Global Change Biology (2010) 16, 15151530, doi: 10.1111/j.1365-2486.2009.02034.

Effects of an experimental drought on the functioning of a

cacao agroforestry system, Sulawesi, Indonesia
L U I T G A R D S C H W E N D E N M A N N *1 , E D Z O V E L D K A M P w 1 , G E R A L D M O S E R z 1 , D I R K
L S C H E R *1 , M I C H A E L K O
H L E R *1 , Y A N N C L O U G H 1 , I S W A N D I A N A S } 1 , G U N A W A N
HO
1
D J A J A K I R A N A } , S T E F A N E R A S M I k1 , D I E T R I C H H E R T E L z1 , D A N I E L A L E I T N E R **1 ,
C H R I S T O P H L E U S C H N E R z1 , B E A T E M I C H A L Z I K **w w 1 , P AV E L P R O P A S T I N k1 , A I Y E N
T J O A zz1 , T E J A T S C H A R N T K E 1 and O L I V E R v a n S T R A A T E N w 1
*Tropical Silviculture and Forest Ecology, Burckhardt Institute, Georg-August-University Gottingen, Busgenweg 1, 37077
Gottingen, Germany, wSoil Science of Tropical and Subtropical Ecosystems, Busgen Institute, Georg-August-University Gottingen,
Busgenweg 2, 37077 Gottingen, Germany, zEcology and Ecosystem Research, Albrecht-von-Haller-Institute for Plant Sciences,
Georg-August-University Gottingen, Untere Karspule 2, 37073 Gottingen, Germany, Agroecology, Georg-August-University
Gottingen, Waldweg 26, 37073 Gottingen, Germany, }Department of Soil Sciences and Land Resources, Faculty of Agriculture,
Bogor Agricultural University (IPB), Jl. Meranti, IPB Darmaga Campus, Bogor 16680, Indonesia, kCartography, GIS and Remote
Sensing, Institute of Geography, Georg-August-University Gottingen, Goldschmidtstr. 5, 37077 Gottingen, Germany, **Landscape
Ecology, Institute of Geography, Georg-August-University Gottingen, Goldschmidtstr. 5, 37077 Gottingen, Germany, wwInstitute of
Geography, Friedrich-Schiller-University Jena, Lobdergraben 32, 07743 Jena, Germany, zzFaculty of Agriculture, Tadulako
University, Palu 94118, Indonesia

Abstract
Agroforestry systems may play a critical role in reducing the vulnerability of farmers
livelihood to droughts as tree-based systems provide several mechanisms that can mitigate
the impacts from extreme weather events. Here, we use a replicated throughfall reduction
experiment to study the drought response of a cacao/Gliricidia stand over a 13-month period.
Soil water content was successfully reduced down to a soil depth of at least 2.5 m. Contrary to
our expectations we measured only relatively small nonsignificant changes in cacao (11%)
and Gliricidia (12%) sap flux densities, cacao leaf litterfall ( 1 8%), Gliricidia leaf litterfall
(2%), soil carbon dioxide efflux (14%), and cacao yield (10%) during roof closure.
However, cacao bean yield in roof plots was substantially lower (45%) compared with
control plots during the main harvest following the period when soil water content was
lowest. This indicates that cacao bean yield was more sensitive to drought than other
ecosystem functions. We found evidence in this agroforest that there is complementary use
of soil water resources through vertical partitioning of water uptake between cacao and
Gliricidia. This, in combination with acclimation may have helped cacao trees to cope with the
induced drought. Cacao agroforests may thus play an important role as a drought-tolerant
land use in those (sub-) tropical regions where the frequency and severity of droughts is
projected to increase.
Keywords: cacao yield, CO2 efflux, fine root biomass, leaf litterfall, plant water uptake, sap flux, shade
trees, soil water, throughfall reduction

Received 30 April 2009 and accepted 19 June 2009

Correspondence: Luitgard Schwendenmann, tel. 1 49 551 399

1118, fax 1 49 551 394 019, e-mail: lschwen@gwdg.de
1
L. S., E. V., D. H., I. A., G. D., S. E., D. He., C. L., B. M., A. T., and T.
T. designed the experiment; L. S., G. M., M. K., Y. C., D. L., P. P., and
O. v. S. performed the field work and laboratory analyses; L. S., E.
V., G. M., D. H., M. K., and Y. C. analyzed and discussed the data;
L. S. and E. V. wrote the paper. All authors commented on the
manuscript.

r 2009 Blackwell Publishing Ltd

Introduction
Rain-fed agriculture is the main source of income in
many developing countries and is vulnerable to the
occurrence of droughts. If changes in precipitation and
temperature occur, this will directly affect the rural
population that depends on rain-fed agriculture (Slingo
et al., 2005). Some climate scenarios for the (sub-) tropics
predict that extreme weather events like drought
1515

1516 L . S C H W E N D E N M A N N et al.
episodes will become more frequent and severe, and
average precipitation may decrease (Sheffield & Wood,
2008). To reduce the impact of droughts on the rural
population it may be necessary to adapt land use
practices to address such future climatic conditions.
Recent publications suggest that diversified and sustainable production systems such as tree-based systems
(e.g. agroforestry) may be more resilient to extreme
climatic conditions than annual crops and tree crop
monocultures as they have several mechanisms to reduce the impact of droughts such as buffering of
humidity, and reduction of air and soil temperature
extremes (Verchot et al., 2007; Lin et al., 2008). Agroforestry systems may thus play a critical role in minimizing the vulnerability of farmers livelihoods to extreme
weather events.
Cacao (Theobroma cacao L.), a neotropical understory
rain forest species, is one of the most important perennial cash crops world wide, and cultivated in the
tropical areas of Central and South America, South-East
Asia and Africa. Millions of farmers, mostly smallholders, depend on cacao for their livelihoods. Traditional cultivation systems are established by planting
cacao under primary or older secondary forest with
minor modifications to the original forest canopy.
Nowadays, however, cacao cultivation takes place in a
range of management systems from shaded agroforests
(under remaining forest cover or planted shade trees) to
nonshaded monocultures (Rice & Greenberg, 2000).
Cacao stands are often established with fast-growing
nitrogen-fixing tree species such as Gliricidia spp. or
Erythrina spp. While recognizing the advantages that
intercropped shade trees provide young cacao plants,
farmers often remove these shade trees after cacao
begins to bear fruit as they fear that competition between cacao and shade trees for water and nutrients
may lower cacao yield (Purseglove, 1968; Alvim, 1977;
Belsky & Siebert, 2003). Although the combination of
crops with shade trees holds the risk of competition
both for aboveground (light) and belowground (water
and nutrients) resources, these risks may be reduced by
the choice of the shade tree species and/or appropriate
management practices (Beer, 1987). Belowground competition for water may for instance be minimized by
planting shade tree species which shed their leaves
during the drier season (Broadhead et al., 2003), or
which take up their water from different soil horizons
than crops (van Noordwijk et al., 1996).
Several studies indicate that water availability affects
key functions of the cacao plant. Results from a field
experiment by Rada et al. (2005) showed that severely
stressed 4-year-old cacao plants (25 days without water)
had the lowest leaf transpiration rates and considerably
lower stomatal conductance as compared with control

plants. Also plant phenology depends on rainfall and

presumably on water availability whereby maximum
leaf fall coincided with low rainfall or drought (Alvim
et al., 1974; Ling, 1986); cacao flowering peaked during
the first rains after a drier period (Alvim et al., 1974;
Young, 1994) and cacao yield correlated with rainfall
during months preceding harvest (Alvim & Alvim,
1978). Using a physiological production model to simulate cacao yield, Zuidema et al. (2005) showed that 70%
of the variability in simulated cacao yield could be
explained by a combination of total annual radiation
and rainfall during the two driest months. The simulated yield reduction due to water limitation was up to
50% for locations with a strong dry season combined
with unfavorable soil conditions. Overall, not much is
known about the drought tolerance and possible
drought adaptation strategies of adult cacao trees.
Our present study evaluates the functioning and
stability of a cacao agroforest under an induced drought
in Central Sulawesi, Indonesia. Indonesia is one of the
major cacao-producing regions world wide and has
been affected in the past by severe droughts related to
ENSO. Our major hypothesis was that a prolonged
drought would strongly affect yield and the functioning
of cacao agroforestry systems. To study the impacts
of a drought on a cacao agroforest we implemented
a replicated throughfall reduction experiment over a
13-month period and measured the following direct
and indirect indicators of water availability, water use
characteristics and ecosystem functioning: soil water
content, fine root biomass and distribution, plant water
uptake, sap flux, leaf litterfall (both in cacao and shade
trees), cacao yield, and soil carbon dioxide (CO2) efflux.
Furthermore, we monitored the occurrence of cacao
pests and diseases, which may also be aggravated by
droughts. We expected that a reduction in throughfall
would strongly reduce available water which in turn
would lead to: (1) a considerable decrease in sap flux in
cacao and shade trees, (2) a drought-induced leaf shedding both of cacao and shade trees, and a decrease in
soil CO2 fluxes, and finally (3) a substantial decline in
cacao yield and maybe even tree mortality. To our
knowledge this is the first replicated experimental
ecosystem-scale investigation on the impact of drought
on the functioning of a cacao agroforestry system.

Materials and methods

Study site
The experimental site was located in the vicinity of
the village of Marena in the Kulawi Valley, Central
Sulawesi, Indonesia (1.5521S, 120.0201E) at 560 m a.s.l.
Average annual precipitation in the region is 2092 mm

r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 15151530

E X P E R I M E N TA L D R O U G H T I N C A C A O A G R O F O R E S T R Y
(measured at the Gimpu meteorological station between
20022006, about 5 km south of Marena at 417 m a.s.l.;
H. Kreilein, O. Panferov & G. Gravenhorst, unpublished
results) and shows a weak bimodal pattern. The mean
annual temperature is 25.5 1C.
We selected a site where we anticipated there would
be no influence of ground water. The water table at this
site was o4.5 m depth as estimated with piezometers.
The soil was classified as a Cambisol with a sandy loam
texture and a high stone content (30%) in the subsoil
(D. Leitner & B. Michalzik, unpublished results). The
main soil physical and chemical characteristics are
summarized in Table 1.
The cacao/Gliricidia agroforest stand was established
in December 2000 on former upland rice and maize
fields planting cacao saplings and Gliricidia sepium

1517

(Jacq.) Kunth ex Steud. cuttings. In Central Sulawesi,

Gliricidia is the most commonly planted shade tree in
cacao plantations. While the density of cacao remained
almost constant over time (1100 trees ha1), approximately 60% of the original Gliricidia trees were removed
from the site in the following years. Main stand structural characteristics before the start of the experiment
are summarized in Table 2. Over 80% of the cacao fine
root biomass was concentrated near the soil surface
(00.4 m depth). In contrast, o35% of Gliricidia fine root
biomass was found in the top 0.4 m. Gliricidia fine root
density was highest between 0.5 and 1.5 m. The maximum fine root depth for both species was around 2 m
(Table 3).
Before the experiment the cacao agroforest had been
fertilized once a year and pesticides were applied every

Table 1 Soil texture, bulk density, carbon and nitrogen concentration, effective cation exchange capacity (ECEC), and pH (H2O) at
the experimental site at different depths, Marena, Central Sulawesi
Soil texture
Depth (m)

Sand (%)

Silt (%)

0.00.05
0.050.1
0.10.2
0.20.4
0.40.75
0.751.5
1.52.5

60.2 
55.0 
55.7 
53.9 
57.9 
68.7 
70.3 

27.0
29.7
28.2
26.5
22.8
19.4
22.8

3.4
2.1
0.9
4.8
2.3
5.3
7.1*









2.5
3.6
3.0
3.3
3.4
3.0
6.4*

Clay (%)

Bulk density
(g cm3)

12.8 
15.3 
16.1 
18.6 
19.3 
11.9 
6.9 

1.25
1.28
1.31
1.32
1.37
1.52
1.60

2.2
2.4
3.5
2.9
3.3
4.4
2.3*

 0.04
 0.04
 0.02
 0.05
 0.09
 0.10
 0.06

Carbon
(g kg1)
16.5
12.6
7.0
4.4
3.3
1.9
0.8

 3.1
 4.0
 1.4
 0.5
 0.4
 0.5
 0.1w

Nitrogen
(g kg1)
1.51
1.13
0.64
0.43
0.37
0.29
0.21









0.22
0.31
0.10
0.03
0.04
0.03
0.02w

ECEC
(cmol kg1)
8.78
7.77
7.57
5.47
7.70
8.86
11.38









2.30
1.52
1.45
0.87
3.46
1.70
5.00w

Soil pH
(H2O)
5.9
6.0
6.1
5.9
5.9
5.9
6.1









0.4
0.4
0.4
0.2
0.3
0.2
0.7w

The values are means  SD, n 5 6.

*n 5 5.
wn 5 3.

Table 2 Stem density, height and diameter, and leaf area index (LAI) of cacao, Gliricidia, and coconut in control and roof plots,
Marena, Central Sulawesi
Gliricidia

Cacao
Parameters

Units

Control

Stem density
Height
Diameter*
LAIw

Trees ha1
m
cm
m2 m2

1038
4.5
9.5
3.5






67
0.1
0.2
0.4

Roof
1022
4.6
9.5
4.1

Control
 68
 0.1
 0.6
 0.4

279
9.7
12.5
1.30z

 15
 1.6
 1.0
 0.1

Coconut
Roof
371
10.6
13.0
1.33z

 46
 1.1
 0.5
 0.1

Control

Roof

43  22
11.2  1.3
34.4  5.7
nd

48
8.3  2.3
21.8  9.7
nd

Measurements were conducted between August and December 2006. The values are means  SD; n 5 3 per treatment.
*Stem diameter was measured at 0.8 m (cacao) and 1.3 m (Gliricidia).
wTrue LAI (LAItrue) values were derived from hemispherical photographs taken with a digital camera (Coolpix S3 with EC-F8
fisheye lens, Nikon Corp.) at 1 and 5 m height at 12 points located on a rectangular grid (10 m  12.5 m) in each plot. WinScanopy
(Regent Instruments Inc., Sainte-Foy, QC, Canada) was used for photograph analyses and the effective LAI (LAIeff) for a zenith angle
of 57.51 was calculated after Bonhomme & Chartier (1972); LAItrue was calculated as LAItrue 5 LAIeff/CI; the clumping index (CI)
was calculated for each photograph after Lang & Xiang (1986).
zGliricidia and coconut.
nd, no data.
r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 15151530

Samples were taken in December 2006 (pretreatment) and June 2008 (post-treatment). The values are means  SD (n 5 3). No significant differences in fine root biomass were
found between control and roof plots in a given period and between sampling dates (ANOVA, Tukeys HSD, Po0.05).
Fine root biomass (diameter o2 mm) and vertical rooting pattern were investigated by excavating two soil pits per plot to 3 m depth. Minimum distance of the soil pit to the
closest stem was 1 m. In each soil pit monoliths of 0.5 m  0.5 m  0.2 m were taken at 0.2 m depth intervals. Roots were sorted in living and dead fine root fractions of cacao,
Gliricidia and coconut. The root biomass was expressed as g root biomass m2 per 0.2 m depth interval, which is equivalent to 0.2 g m3.

52.2  25.1
37
 2.2
87.8  31.9
78
 2.4
79.7  23.5
76
 2.2
63.4  30.2
31
 2.4
78.6  30.2
34
 2.4
118.5  83.5
83
 2.4

Roof
Control
Control
Control

Roof

Gliricidia
Cacao

Roof

Cacao

Post-treatment (June 2008)

Pretreatment (December 2006)

91.9  62.4
96
 2.2
Biomass (g m2)
Proportion of fine roots at 00.4 m depth (%)
Maximum root depth (m)

Rainfall (AR100, Campbell Scientific Inc., Logan, UT,

USA), air humidity and temperature (Campbell CS215),
and global radiation (Campbell CS300) were measured
at a distance of 30 m from the study site at a height of
2 m above the canopy of cacao trees in an area where
shade trees were absent. Data were recorded every 5 s
and logged at 30 min intervals (Campbell CR800).
The amount of throughfall excluded from the roof
plots was estimated weekly using a network of plastic
gutters (4 m  0.2 m; n 5 3 per plot) installed above and
below the panels.
Temperature and relative humidity sensors (Hobo
Pro v2 Logger, Onset Computer Corporation, Bourne,

Fine root biomass (diameter o2 mm) of cacao and Gliricidia in control and roof plots, Marena, Central Sulawesi

Measurement of micrometeorological parameters and

throughfall

Table 3

We established the experiment between November 2006

and February 2007. The throughfall reduction plots and
control plots were laid out in three replicates across a
1 ha area in a stratified random design. Each of the six
plots was 40 m  35 m and all measurements were conducted in a central core zone (30 m  25 m) surrounded by a 5 m buffer zone. Treatment plots
(hereafter called roof plots) were separated from control plots and the adjacent area by trenches lined with
plastic to avoid lateral water uptake and to prevent
water addition to the roof plots by overland flow. We
chose a trench depth of 0.4 m based on the vertical fine
root distribution (Table 3). Throughfall was partially
excluded using panels made of clear polyethylene foil
mounted on bamboo frames (0.5 m  5 m) (Fig. 1a and
b). Each panel drained into plastic lined, wooden gutters that were constructed at approximately 1.2 m
aboveground. From the wooden gutters the water
was drained into plastic-lined trenches and was channeled away from the plots. In the roof plots, bamboo
panels and wooden gutters covered approximately
80% of the plot area while small gaps remained around
tree stems and between some panels. Atmospheric
conditions in the canopy (i.e. relative humidity, temperature, and incident radiation) were not altered by the
experimental setup. The throughfall reduction treatment began on March 1, 2007 and lasted until April
10, 2008. Panels were flipped on their sides every 2
weeks to transfer accumulated litter to the soil surface
below.

Gliricidia

Experimental design

Control

Roof

34 months. During the experiment, neither fertilizer

nor pesticides were applied. Weeding was done manually three times a year and pruning of cacao trees was
done in July and December 2007.

61.8  17.3
35
 2.4

1518 L . S C H W E N D E N M A N N et al.

r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 15151530

E X P E R I M E N TA L D R O U G H T I N C A C A O A G R O F O R E S T R Y

1519

Fig. 1 Picture (a) above and (b) below roof, throughfall reduction experiment, Marena, Central Sulawesi, Indonesia.

MA, USA) were installed 1 m aboveground in each plot

to monitor changes in environmental conditions. No
significant roof effect on the stand temperature and air
humidity beneath the roof was detected.

Determination of soil water characteristics

Volumetric soil water content was measured using time
domain reflectometry (TDR) sensors (Campbell CS616).
On each plot one large (0.8 m width  1.6 m length  3.0 m
depths) and two smaller (0.8 m  1.0 m  3.0 m) soil pits
were dug. In the larger soil pit TDR sensors were placed at
0.10, 0.20, 0.40, 0.75, 1.50, and 2.50 m depth. Additional
sensors were also installed at 0.10, 0.40, and 0.75 m in the
two smaller soil pits. Each sensor was embedded into
undisturbed soil at the end of a 0.30 m hole that was dug
horizontally into the wall of soil pits. Volumetric soil
water content was recorded hourly (Campbell CR1000).
Depth-specific calibration functions were developed according to the methodology described by Veldkamp &
OBrien (2000). Volumetric soil water content (corrected
for stone content) at a given depth was multiplied by
the corresponding horizon depth and then summed
up to obtain the total soil water storage in the top 2 m
of soil.
We calculated relative extractable soil water (REWj) in
the top 2 m of soil following the approach of Vincke &
Thiry (2008):
REWj

SWSj  SWSmin
;
SWSmax  SWSmin

where SWSj (in mm) is the soil water storage at day j,

SWSmin is the average minimum soil water storage
(283 mm) measured in the roof plots, and SWSmax is
the average maximum soil water storage in the control
plots (510 mm). Granier et al. (1999) reported that soil
water content begins to limit maximum transpiration of
various tree species when REW values were o0.4.

Sap flux density, stand transpiration, and plant water

uptake depth
In each plot, three randomly selected cacao and Gliricidia trees were equipped with two thermal dissipation
sensors. Each sensor consisted of two probes (1.5 mm in
diameter and 24 mm in length) constructed according to
Granier (1985). At each monitored tree, one sensor was
installed at the north and south side of the trunk,
respectively. Probes were inserted at 0.6 m (cacao) and
1.2 m (Gliricidia) above the ground into predrilled holes
spaced approximately 14 cm apart. Sensors were
shielded from incident radiation and thermal influences
by a styrofoam box and reflective foil. The setup was
covered with plastic foil to protect the sensors against
rainfall. The upper probe of each sensor was heated
with a constant power of 250 mW. Temperature difference between the upper and lower probe of each sensor
was measured every 30 s and 30 min averages were
logged (Campbell CR1000, AM 16/32). Temperature
differences were converted to sap flux density using
an empirically derived equation (Granier, 1987) and
summed up to yield cumulative daily sap flux density
(Js, g cm2 day1). Sap flux density was measured continuously from February 1, 2007 to June 5, 2008. However, due to sensor exchanges no data were recorded in
September 2007 and a power failure occurred for a few
days in December 2007 and February 2008.
The stand transpiration rate (mm day1) expressed
per unit ground area was calculated following Garca
Santos (2007). We derived daily relationships between
measured tree water use rates and tree diameters for
both species. The diameters of all individuals of the
control and roof plots were entered into the respective
equations and the calculated water use rate of all trees
was summed up. Total water use was then divided by
the area of the plot. Details are given in Kohler et al. (in
press).
Stable isotope analysis (natural abundance of d2H 5
Deuterium, D) was used to assess depth of water uptake
of cacao and Gliricidia. Samples of five individuals per

r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 15151530

1520 L . S C H W E N D E N M A N N et al.
species were collected from control and roof plots during
the pretreatment period (February 2007). Suberized twigs
approximately 10 mm in diameter and 50 mm long were
cut from the canopy. Bark was removed to avoid contamination of xylem water with phloem water. Soil
samples were taken directly below the canopy of each
tree at the following intervals: 00.1, 0.10.3, 0.30.5, 0.5
0.7, and 0.71.0 m. Twigs and soil samples were placed
into 40 mL glass bottles, closed with a Teflon coated lid,
wrapped in Parafilm, and then kept frozen until water
extraction. Water was extracted from plant and soil
samples by cryogenic vacuum extraction (Ehleringer &
Osmond, 1989). Hydrogen isotopic composition was
measured by injecting the extracted water into a hightemperature elemental analyzer (TC/EA, Thermoelectron cooperation, Breman, Germany) coupled via a ConFlo III interface to a Delta V Plus isotope ratio mass
spectrometer (Thermo-Electron Cooperation) (Gehre
et al., 2004). Deuterium isotope ratios were expressed in
% relative to Vienna Standard Mean Ocean Water. Measurement precision was 2%. Analyses were carried out at
the Center for Stable Isotope Research and Analysis
(KOSI, Georg-August-University, Gottingen, Germany).
The isotopic composition of plant water was then compared with soil water at multiple depths in order to
identify the best match based on direct inference (Brunel et al., 1995).

Leaf litterfall and soil CO2 efflux

We deployed twelve 0.75 m  0.75 m (0.56 m2) litterfall
traps systematically on a rectangular grid (10 m  12.5 m)
in each plot. Litter was collected fortnightly from February 14, 2007 to May 31, 2008. Samples were separated
into cacao and Gliricidia leaves. Leaves were dried at
70 1C for 48 h and weighed.
Six respiration chambers were installed along three
parallel transects in each plot. Polyvinyl chloride (PVC)
chambers (area 0.045 m2, height 0.15 m) were inserted to
a depth of about 0.02 m into the soil. Once inserted, the
chambers were left in place and kept free of vegetation
throughout the whole study period. During the measurements, flux chambers were closed with a PVC cover
for about 5 min. Air was circulated at a flow rate of
about 0.8 L min1 between an infrared CO2 gas analyzer
(LI-800, Li-Cor Inc., Lincoln, NE, USA) and the flux
chambers. To prevent pressure differences between
chamber and atmosphere, chambers were vented to
the atmosphere through a 0.025 m long stainless-steel
tube. CO2 concentrations were recorded at 5 s intervals
with a datalogger (Campbell CR800). Soil CO2 efflux
was calculated from the linear change in CO2 concentration multiplied by the density of air and the ratio of

chamber volume to soil surface area. Air density was

adjusted for air temperature measured at the time of
sampling. A linear increase in CO2 concentration typically occurred between 2 and 4 min after placing the
cover over the ring. The coefficient of determination (r2)
of the regression was typically 40.99. The infrared gas
analyzer was calibrated in the lab using a loop with a
column with CO2 scrubber (Soda Lime indicating 48
mesh) as zero standard and a CO2 standard (700 ppm,
Deuste Steiniger GmbH, Muhlhausen, Germany). Measurements were conducted between 9:00 and 17:00
hours local time. Each chamber was measured fortnightly from January 27, 2007 to March 30, 2008. Between April 8 and July 15, 2008 soil CO2 efflux was
measured weekly. Because of equipment failure no CO2
efflux data are available for August 2007.

Cacao yield, pest, and diseases

The number of healthy cacao pods and the number of
infested pods by black pod disease (Phytophthora palmivora Butler), cacao pod borer (Conopomorpha cramerella
Snellen), and mirid bug (Helopeltis sulawesi Stonedahl)
were recorded fortnightly from all cacao trees per plot
(on average 140 trees). The numbers of healthy and
invested pods refer to larger cacao pods only and do not
include the effects of the reproductive strategy of cacao.
Cacao beans were sun-dried for several days. A subsample was then oven-dried (70 1C, 3 days) to determine the sun-dry/oven-dry conversion factor. The
results are expressed as oven-dry bean yield.

Data analyses
We divided the observation time into three periods:
pretreatment (February 2007), treatment (March 1,
2007April 10, 2008), and post-treatment (April 11,
2008June 5, 2008). For all parameters, except
stand transpiration, mean values were calculated for
each plot and sampling date. Mean values of stand
transpiration were estimated based on the nine
trees measured on control and roof plots, respectively.
Mean values were then calculated for each plot-period
combination.
Before analyses, parameters were tested for normality
using ShapiroWilks normality test. The effects of the
treatment were analyzed for each period using one-way
analysis of variance followed by Tukeys HSD post hoc
test. Regression analyses were used to examine relationships between parameters. Effects were considered significant if Po0.05. All statistical analyses were carried
out using STATISTICA version 7 (StatSoft Inc., Tulsa, OK,
USA).

r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 15151530

E X P E R I M E N TA L D R O U G H T I N C A C A O A G R O F O R E S T R Y
Results

Micrometeorological characteristics and throughfall

reduction
Total rainfall over the first 12 months of the study
period (February 2007January 2008) was 2937 mm,
40% higher than the recorded 5-year average measured
at the nearby meteorological station in Gimpu. Average
daily rainfall ranged from 1.5 mm day1 (January 2008)
to 15 mm day1 (April 2007). Average daily air temperature was 24.5 1C. Mean daily maximum vapor pressure
deficit (VPD) was 2.30 kPa and ranged from 0.58 kPa
(July 2007) to 4.12 kPa (September 2007). Integrated
daily global radiation (Rg) varied from 15.8 (June
2007) to 20.4 MJ m2 day1 (October 2007).
At the beginning of the treatment period (March and
April 2007) throughfall in roof plots was on average
53% of control plots. In May 2007, roof closure was
improved and from May 2007 until March 2008 the
proportion of the diverted throughfall was increased to
an average of 78%.

Soil water characteristics

Mean volumetric soil water content in the control plots
varied between 0.3 and 0.45 m3 m3 depending on soil
depth and time of observation (Fig. 2ac). Lower volumetric soil water content in deeper layers was probably
a result of the high proportion (up to 30%) of stones.
Volumetric soil water content in the roof plots started to
diverge from the control plots approximately 10 days
after roof closure. With the exception of 2.5 m depth,
soil water content decreased at all depths simultaneously (Fig. 2ac). At 2.5 m depth soil water content
started to decline only after two and a half months
after roof closure (data not shown). Despite a throughfall diversion of 78% minor recharges were measured
at the soils surface (Fig. 2a) in the roof plots following
intensive rain showers. These recharges were however
insufficient to reach the soil water content measured
in the control plots. Between August and December
2007 we observed only a minor decline in volumetric
soil water content (Fig. 2ac). The lowest soil water
content and REW in the roof plots were measured
at the end of February 2008 (Fig. 2ad) after a dry spell
in January and February 2008 (Fig. 2e). The effect
of low precipitation in January and February 2008 was
also observed in the control plots where soil water
content decreased by 25% in 0.1 m depth (Fig. 2a) and
by 15% in 0.75 m depth (Fig. 2b). Within 3 weeks
after removing the panels, total soil water storage in
the roof plots increased to about 90% of its initial level
(Table 4). Soil water storage did not fully recover to

1521

pretreatment levels which may be related to a hysteresis

effect (Topp, 1971).
In summary, although our experiment was conducted
in a year with above-average rainfall, our throughfall
reduction setup was successful in creating conditions
where, over the course of the roof closure, soil water
storage in the roof plots was significantly lower compared with control plots (Table 4) and REW in the roof
plots was o0.4 over an extended period (Fig. 2d).
Heavy rain showers led to slight water recharge but
this was limited to the topsoil only.

Sap flux density, stand transpiration, and plant water

uptake depth
Daily cacao sap flux densities did not differ significantly
between control and roof plots during the pretreatment
period (Table 4). In the roof plots, mean cacao sap flux
densities started to decline 10 weeks after roof closure
(mid June 2007). The roof to control ratio approached
0.85 and then remained relatively constant until December 2007 (Fig. 3b). For a few days during a dry
period (January and February 2008) the roof to control
cacao sap flux ratio dropped below 0.7 (Fig. 3b). Over
the course of the treatment period, sap flux densities of
cacao trees in the roof plots were on average 11% lower
than those of the control plots (Table 4). After removing
the panels, sap flux densities of cacao trees in the roof
plots returned to pretreatment values within a few days
and were similar to control plot measurements (Fig. 3b,
Table 4). An assessment of the influence of overstory
gap fraction on cacao sap flux density conducted 8
months after roof closure revealed that cacao trees
reached highest sap flux densities at an intermediate
gap fraction (Fig. 4).
Daily Gliricidia sap flux densities were within the same
range of values measured for cacao trees (Table 4).
During the pretreatment Gliricidia sap flux densities in
the roof plots did not differ significantly from the control
(Table 4). Approximately 8 weeks after roof closure, roof
to control Gliricidia sap flux density ratio decreased to
0.85 and thereafter remained relatively constant until
November 2007 (Fig. 3c). Between January and March
2008, roof to control ratios of Gliricidia sap flux densities
reached the lowest level (0.650.75) (Fig. 3c) when a
natural dry spell helped further reduce soil water storage to minimum recorded levels. Over the course of the
treatment period, sap flux densities of Gliricidia in the
roof plots were on average 12% lower than those of
control plots (Table 4). Upon roof opening in April 2008,
Gliricidia sap flux densities in the roof plots returned to
control levels quickly (Table 4).
No significant differences in daily stand transpiration
rates were estimated between control and roof plots

r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 15151530

1522 L . S C H W E N D E N M A N N et al.

Fig. 2 Volumetric soil water content (m3 m3) in (a) 0.10 m, (b) 0.75 m, (c) in 1.50 m, (d) relative extractable water to 2 m depth, and (e)
rainfall in the control and roof plots, Marena, Central Sulawesi. (ac) Values are means  SD (n 5 3). The treatment period (roof closure,
March 1, 2007April 10, 2008) is indicated by an arrow.

r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 15151530

E X P E R I M E N TA L D R O U G H T I N C A C A O A G R O F O R E S T R Y

1523

Table 4 Soil water storage to 2 m depth, sap flux density of cacao and Gliricidia, stand transpiration, leaf litterfall of cacao and
Gliricidia, soil CO2 efflux, cacao yield, infection rates of black pod disease, cacao pod borer, and Helopeltis during the pretreatment
(February 2007), treatment (March 1, 2007April 10, 2008) and post-treatment (April 11, 2008June 5, 2008) period
Pretreatment
Parameters

Units

Control

Soil water storage

Sap flux cacao
Sap flux Gliricidia
Stand transpiration
Leaf litterfall cacao
Leaf litterfall Gliricidia
Soil CO2 efflux
Cacao yield
Black pod disease
Cacao pod borer
Helopeltis

mm
g cm2 day1
g cm2 day1
mm day1
g m2 2 week1
g m2 2 week1
mg C m2 h1
kg ha1 2 week1
n ha1 2 week1
n ha1 2 week1
n ha1 2 week1

495 
151.8 
143.8 
1.4
4.7 
11.9 
113.1 
n.d
nd
nd
nd

Treatment
Roof

27
20.3
16.6
1.5
2.8
13.7

470 
151.5 
154.9 
1.4
4.0 
15.2 
130.5 
n.d
nd
nd
nd

Control
24
25.3
43.6
2.0
5.4
28.0

494 
166.9 
151.7 
1.5
9.9 
14.1 
131.1 
26.4 
58 
281 
206 

Post-treatment
Roof

33
22.0
13.6
0.5
2.0
15.1
5.9
10
63
68

348 
147.9 
133.7 
1.3
10.7 
13.9 
112.8 
23.7 
49 
196 
184 

Control
26*
22.9
36.9
1.6
2.7
16.5
4.2
4
58
34

502 
153.3 
150.1 
1.4
7.7 
9.5 
105.4 
45.2 
523 
147 
108 

Roof
33
23.9
14.9
0.8
0.9
18.1
16.0
155
40
54

406 
147.0 
147.0 
1.4
6.6 
9.5 
132.6 
25.0 
376 
149 
102 

19*
34.0
46.2
0.3*
0.9
26.7
1.9*
177
61
6

Values are means  SD (n 5 3).

*Indicates significant differences between control and roof plots in a given period (ANOVA, Tukeys HSD, Po0.05).
nd, no data.

during the pretreatment period (Table 4). Over the

13-month roof closure, stand transpiration rate in the
roof plots was on average 13% lower than those of
control plots (Table 4). Between May 2007 and March
2008 the amount of water transpired by the cacao/
Gliricidia stand was equal to or higher than the amount
of water provided by the throughfall and stemflow that
reached the soil surface in the roof plots. Stand level
transpiration rates in the roof plots considerably exceeded water input in February 2008.
Cacao plant water (Fig. 5a) was more enriched in dD
as compared with Gliricidia (Fig. 5b). Under cacao, soil
water dD values decreased sharply between the soil
surface and 0.3 m depth and then remained constant
(Fig. 5a). Soil water dD values under Gliricidia decreased
gradually to approximately 0.5 m depth, followed by a
constant isotopic signature at greater depth (Fig. 5b).
Comparing the plant water values with the isotopic
gradients in the soil profiles and assuming that plants
were obtaining water from a single dominant source,
direct inference suggests that cacao was primarily obtaining water from the upper 0.3 m of the soil profile
(Fig. 5a) while Gliricidia was obtaining water primarily
from a depth o0.3 m.

Leaf litterfall
During the pretreatment period cacao leaf litterfall did
not differ significantly between roof and control plots
(Table 4). Cacao leaf shedding in roof plots exceeded
leaf loss in control plots by 22% between March and
July 2007 (P 5 0.12) (Fig. 3d). However, over the course

of the roof closure cacao leaf litterfall was on average

8% higher for the roof plots as compared with control
plots. Cacao leaf litterfall in the roof plots was significantly lower (14%) compared with the control during
the post-treatment period (Table 4).
Gliricidia leaf litterfall was significantly higher than
cacao leaf litterfall in both control and roof plots
throughout the study period. During the pretreatment,
Gliricidia trees in the roof plots shed marginally more
leaves than in the control plots (Table 4, Fig. 3e);
however, over the 13-month roof closure leaf litterfall
from Gliricidia did not differ significantly between control and roof plots. After panel removal no difference
between roof and control was measured (Table 4).

Soil CO2 efflux

Before roof closure soil CO2 efflux did not differ significantly between roof and control plots (Table 4).
During the treatment period the ratio roof to control
soil CO2 efflux dropped to approximately 0.8 for most
of the time, indicating a reduction in soil CO2 efflux in
the roof plots compared with the control plots (Fig. 3f).
Significantly lower soil CO2 emissions were measured
for the period between January and April 2008 corresponding to the phase when soil moisture storage was
at its lowest. For the overall treatment period soil CO2
efflux rates in the roof plots were on average 86% of
control plots. Upon roof opening, a flush of CO2 was
recorded in the roof plots (Fig. 3f). A parabolic function
best described the relationship between soil water storage and soil CO2 efflux (Fig. 6).

r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 15151530

1524 L . S C H W E N D E N M A N N et al.

Fig. 3 (a) Soil water storage to 2 m depth, (b) sap flux density of cacao, (c) sap flux density of Gliricidia, (d) leaf litterfall of cacao, (e) leaf
litterfall of Gliricidia, and (f) soil CO2 efflux. Values are expressed as roof to control ratio (i.e. a value of 1.0 represents roof-control plot
parity, and dashed line). The lines represent the centered moving average over 30 days. The treatment period (roof closure, March 1,
2007April 10, 2008) is indicated by an arrow.

Cacao yield
During the main harvest (April to June 2008; post-treatment period) cacao yield in roof plots was reduced to
55% of that in the control plots (Table 4). Yields peaked in
May of both 2007 and 2008. Highest cacao yield (80 kg
oven-dry beans ha1 2 week1) occurred in control plots
in May 2007. Between July 2007 and March 2008 bean
yield was o20 kg oven-dry beans ha1 2 week1 in both

roof and control plots. Over the course of the 13-month

throughfall manipulation total cacao yield in the roof
plots was 10% lower than the control plots (Table 4).

Cacao pests and diseases

During roof closure the harvested pods were mainly
infected by cacao pod borer and Helopeltis (Table 4), but

r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 15151530

E X P E R I M E N TA L D R O U G H T I N C A C A O A G R O F O R E S T R Y

1525

significant differences between roof and control plots

were not found. In contrast, the number of pods infected with black pod disease was significantly lower in
the roof plots when compared with control plots for the
last 4 months of the experimental drought period

(JanuaryApril 2008). During the post-treatment period

most pods were damaged by black pod disease.
However, the number of infected pods in roof plots
did not differ from the control plots after roof opening
(Table 4).

Fig. 4 Gap fraction above cacao trees and sap flux density of
the same individuals in roof plots during November/December
2007. Factor 1 indicates highest gap fraction. An inverse parabolic function was used to fit the data (R2 5 0.87, R2adj. 5 0.85,
P 5 0.001). Gap fraction (here 101 from azimuth) was estimated
from hemispherical photographs using CANEYE 5.0 (INRA, 2007).
Photographs were taken with a digital camera (Coolpix S3 with
EC-F8 fisheye lens, Nikon Corp., Tokyo, Japan) above each of the
nine cacao trees.

Fig. 6 Relationship between soil water storage to 2 m depth

and soil CO2 efflux in control (open triangles) and roof plots
(filled triangles). An inverse parabolic function was used to fit
the data (R2 5 0.35, R2adj. 5 0.33, P 5 0.001).

Fig. 5 Values of dD (%) of plant water and soil water from the agroforest stand, Marena, Central Sulawesi: (a) cacao and (b) Gliricidia.
Values are means  SD (n 5 5). Samples were taken during the pretreatment period (February 2007). The dashed lines indicate the depth
of water uptake of the respective species.
r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 15151530

1526 L . S C H W E N D E N M A N N et al.
Discussion

Ecosystem scale throughfall reduction as an experimental

approach to study droughts
We chose to conduct a replicated ecosystem scale
throughfall reduction in the field. Large-scale, whole
ecosystem experiments are the best means of examining
individual tree responses, interplant interactions as well
as ecosystem level carbon, water, and nutrient responses simultaneously (Hanson, 2000). However, our
approach deviates from a real drought. Under real
drought conditions, we would expect increases in temperature, radiation as well as VPD.
The main effect of a partial throughfall removal is the
strong reduction of water availability in the soil. In case
a natural dry spell occurs during the throughfall reduction, the ecosystem cannot rely on the soil for water
supply and will quickly experience drought stress.
Measurements show that our throughfall reduction
was successful in reducing soil water storage at least
down to 2.5 m. Similar experiments have also been
conducted in temperate climates (Gundersen et al.,
1995; Lamersdorf et al., 1998; Hanson et al., 2001) and
in the humid tropics of Brazil (Nepstad et al., 2002;
Fisher et al., 2007) where throughfall reductions were
imposed on old-growth Amazonian rain forests.

Leaf litterfall and cacao yield in the control plots

Leaf litter production of our cacao/Gliricidia agroforest
(6.2 Mg dry mass ha1 yr1 without pruning residues,
March 2007February 2008) was within the range
(521 Mg dry mass ha1 yr1; including pruning residues)
reported from shaded cacao stands in Malaysia, Venezuela, Costa Rica, Brazil, and Cameroon (Beer, 1988;
Hartemink, 2005) and exceeded litterfall of many tropical
forests (Vitousek, 1984). Net primary production (NPP) of
this cacao/Gliricidia stand (13.7 Mg ha1 yr1, G. Moser,
unpublished results) was at the upper end of NPP values
(715 Mg ha1 yr1) reported for tropical rain forests
(Clark et al., 2001) which illustrates the high productivity
of the stand.
We did not find evidence in our data that short-term
changes in rainfall or temperature controlled the timing
of leaf shedding and/or cacao yield. It was much more
likely that the timing of major leaf shedding events
(cacao: SeptemberOctober; Gliricidia: FebruaryApril
and SeptemberOctober) reacted to the long-term meteorological record where August was the driest month
(o100 mm rainfall month1). Studies from Brazil, Costa
Rica, and Malaysia demonstrated that maximum cacao
leaf fall coincides with low rainfall or drought (Alvim
et al., 1974; Ling, 1986; Heuveldop et al., 1988). Our

results also agree with observations of Gliricidia in

semiarid agroforestry systems in Kenya, where Gliricidia leaf cover declined during the dry season (Broadhead et al., 2003).
Cacao bean production (0.74 Mg ha1 yr1, March
2007February 2008) accounted for only 6% of the stand
NPP (G. Moser, unpublished results). The yield of this
6-year-old cacao stand is within the range (0.28
1.4 Mg ha1 yr1) reported from shaded cacao stands
across the tropics (Aranguren et al., 1982; Ling, 1986;
Hartemink, 2005) but considerably lower than the dry
bean yield (26 Mg ha1 yr1) used by Zuidema et al.
(2005) for the validation of their physiological production model. Cacao yield varies widely depending on the
age of the stand, climatic conditions, soil fertility, and
management (Alvim, 1977; Hartemink, 2005). Peak cacao yield in 2008 was 40% lower compared with 2007,
which may be attributed to the fact that pesticides were
not applied during the experiment. This may have
resulted in higher infestation rates, which was confirmed by our data as 485% of pods were affected by
pests and/or diseases.

Drought effects on water use characteristics

Perhaps the most striking result of our study was the
only moderate reduction of sap flux density and stand
transpiration in the roof plots even after 13 months of
throughfall reduction. This was especially clear for
cacao where sap flux density was on average 89% of
control. Although the overall decline of Gliricidia sap
flux densities in roof plots were similar to cacao, sap
flux of Gliricidia was more affected than cacao during
the period of lowest soil water availability between
January and March 2008. During this time Gliricidia
sap flux densities in roof plots were only 6575% of
control plots. Several factors may explain these observations:
(1) The species differed in their water use characteristics.
Mean daily water use rate of cacao plants in control
plots was 10 kg day1 compared with 14 kg day1 for
Gliricidia (Kohler et al., in press). Higher water use rate
of Gliricidia trees can be explained by a larger conductive sapwood area and tree crown exposure. At our site
cacao fine root biomass was not affected by the drought.
However, cacao coarse root water potential in the roof
plots was significantly lower compared with control
plots suggesting osmotic adjustment (G. Moser, unpublished results). Osmotic adjustment is supposed to
maintain turgor-dependent processes to a lower water
potential and contribute to desiccation delay (Kozlowski et al., 1991). In contrast, no osmotic adjustment was
found for Gliricidia.

r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 15151530

E X P E R I M E N TA L D R O U G H T I N C A C A O A G R O F O R E S T R Y
(2) Water uptake was partitioned vertically between cacao
and Gliricidia. An analysis of the dD in water extracted
from cacao and Gliricidia plant water and soil water at
different depths conducted during the pretreatment
period indicated that cacao obtained most of its water
from the upper horizon, while Gliricidia obtained more
of its water from deeper soil layers. This vertical partitioning in water uptake may limit competition between
cacao and Gliricidia for soil water resources. The reliance on deeper water sources may also explain why
Gliricidia sap flux densities recovered slower after roof
opening as it took some time to rewet lower soil
compartments.
(3) Cacao may have benefited from water input through
small openings in the roof. In our experiment approximately 20% of the throughfall (plus stemflow) reached
the soil surface. As storm events were frequent (Fig. 2e),
and cacao predominantly took up water from the
topsoil (Fig. 5a), it probably profited more from the
small amount of throughfall reaching the soil surface
(Fig. 2a) than the Gliricidia trees.
(4) Benchmark values such as permanent wilting point
(PWP) or relative extractable soil water (REW) are misleading. A REW value of about 0.4 is considered as a
physiological threshold at which soil water content
begins to limit transpiration rates due to stomatal
closure (Granier et al., 1999). Our data shows that sap
flux densities only declined considerably (roof to contol
ratio o0.75) when REW was o0.1 indicating that both
species are quite tolerant to low soil water availability.
(5) Acclimation to drought may have taken place in both
tree species. In cacao and Gliricidia leaves, d13C values
and intrinsic water use efficiencies (iWUE) increased
with decreasing water availability in the roof plots
(A. Camejo Diaz & L. Schwendenmann, unpublished
results) implying changes in stomatal control of transpirational water loss (Farquhar et al., 1989).

Drought effects on leaf litterfall and soil CO2 efflux

Similarly, leaf litterfall did not react as strongly to the
induced drought as we expected: changes in leaf litterfall biomass and timing were marginal. Cacao tended to
shed more leaves shortly after roof closure. The lack of
strong changes may partly be explained by the similarity of atmospheric conditions of roof and control plots.
Maintaining soil water potentials at or above field
capacity by irrigation for five consecutive dry seasons
at two 2.25 ha plots of a tropical moist forest on Barro
Colorado Island, Panama, showed that the timing of
leaf fall in 495% of the species considered did not differ
between irrigated and control plots (Wright & Cornejo,
1990). The absence of an effect of forest irrigation on tree
leaf fall suggests that plant water stress may not solely

1527

explain leaf fall. Wright & Cornejo (1990) suggested that

changes in atmospheric conditions (that were not altered throughout the experiment) may be more important phenological triggers.
In contrast, Gliricidia leaf litterfall appeared to have
declined in response to the throughfall reduction, especially in the first months of roof closure (Fig. 3e). The
reduction in leaf litter biomass that we found for Gliricidia
was also observed in a moist tropical forest in Amazonia.
Fine litterfall (leaves, reproductive parts, and twigs
o1 cm) quickly decreased in response to partial throughfall exclusion, indicating a reduction in the rate of leaf
production (Nepstad et al., 2002; Brando et al., 2008).
Soil CO2 efflux decreased marginally in the roof plots
compared with the control and the decrease was not as
strong as we anticipated at the beginning of the experiment. No effect of the treatment on soil CO2 efflux was
observed over the course of 5 years (Davidson et al.,
2008). The nonlinear relation between soil water storage
and soil CO2 efflux may explain the marginal difference
in measured soil CO2 efflux between roof and control
plots. While soil CO2 efflux decreased with decreasing
soil water storage in the roof plots the opposite was
observed in the control plots where CO2 efflux decreased with increasing soil water storage (Fig. 6). As
a result a decrease in soil CO2 efflux due to water stress
may have occurred at a time when soil water storage in
the control plots was not optimal for soil CO2 efflux
either. Similar nonlinear controls of soil water on soil
CO2 efflux were also observed in a tropical forest
ecosystem in Costa Rica (Schwendenmann et al., 2003)
and in a drought experiment in an Amazonian rain
forests (Sotta et al., 2007).

Drought effects on cacao yield and losses to pests and

diseases
The cacao yield in roof plots was substantially lower
compared with control plots during the main harvest in
2008 (post-treatment period). Cacao takes about 56
months to complete the development from flower to
pod in the study region (Y. Clough, unpublished results). In a given stand, assuming similar losses to pests
and diseases, yield is thus influenced by environmental
conditions during flowering, pollination, fruit set as
well as during fruit development, and maturation. Most
flowers in control and roof plots were found between
November 2007 and March 2008. However, the number
of flowers per tree was considerably lower in roof plots
during the dry spell in January/February 2008 as compared with control plots (G. Moser, unpublished results). Further, low water availability most likely
affected development and maturation resulting in a
lower number of cacao pods. As we did not observe

r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 15151530

1528 L . S C H W E N D E N M A N N et al.
differences in stem increment (G. Moser, unpublished
results) and leaf litterfall between roof and control plots,
this indicates that the reproduction was more sensitive
to drought than vegetative growth and fine litter production corroborated by Nepstad et al. (2002) who
found some preliminary evidence that fruiting phenology in a tropical rain forest was more sensitive to
drought than fine litter production.
The observed yield loss was higher than the reduction
obtained from a socioeconomic survey in Central Sulawesi (Keil et al., 2008) where farmers reported a decline
of up to 38% of average cacao yield levels after strong
ENSO-related droughts in 1997 and 2002. In January
2002 rainfall in the study region was as low as in January
2008. According to Keil et al. (2008) most of the droughtaffected farm households were forced to substantially
reduce expenditures for food and other basic necessities
to adapt to the reduction in agricultural income. In
addition, the level of preparedness for the case of
drought is low and risk management often includes
environmentally damaging and illegal activities such as
rattan extraction from protected forests (Keil et al., 2008).
The physiological modeling study by Zuidema et al.
(2005) suggested that droughts may induce yield reductions as high as 50% which is close to the result we
observed in our experiment. Yield reduction in our
experiment was closely linked to the two driest months
(January and February 2008) which was also found by
Zuidema et al. (2005).
Infection rates of cacao pods by pest insects and fungal
disease were high (485%) in both control and roof plots.
Bos et al. (2007) and Y. Clough (unpublished results) both
investigated cacao pests and pathogens in the vicinity of
our study site and also reported a high proportion of
damage caused by black pod, cacao pod borer, and
Helopeltis (see also Hebbar, 2007). Yield losses due to
black pod disease P. palmivora have increased in Sulawesi
in 2007 and 2008 due to increased rainfall (Y. Clough,
unpublished results). Our data suggests that low humidity due to low rainfall led to a significant decrease in
black pod occurrence between January and April 2008.
Herbivory caused by Helopeltis and cacao pod borer was
initially expected to be higher at the wet end of precipitation gradients (Connell, 1971), but recent evidence
suggests higher herbivory takes place at intermediate
precipitation levels (Marquis et al., 2002). In this study,
we did not find evidence of changes in occurrence of
insect damage caused by the induced drought.

Mechanisms that may have helped the agroforest to cope

with drought
As mentioned above, several species-specific physiological mechanisms and traits may explain why this stand

was able to cope relatively well with reduced soil water

availability. We further found indications that this
cacao/Gliricidia agroforest may have additional mechanisms that may reduce the impact of droughts,
which are absent in monocultures and/or annual crop
systems.
Firstly, shade trees may reduce evaporative demand.
Contrary to our expectations, cacao showed only minor
reductions in sap flux density. Although our setup was
not designed to test the overall effect of shade trees, we
did find some indications that cacao plants may have
profited from the (moderate) shade provided by the
Gliricidia trees. A low gap fraction, indicating a dense
Gliricidia cover, led to a reduction in cacao sap flux
densities in the roof plots (Fig. 4). This may be attributed to reduced radiation and VPD. Therefore it may be
expected that a high gap fraction would result in high
cacao sap flux density as evaporative demand would be
highest at locations without shade. In contrast, cacao
trees in roof plots did not exhibit an increase in sap flux
density with gap fraction. In fact, we even measured
lower sap flux densities where the gap fraction was
high. This may be explained by the concurrent soil
drought. Our data suggest that cacao trees reached
highest sap flux densities and may thus have realized
transpirational demands best under intermediate shade
levels. However, the relationship between sap flux
density and gap fraction does not necessarily indicate
higher photosynthetic carbon gain at intermediate gap
fraction above cacao trees. Trees with relatively closed
stomata in the open may still have been photosynthesizing faster than the shaded trees in the intermediate
gap fraction areas. More detailed studies will be necessary to derive a precise conclusion.
Secondly, use of soil water sources is complementary
between species. This was inferred from the vertical
partitioning in water uptake between cacao and Gliricidia under moist soil conditions. Gliricidia had a higher
fine root density (biomass per soil volume) in deeper
layers compared with cacao and is thus able to explore a
larger soil volume for water and nutrients, which
would be advantageous during dry periods. Finally,
the superficial rooting pattern of cacao enabled it to
extract the small throughfall and stemflow inputs that
only recharged the topsoil.
In summary, our study suggests that shade trees may
be relevant for this agroforestry system to cope with
droughts. An earlier study indicated that Gliricidia trees
can be beneficial for the nitrogen balance of cacao/
Gliricidia agroforests (Corre et al., 2006). Here, we have
shown that competition between cacao and Gliricidia for
soil water resources may be limited, and suggest that
the shade trees may even help the system to cope with
droughts. Cacao/Gliricidia agroforests may thus play a

r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 15151530

E X P E R I M E N TA L D R O U G H T I N C A C A O A G R O F O R E S T R Y
critical role in minimizing the vulnerability of farmers
livelihood to extreme weather events such as droughts.

Acknowledgements
This study was conducted in the framework of the joint Indonesian-German research project Stability of Tropical Rainforest
Margins in Indonesia (STORMA) funded by the Deutsche Forschungsgemeinschaft, DFG (SFB 552). We thank Thomas Kluter
and his team for roof construction and maintenance; Pak Andi
Sofyan for his assistance with the field work; three anonymous
reviewers and Eric A. Davidson for constructive reviews and
suggestions; the German and Indonesian project coordination for
technical support, and LIPI, the Indonesian Research Institute,
for the research permit.

References
Alvim PT (1977) Cacao. In: Ecophysiology of Tropical Crops (eds Alvim PT,
Kozlowski TT), pp. 279313. Academic Press, London.
Alvim PT, Alvim R (1978) Relation of climate to growth periodicity in
tropical trees. In: Tropical Trees as Living Systems (eds Tomlinson PB,
Zimmermann MH), pp. 445464. Cambridge University Press, Cambridge, UK.
Alvim PT, Machado AD, Vello F (1974) Physiological responses of cacao to
environmental factors. Revista Theobroma, 4, 325.
Aranguren J, Escalante G, Herrera R (1982) Nitrogen cycle of tropical
perennial crops under shade trees. II. Cacao. Plant and Soil, 67, 259269.
Beer J (1988) Litter production and nutrient cycling in coffee (Coffea
arabica) or cacao (Theobroma cacao) plantations with shade trees. Agroforestry Systems, 7, 103114.
Beer JW (1987) Advantages, disadvantages and desirable characteristics
of shade trees for coffee, cacao and tea. Agroforestry Systems, 5, 313.
Belsky JM, Siebert SF (2003) Cultivating cacao: implications of sun-grown
cacao on local food security and environmental sustainability. Agriculture and Human Values, 20, 277285.
Bonhomme R, Chartier P (1972) The interpretation and automatic measurement of hemispherical photographs to obtain sunlit foliage. Israel
Journal of Agricultural Research, 22, 5361.
Bos MM, Steffan-Dewenter I, Tscharntke T (2007) Shade tree management
affects fruit abortion, insect pests and pathogens of cacao. Agriculture,
Ecosystems and Environment, 120, 201205.
Brando PM, Nepstad DC, Davidson EA, Trumbore SE, Ray D, Camargo P
(2008) Drought effects on litterfall, wood production, and belowground
carbon cycling in an Amazon forest: results of a throughfall reduction
experiment. Philosophical Transactions of the Royal Society B, 363, 1839
1848.
Broadhead JS, Black CR, Ong CK (2003) Tree leafing phenology and crop
productivity in semi-arid agroforestry systems in Kenya. Agroforestry
Systems, 58, 137148.
Brunel JP, Walker GR, Kenneth-Smith AK (1995) Field validation of
isotopic procedures for determining sources of water used by plants
in a semi-arid environment. Journal of Hydrology, 167, 351368.
Clark DA, Brown S, Kicklighter DW, Chambers JQ, Thomlinson JR, Ni J,
Holland EA (2001) Net primary production in tropical forests: an
evaluation and synthesis of existing field data. Ecological Applications,
11, 371384.
Connell JH (1971) On the role of natural enemies in preventing competitive exclusion in some marine animals and rain forest trees. In:
Dynamics of Populations (eds den Boer PJ, Gradwell GR), pp. 298312.
PUDOC, Wageningen.

1529

Corre MD, Dechert G, Veldkamp E (2006) Soil nitrogen cycling following

montane forests conversion in Central Sulawesi, Indonesia. Soil Science
Society of America Journal, 70, 359366.
Davidson EA, Nepstad DC, Ishida FY, Brando PM (2008) Effects of an
experimental drought and recovery on soil emissions of carbon dioxide, methane, nitrous oxide, and nitric oxide in a moist tropical forest.
Global Change Biology, 14, 25822590.
Ehleringer JR, Osmond CB (1989) Stable isotopes. In: Plant Physiological
Ecology Field Methods and Instrumentation (eds Pearcy RW, Ehleringer JR,
Mooney HA, Rundel PW), pp. 281300. Chapman and Hall Ltd,
London.
Farquhar GD, Ehleringer JR, Hubick KT (1989) Carbon isotope discrimination and photosynthesis. Annual Review of Plant Physiology and Plant
Molecular Biology, 40, 503537.
Fisher RA, Williams M, Lola da Costa A, Mahli Y, da Costa RF, Almeida S,
Meir P (2007) The response of an Eastern Amazonian rain forest to
drought stress: results and modelling analyses from a throughfall
exclusion experiment. Global Change Biology, 13, 23612378.
Garca Santos G (2007) An ecohydrological and soils study in a montane cloud
forest in the National Park of Garajonay, La Gomera (Canary Islands, Spain).
PhD Dissertation, Vrije Universiteit, Amsterdam, the Netherlands.
Available at http://hdl.handle.net/1871/12697 (accessed 7 April 2009).
Gehre M, Geilmann H, Richter J, Werner RA, Brand WA (2004) Continuous flow 2H/1H and 18O/16O analysis of water samples with
dual inlet precision. Rapid Communication Mass Spectrometry, 18,
26502660.
Granier A (1985) A new method of sap flow measurement in tree stems.
Annals of Forest Science, 42, 193200.
Granier A (1987) Evaluation of transpiration in a Douglas-fir stand by
means of sap flow measurements. Tree Physiology, 3, 309320.
Granier A, Breda N, Biron P, Villette S (1999) A lumped water balance
model to evaluate duration and intensity of drought constraints in
forest stands. Ecological Modelling, 116, 269283.
Gundersen P, Andersen BR, Beier C, Rasmussen L (1995) Experimental
manipulations of water and nutrient input to a Norway spruce
plantation at Klosterhede, Denmark: I. Unintended physical and
chemical changes by roof experiments. Plant and Soil, 168169,
601611.
Hanson PJ (2000) Large-scale Water Manipulations. In: Methods in Ecosystem Science (eds Sala OE, Jackson RB, Mooney HA, Howarth RW),
pp. 341352. Springer-Verlag, New York.
Hanson PJ, Todd DE, Amthor JS (2001) A six-year study of sapling and
large-tree growth and mortality responses to natural and induced
variability in precipitation and throughfall. Tree Physiology, 21, 345358.
Hartemink AE (2005) Nutrient stocks, nutrient cycling, and soil changes
in cocoa ecosystems: a review. Advances in Agronomy, 86, 227253.
Hebbar PK (2007) Cacao diseases: a global perspective from an industry
point of view. Phytopathology, 97, 16581663.
Heuveldop J, Fassbender HW, Alpizar L, Enriquez G, Folster H (1988)
Modelling agroforestry systems of cacao (Theobroma cacao) with laurel
(Cordia alliodora) and poro (Erythrina poeppigiana) in Costa Rica. II.
Cacao and wood production, litter production and decomposition.
Agroforestry Systems, 6, 3748.
INRA (2007) Can-Eye. French National Institute for Agricultural Research,
Avignon, France. Available at http://www.avignon.inra.fr/can_eye
(accessed 9 December 2008).
Keil A, Zeller M, Wida A, Sanim B, Birner R (2008) What determines
farmers resilience towards ENSO-related drought? An empirical assessment in Central Sulawesi, Indonesia. Climatic Change, 86, 291307.
Kohler M, Dierick D, Schwendenmann L, Holscher D (in press) Water use
characteristics of cacao and Gliricidia trees in an agroforest in Central
Sulawesi. Ecohydrology, doi: 10.1002/eco.67.
Kozlowski TT, Kramer PJ, Pallardy SG (1991) The Physiological Ecology of
Woody Plants. Academic Press, San Diego.

r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 15151530

1530 L . S C H W E N D E N M A N N et al.
Lamersdorf NP, Beier C, Blanck K et al. (1998) Effect of drought experiments using roof installations on acidification/nitrification of soils.
Forest Ecology and Management, 101, 95109.
Lang ARG, Xiang YQ (1986) Estimation of leaf area index from transmission of direct sunlight in discontinuous canopies. Agricultural and Forest
Meteorology, 37, 229243.
Lin B, Perfecto I, Vandermeer J (2008) Synergies between agricultural
intensification and climate change could create surprising vulnerabilities for crops. BioScience, 58, 847954.
Ling AH (1986) Litter production and nutrient cycling in a mature cocoa
plantation on inland soils of Peninsular Malaysia. In: Proceedings of the
International Conference on Cocoa and Coconuts, Kuala Lumpur (eds
Pushparajah E, Chew PS), pp. 451465. Incorporated Society of Planters, Kuala Lumpur, Malaysia.
Marquis RJ, Morais HC, Diniz IR (2002) Interactions among cerrado
plants and their herbivores: unique or typical? In: The Cerrados of Brazil:
Ecology and Natural History of a Neotropical Savanna (eds Oliveira PS,
Marquis RJ), pp. 306328. Columbia University Press, New York.
Nepstad DC, Moutinho P, Dias-Filho MB et al. (2002) The effects of partial
throughfall exclusion on canopy processes, aboveground production,
and biogeochemistry of an Amazon forest. Journal of Geophysical
Research, 107, 53.153.18.
Purseglove J (1968) Tropical Crops: Dicotyledons. Longman, Harlow, UK.
Rada F, Jaimez RE, Garcia-Nunez C, Azocar A, Ramrez ME (2005) Water
relations and gas exchange in Theobroma cacao var. Guasare under
periods of water deficit. Revista de la Falcultad de Agronomia, 22, 112120.
Rice RA, Greenberg R (2000) Cacao cultivation and the conservation of
biological diversity. Ambio, 29, 167173.
Schwendenmann L, Veldkamp E, Brenes T, OBrien JJ, Mackensen J (2003)
Spatial and temporal variation in soil CO2 efflux in an old-growth
neotropical rain forest, La Selva, Costa Rica. Biogeochemistry, 64, 111128.
Sheffield J, Wood EF (2008) Projected changes in drought occurrence
under future global warming from multi-model, multi-scenario, IPCC
AR4 simulations. Climate Dynamics, 31, 79105.

Slingo JM, Challinor AJ, Hoskins BJ, Wheeler TR (2005) Introduction: food
crops in a changing climate. Philosophical Transactions of the Royal Society
B, 360, 19831989.
Sotta ED, Veldkamp E, Schwendenmann L et al. (2007) Effects of an
induced drought on the soil CO2 production and soil CO2 efflux in an
Eastern Amazonian Rainforest, Brazil. Global Change Biology, 13, 2218
2229.
Topp GC (1971) Soil water hysteresis in silt loam and clay loam soils.
Water Resources Research, 7, 914920.
van Noordwijk M, Lawson G, Soumare A, Groot JJR, Hairiah K (1996)
Root distribution of trees and crops: competition and/or complementarity. In: Tree-Crop Interactions (eds Ong CK, Huxley P), pp. 319364.
CAB International, Wallingford.
Veldkamp E, OBrien JJ (2000) Calibration of a frequency domain reflectometry sensor for humid tropical soils of volcanic origin. Soil Science
Society of America Journal, 64, 15491553.
Verchot LV, van Noordwijk M, Kandji S et al. (2007) Climate change:
linking adaptation and mitigation through agroforestry. Mitigation and
Adaptation Strategies for Global Change, 12, 901918.
Vincke C, Thiry Y (2008) Water table is a relevant source for water uptake
by a Scots pine (Pinus sylvestris L.) stand: Evidences from continuous
evapotranspiration and water table monitoring. Agricultural and Forest
Meteorology, 148, 14191432.
Vitousek PM (1984) Litterfall, nutrient cycling and nutrient limitation in
tropical forests. Ecology, 65, 285298.
Wright SJ, Cornejo FH (1990) Seasonal drought and leaf fall in a tropical
forest. Ecology, 71, 11651175.
Young AM (1994) The Chocolate Tree. A Natural History of Cacao. Smithsonian Institution Press, Washington, DC.
Zuidema PA, Leffelaar PA, Gerritsma W, Mommer L, Anten NPR (2005) A
physiological production model for cocoa (Theobroma cacao): model
presentation, validation and application. Agricultural Systems, 84, 195
225.

r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 15151530