small for canopy saturation to occur 3.
By far the largest evaporative losses occurred either while the canopy remained saturated 54 or as it
dried out following rainfall 30. Evaporation from
Fig. 5. a Modelled interception losses by G. robusta canopies in the intercropped CT
d
, and sole-tree T
d
, treatments. Also shown
are monthly estimates of the gap fraction space between tree canopies in the CT
d
and T
d
treatments. Arrows denote times at
which the canopies were pruned. b Monthly gross rainfall amounts bars shown for comparison.
the trunks accounted for a little under 9 of the total evaporative loss.
Given that this revised overall model estimate was close to the measured cumulative interception,
and that it covered a period where the fractional canopy cover, c, ranged between 0.02 and 0.54 the
maximum observed, it was considered acceptable to use the model to estimate cumulative intercep-
tion over the period before interception measure- ments commenced, i.e. from October 1991 to Oc-
tober 1994. During this period, it was impossible to determine canopy storage capacity S
c
values, and therefore the earliest recorded measurement
November–December 1994 was substituted, al- though this may have lead to a slight over-estimate
of interception in the early stages of the exper- iment. As before, monthly values of the canopy
cover, c, were determined from projected area val- ues see Fig. 3. Monthly mean rainfall rates ¯
R
m
were calculated from hourly rainfall data, and ranged
334 N.A. Jackson Agricultural and Forest Meteorology 100 2000 323–336
between 0.5 and 5.7 mm h
− 1
. The results are shown in Fig. 5.
Overall, between October 1991 and June 1997, the cumulative gross rainfall was 4106 mm and the cu-
mulative modelled interception by the trees grown with CT
d
and without T
d
an understorey of maize was 236 mm 5.8 and 286 mm 7.0, respectively.
However, monthly interception losses were generally low during the first 3 years after the trees were planted,
and the cumulative interception up to 1 November 1994, was 59 mm 2.6 and 80 mm 3.6 for the
CT
d
and T
d
treatments, respectively, during which time viable yields of maize were obtained from the
intercropped plots Lott et al., 1997. Highest monthly interception losses were recorded
during the 1994–95 short rains, which were signifi- cantly greater than the seasonal average rainfall see
Table 1. Interception losses decreased after this point, then increased again, following the trends in monthly
rainfall, as well as variation in the free throughfall coefficients that resulted from pruning management
practices. Maize yields were substantially reduced from this point onwards see Lott et al., 1997, as
a result of competition for water between trees and crops. Drastic pruning in late 1996 reduced intercep-
tion losses as a percentage of gross rainfall to values similar to those observed during the first 3 years after
planting, with increased maize growth beneath the trees data not shown.
The Penman–Monteith equation used to estimate the evaporation rate is strongly dependent on the
boundary layer conductance, g
a
. When looking at the effects that thinning had on rainfall interception,
Whitehead et al. 1989 assumed that g
a
was unaf- fected by tree spacing, although it is generally consid-
ered to be the most important characteristic of forest canopies that determines the extent of interception
losses Jarvis and Stewart, 1978. As mentioned ear- lier there are problems associated with determining d
and z
at wide tree spacings Jarvis et al., 1976, as occurred in some agroforestry systems. Therefore it is
worth considering how greater temporal variation in g
a
might have affected the model simulations carried out in this study.
By inverting the Penman–Monteith equation, and ignoring the net radiation term a as being negligible
during rainfall, Teklehaimanot et al. 1991 used an averaging method to determine g
a
at different spacings of 10 m tall Sitka spruce trees. As the gap between the
tree canopies increased from 0 to 2 m, g
a
increased from ∼0.6 to 2.1 m
3
s
− 1
. In the experiment reported here, a comparable
increase in gap size occurred when the trees were severely pruned, moving from a situation where ad-
jacent tree canopies overlapped to a gap of between 2 and 3 m between canopy edges. However, unlike a
study dealing with similar sized trees at different spac- ings, in this experiment as gap size increased it was
at the expense of canopy size, i.e. through pruning.
It is likely that a large increase in g
a
would have occurred immediately after severe pruning took place,
and the average value of ¯ E used would no longer be
appropriate, as subsequent evaporation rates would be significantly higher. This would have resulted
in under-estimated interception losses in the period immediately after severe pruning. However, given
that the aim of the modelling exercise was to sim- ulate cumulative interception from planting through
to a harvestable timber stage, and that even a large under-estimation during periods where interception
losses were small between 1.3 and 3.4, see Table 4 would be negligible in terms of the overall fraction
of incoming rainfall, it is convenient to accept the model as it stands.
As remarked on earlier, agroforestry systems are spatially complex. When looking at how conventional
methods of determining g
a
may be inappropriate in some cases, other factors affecting the Penman–
Monteith equation may also be worth examining. An actively transpiring understorey crop may lead
to a lower vapour pressure deficit below and around the tree canopy, affecting evaporation rates of rain-
fall intercepted by the tree canopy, when compared with a woodlot with nothing but crusted bare soil
below. Clearly there is a need for more research into interception processes within agroforestry systems
and their interaction both with the other components of the water balance, and with canopy manipulation
through pruning.
4. Concluding remarks