Professional Documents
Culture Documents
Design and Analysis of Agroforestry
Design and Analysis of Agroforestry
OF AGROFORESTRY EXPERIMENTS
Seema Jaggi
I.A.S.R.I, Library Avenue, New Delhi-110 012
1. Introduction
Agroforestry is a collective name for land-use systems and technologies where woody
perennials (trees, shrubs, bamboos etc.) are deliberately grown on the same land-
management units as agricultural crops and/or animals, in some form of spatial
arrangement or temporal sequence. This definition implies that an agroforestry system
normally involves two or more species of plants, atleast one of which is a woody
perennial, always has two or more outputs, and has the cycle of more than one year.
The influence of one component of the system on the performance of the other as well as
the system as a whole is referred to as component interactions. These interactions are also
called as the tree-crop interaction. These interactions could be positive or negative.
Positive interactions include biomass contribution, water conservation, soil conservation,
etc. and negative interactions are competition for light, nutrient, water etc.
Agroforestry systems are more complex than the mono-cropping system. Several
characteristics of the trees like slow growth, long term effects on their surroundings, long
life, age of trees, the area over which the influence of trees extend etc. complicate the
issue of designing experiments for these systems. In addition to the basic principles i.e.
randomization, replication and local control, there are several other factors that need to
be taken care of while planning agroforestry experiments. For example, within one
experimental plot there may be crop rows, individual trees or hedges with different
treatments applied to each and each component may respond to its own treatment as well
as to the treatments applied to other components.
The agroforestry experiments may be broadly classified under following four heads:
The arrangement of tree and crop components in relation to one another within plots
deserves important consideration, especially in interaction studies. Therefore, use of an
appropriate design is a very important aspect of agroforestry experimentation. A general
recommendation or solution cannot be given for all the situations since a specific
problem requires a particular design.
The plot size is also an important consideration in agroforestry experiments. The size of
the plot depends on many factors like the objectives of the experiment, types of
measurement to be made, the expected duration of the experiment, likely ultimate size of
the trees, the requirement for extra space to avoid interferance between plots etc. For
example, MPT selection trials involve large number of species or provenances and if the
focus is on tree survival and growth, each species can be allocated a small plot (20-30
m2). Larger plots (50-200 m2) are needed for experiments designed to test species for
particular agroforestry technologies or to study the effects of management practices
[Roger and Rao, 1990]. Choosing suitable sample from which to measure the response is
another important aspect of these experiments. Rao and Coe (1992) have discussed some
aspects related to this.
The designs used in agroforestry experimentation along with the layouts where-ever
possible will be described in next section. Some analytical techniques, other than the
usual analysis as per the design adopted, are also described. It is worthwhile to mention
here that several types of analyses of data resulting from agroforestry experiments can
now be performed easily because of sophisticated and flexible statistical computer
software packages like SAS and SPSS.
2. Experimental Designs
To obtain valid results from agroforestry experiments, it is important to follow scientific
principles. The research methods should take into account the unexplainable variability in
the experimental material. In view of the large number of factors to be studied in the
system, the number of combinations may become too unwieldy to be evaluated in
conventional designs. However conventional designs are being used in some situations.
Jaggi et al (2001) have given an overview of the designs used in agroforestry
experimentation along with the layout plans where-ever possible.
• Randomized Designs
The most widely used design because of its simplicity is Randomized Complete Block
(RCB) Design. In one of the experiment planned in 1986 at Fatehpur Shekhawati
(Rajasthan), the objective was to identify suitable crop under different tree species. The
design adopted was RCB DESIGN with five crops viz. Moong, Moth, Cowpea, Guar and
454
Agroforestry Experiments
Bajra each with two varieties viz. local and improved taken under each of the three trees
viz. Acacia Nilotica, Acacia Tortilis and Dalbergia Sissoo, the replications in the design
were two.
There are many situations in agroforestry where a RCB DESIGN may not be appropriate.
For example, in MPT evaluation studies or in germplasm evaluation trials, the number of
treatments may be too large so that it is difficult to locate a block with sufficient
uniformity to accomodate a complete set of all treatments to be tested. In such cases
Incomplete Block designs such as lattice designs, alpha designs, confounded designs
may be adopted.
Another commonly used design is the Split-Plot design. For example, in an alley-
cropping experiment under the component and system management trials, for testing
hedgerow species and hedgerow pruning management, species could be allocated to main
plots and pruning management (frequency, height of cutting, etc.) to subplots of a split-
plot design. Split-plot experiments are particularly useful if the treatments are susceptible
to ‘neighbour effects’ as in irrigation experiments or in experiments involving tree
species of varying growth habits, especially height. But these experiments should be
avoided as far as possible because the treatments in the main-plots are compared with
less precision than those in the subplots. The split-plot design could be further extended
to Split-split plot design. The strong recommendation is that split plot designs should
only be used when practical considerations of plot size require it.
Related to the split-plot design is the Strip-plot design that is specifically suited for two
factor experiments when both factors require large plot sizes and when the interaction
between the two factors is desired to be measured more precisely than the effects of
either of the two factors.
• Systematic designs
In some circumstances, randomization is impracticable or unwise. For example in
spacing trials, the extreme treatments must not adjoin. The designs for such situations are
called Systematic designs and these can be used in agroforestry experiments (Huxley,
1985b). Many factors of interest in agroforestry research do not allow randomization,
such as slope, orientation, rainfall or distance from tree or hedge. These designs despite
their limitations have been proposed for agroforestry research because they minimize the
requirement for field and experimental resources. The concept of systematic design is
that over a series of plots the level of the quantitative factor varies systematically but
slowly (e.g.10% change) through the sequence of plots. Each particular treatment level is
then surrounded by treatment levels differing only slightly, and therefore guard areas are
unnecessary. The harvested area on the systematic design is much more than in any other
randomised design conducted on the same piece of land. For example, in alley cropping
trials, the spacing between hedgerows (alley width) can increase gradually across a site
rather than having plots of varying alley widths located at random. Another example is an
alley cropping trial where distance of crop rows from hedgerows is the treatment variable
so that the crop row nearest the hedgerow will be compared against other rows that are
455
Agroforestry Experiments
progressively further from the hedgerow. The direction of the systematic change of the
treatment level within each plot can and should be randomly selected.
Fan designs (Nelder, 1962) and Parallel row layouts (Bleasdale, 1967) are the
systematic designs normally used to investigate sole crop spacing problems with a wide
variety of plant material. For agroforestry mixed cropping experiments, including tree-
crop interface studies, spacings of both components is an important variable and
systematic design can be used to explore the effects. For example to select a suitable
tree/bush species and agricultural crop associates, a simple Parallel row design for the
tree species, with replicated strips of the various agricultural crops sown across this at
right angles could be a suitable layout. Fan designs are a kind of systematic design that
involve planting on rays emanating from the centre of an arc of a circle, the angle
between successive rays usually being about 50. In tree spacing trials, the distance
between trees could be increased systematically. If the object of the experiment is to
compare inter-row spacings, the plant density along a row can be kept constant.
The statistical analysis of the data from systematic design is little complex. Whatever
kind of systematic design or fan is used, the data should not be regarded as coming from
a randomized experiment. The analysis of variance would not be the most appropriate
initial step in the analysis. What it expresses is a relationship between, say, growth and
the character i.e. area, inter-row spacing or rectangularity, etc. Hence the appropriate
statistical tool is fitting of a response function to represent the effect of the systematic
factor for each set of data. The subsequent analysis is based on the comparisons of
parameters of the fitted response functions. Such comparisons may be between replicates
of the systematic plot or between different treatments applied to the different plots within
which the systematic treatment level variation occurs. Mead (1967) has examined the
general problem of spacing and has given a mathematical model for the estimation of
inter-plant competition. Randomization in the systematic design can be introduced at a
higher level (Nelder, 1962). This is done by repeating the sets of systematically arranged
treatments at different locations within the experimental site. We now describe some
more systematic designs.
Y Design
The kind of field layout used for tree-crop interface (TCI) investigations i.e. component
interaction studies will depend on the precise research objectives and the resources
available, including the extent and nature of the site. Orientation may have a significant
role at TCI because of the shading effect of tall trees on the understorey crop component.
Therefore, orientation as a factor may also be considered. Huxley (1985a) suggested that
TCI and orientation effects can be studied in simple geometrical designs. Huxley
described a Y design (or 1200 design) with three strips (or arms) each containing the tree
and crop components juxtaposed to each other. This design enables a study of the TCI in
a relatively small space and reduces interfaces between the different arms, especially
with tall trees. This design takes care of the orientation effects.
456
Agroforestry Experiments
described and some results have been presented to illustrate the kind of information that
can be obtained. The objective of one of the experiments was to study TCI effects and
interactions and to investigate assessment methodologies. The species used was Cassia
Siamea/ maize. The trial was planted in May 1984 and the layout consisted of a
geometric 1200 design with trees in strip plots as shown in Fig.1. The details regarding
plot size is shown in the layout.
Fig.1: Experimental layout using a ‘Y’ design with three arms. Each arm is bracketed by crop plots that are
subplots and are numbered anticlockwise starting in the north-east corner i.e. 1 & 2 at 00, 3 & 4 at
2400 and 5 & 6 at 1200 respectively. Subplots are further divided by tilled and no-tilled treatments.
The possibilities of shelter and shading effects due to either orientation or the presence of
a mixture of differently structured plant components (tree and crop) has to be taken into
account in selecting a suitable form of data analysis. The data has been analyzed by
taking the factor strips as the main plot treatments and the two sides of the strip as the
subplot treatment of a split plot design.
Star Design
Rao et al. (1991) have described a TCI experiment designed to investigate the effects of
row orientation and to evaluate the potential of hedgerow intercropping. A study was
conducted at ICRISAT with a hedge crop and annual crops to evaluate the potential of
hedgerow intercropping in semi-arid tropical India. The layout was designed to allow a
test of a wider range of orientations than in the Y design. The experimental design used is
a Star design (Fig.2).
From the centre of the layout there are eight arms (or plots) at 450 angle with rows
orientated in all directions giving the shape of a Star. Half of the ‘star’ with four arms
representing the four row directions are taken as replication. The whole star can again be
replicated for more replications for the trial. In each plot or arm, hedge crop is planted in
the middle and annual crops in rows are sown on either side of the hedge. A sole hedge is
taken in the middle arm. The crop rows beyond the influence of hedge are regarded as
sole crop. Each plot is 11.4m wide, but the length varied with row direction, which was
13.5m in E-W and N-S directions and 10.5m in the other two directions.
457
Agroforestry Experiments
Fig.2: Field layout of two replications (one star) showing plot arrangement for studying the orientation and
interface effects in Leucaena/ Annual crop.
For the analysis there are three factors in the layout of a star design: row orientation as
main-plots, the two halves of a plot, one on each side, as sub-plots to see the effect on
crop rows due to their location on the windward or leeward side of the hedge and crop
rows at different distances away from the hedge as sub-sub plots. Because of the row
orientations the sides and row positions can not be randomised. Annual crop yields are
analysed as per the analysis of variance of split-split plot design as shown below:
ANOVA
Source of Variation Degrees of Freedom
Replication 2r-1
Orientation 3
Error (a) 3(2r-1)
Side 1
Orientation x Side 3
Error (b) 4 x 1 x (2r-1)
Rows p-1
Orientation x Rows 3 x (p-1)
Side x Rows p-1
Orientation x Side x Rows 3 x 1 x (p-1)
Error (c) 8 x (p-1) x (2r-1)
Total (4 x 2 x 2r x p) –1
r is the number of times the whole star is replicated and p is the number of crop rows on
either side of the hedge. Since all crop rows of a plot are not affected by hedge only (end
rows may be influenced by edge effect), therefore, individual row yields can be analysed
as per split-plot to identify the effects of hedge, orientation and their interaction.
Individual hedge yields from interface and sole plot can be subjected to ANOVA of RCB
DESIGN.
458
Agroforestry Experiments
Regarding analysis, crop yields and leucaena fodder and wood yields were analyzed as a
RBD in order to have an approximate measure of experimental variability. The
relationship of leucaena yield-plant population and hedgerow performance-space
between
hedge and annual crops was fitted with regression equations.
Fig.3: Layout of 2-way systematic design for studying the hedgerow spacing and cropping intensity.
Augmented Design
Pinney (1991) has made use of Augmented design for On-farm trials or prototype
evaluation trials. On-farm trials are an essential means of technology transfer and can be
viewed as a means of providing feed-back from the farm to the research station. It is a
very powerful and appropriate strategy for agroforestry research, especially for the
applied type. The technology developed at research stations is tested under farm
conditions to obtain information about its performance and acceptability by the farmers.
459
Agroforestry Experiments
minimises plot number and enables the researcher’s and farmer’s questions to be
answered.
An augmented design contains both treatments that are replicated in every block and
those that are not. It allows the farmer some flexibility to decide what treatments are
tested on his/her farms. The technology developed at research station forms the set of
core treatments and the farmer defined treatments are called the augmented
treatments.
Pinney (1991) has described a hypothetical alley cropping example to illustrate the
application of augmented designs to participatory on-farm agroforestry research by
taking two core treatments (farmer’s existing practice and alley cropping) and five
augmented treatments. The number of plots per farm depend upon the region, population
density and farming system. The more the number of plots available per farm, the more is
the scope for within-farm replication or for more the treatment augmentation.
In the example considered by Pinney, there are two to four plots per farm. The design is
non-orthogonal and unbalanced. The data analysis has to be performed through a
generalised linear regression approach. The terms considered in the model for the
analysis are farm, treatments and their interactions. There could be grouping of plots
other than plots within farms that may add to the information gained from such a trial.
The analysis of variance for the given design can be of the form given below.
ANOVA
Sources of Variation Degrees of Freedom
Farms 19
Treatments 6
Core treatments 1
Augmented treatments 4
Core Vs. Augmented treatments 1
Error 37
Total 62
The seven treatments can be compared by the six orthogonal contrasts as (1 –1 0 0 0 0 0),
(0 0 1 –1 0 0 0), (0 0 1 1 –2 0 0), (0 0 1 1 1 –3 0), (0 0 1 1 1 1 -4), (5 5 -2 -2 -2 -2 -2).
Beehive Design
Beehive designs suggested by Martin (1973) are used for competition experiments. In
these designs there are two kinds of species of plants that are intermingled on a triangular
format in such a way that there is a wide range of competition. The figure shown below
has two species.
a
A A A B B
460
Agroforestry Experiments
A A* A B A+ B
A A B B B
B A A A A B
B B B B B
There are nine internal plants of A. The one marked * has all six neighbours of its own
kind, whereas that marked + has all six neighbours of B. All intermediate cases are also
shown. In this triangular format each plant is at the same distance from all its neighbours
i.e. each plant lies at the centre of a regular hexagon, three adjacent plants forming an
equilateral triangle. Plants are a distance ‘a’ apart in rows and rows are a distance 3 a/2
apart. For example: if such a format is adopted and one diseased plant is introduced
somewhere in the middle, useful information should become available about the spread
of the disease, whether it is downwind, along the direction of cultivation, strictly to
adjacent plants. Since each plant has six neighbours set on six equally spaced radii,
spread can take place in all directions with nearly equal ease.
Excluding the plants on the perimeter, there are 11 plants in all, eight of A. Of these
eight, one is surrounded by its own kind, one has one neighbour of B, another has two
such neighbours and so on. It is, therefore, possible to plot performance of these eight
plants against the number of similar neighbours and so determine competition effects. A
better design might enable the B-plants to be studied also, but with more plants. These
matters have been dealt with in Martin (1973) and Veevers and Boffey (1975). Mead
(1967) has considered a mathematical model for the estimation of interplant competition.
Mead (1979) has also considered competition experiments in generel.
The following table gives a summary of the experimental designs as discussed above that
have been adopted for different situations according to the objectives of the experiment.
461
Agroforestry Experiments
3. Analytical Techniques
In agroforestry experiments, there is not a single form of statistical analysis which is
appropriate to all forms of agroforestry data. Even for a single set of experimental data it
will be important to use several different forms of analysis. For the two components of an
agroforestry system, the data may occur in different structural forms. The data structures
from these experiments will be complex with different forms of yield information
available for different subsets of experimental units. In considering alternative
possibilities for the analysis of data from agroforestry experiments, it is essential that the
principle of comparing ‘like with like’ is followed. A comparison is valid only when the
units of measurement are identical.
Mead (1986) has described the statistical approaches to the analysis of agroforestry data.
Edge effects are important in agroforestry experiments. Langton (1990) has described the
methods of eliminating edge effects, either at the design or at the analysis stage using
computer simulation technique.
Apart from analysing the data according to the design adopted and studying the
significance of various treatment effects, some further analysis can be done through the
use of some of the powerful statistical techniques.
In some agroforestry systems, there is only one major output, such as crop yield in a
hedgerow intercropping system designed to enhance soil conservation or fertility. In such
a situation, the evaluation techniques used in agronomic trials are appropriate. One
technique is based on analysing the response under sole cropping and under agroforestry
and estimating the effect of the agroforestry system [Rao and Coe, 1992].
462
Agroforestry Experiments
Mead (1986) described the value of Bivariate analysis in intercropping research. This
technique could also be applied to agroforestry trials that have two key outputs. It
involves looking at the joint variation of the two outputs related to two variables. The two
variables (e.g. yield of a crop and height of the tree) will have an associated sum of
squares and the sum of products because of the relationship between them. The
construction of bivariate analysis of variance table is described in Pearce and Gilliver
(1978).
The approach to the problem of multiple outputs is to produce an index that synthesizes
them into a single value. This can be done by assigning weights to each characteristic,
reflecting an assessment of their relative importance. A way of indexing is the Land
equivalent ratio (LER).
where Ci and Ti are crop and tree yields under intercropping and Cs and Ts are yields in
the sole crop and sole tree system. It is defined as the land area in the system that would
be required to produce the same yields as one hectare of intercropping. The LER is only a
relative measure, it does not reveal the magnitude of yields. LER are only comparable if
they are based on the same tree-crop yields. Oyejola and Mead (1981) suggested that
analysis of variance for LER values should be reasonably valid provided that the
different divisors used in calculating the individual LER values are not large.
Multivariate techniques
Agroforestry experimentation results in different forms of information obtained from
different components of the system. When two species in the system are planted in the
same plot together, their yield will possibly be correlated. The data obtained is therefore
multivariate in nature and the univariate analysis of the components separately may not
result in correct inferences about different effects. The only form of analysis which
retains all the available information is multivariate data analysis. Oranga (1981) has
described the use of multivariate techniques for the analysis of data generated through
agroforestry experiments. These techniques include the principal component analysis
(PCA), discriminant function, canonical correlation, etc. For example, data on a number
of characters, that are related, is observed, both for tree as well crop components. These
characters could be combined through PCA to get uncorrelated principal components.
The first principal component contributes maximum towards the variability and can be
used for some further analysis. Jaggi and Pateria (2003) have described the application of
some of the multivariate techniques for analyzing the data obtained from agroforestry
experimentation.
Contrast Analysis
The procedure for making a comparison between treatments is to partition the treatment
sum of squares (SS) into meaningful components through contrast analysis. By
partitioning, the specific causes of the difference between treatment means can be
determined and the important ones can be identified. Each component of a partitioned
463
Agroforestry Experiments
In one of the experiments with 4 tree species and two crops, the experiment was laid out
in an RBD with 4x2 plots of combinations of trees and crops along with 2 sole crop plots
and 4 sole tree plots in three replications. Analysis was performed for one of the crop
under 4 tree species i.e. 5 treatments (crop under 4 tree species and a sole crop) in 3
replications The five treatments were compared through four orthogonal contrasts (1 –1 0
0 0), (1 1 –2 0 0), (1 1 1 –3 0), (1 1 1 1 –4). The last contrast compares the performance
of the crop with and without trees and was found to be significant.
Analysis of Covariance
One existing technique that may reduce some of the unexpected variation in the data is
the use of co-variates. For example, simple scores of pest damage or weediness of plots
can be used to reduce their impact on the overall variation in the trial. Successful use of
co-variates depends upon choosing appropriate ones. For example, the different
characters observed from tree (height, canopy, collar diameter, etc.) or a combination of
them like PCA, can be taken as a covariate in analysing the crop component in order to
eliminate the effect of trees on the crops.
Stability Analysis
Modified Stability Analysis (MSA) [Hilderbrand, 1984] is a procedure by which the data
from a wide range of environments (caused by variable biophysical conditions as well as
variations in management operations by different farmers) can be evaluated using both
researcher and farmer-oriented criteria. MSA is based on statistical method to assess
genotype x environment interactions. These physical or biophysical environments can be
expressed through an index based on the yield of different components at each location.
References
Bleasdale,J.K.A.(1967). Systematic designs for spacing experiments. Experimental
Agriculture, 3, 73-85.
Hilderbrand,P.E. (1984). Modified stability analysis of farmer-managed, on-farm trials.
Agronomy Journal, 76, 271-274.
Huxley,P.A.(1985a). The tree/crop interface -or simplifying the biological/
environmental study of mixed cropping agroforestry systems. Agroforestry
Systems, 3, 197-207.
Huxley,P.A. (1985b). Systematic designs for field experimentation with multipurpose
trees. Agroforestry Systems, 3, 251-266.
Huxley, P., Darnhofer, T., Pinney, A., Akunda, E. and Gatama, D. (1989). The Tree/Crop
Interface: A project designed to generate experimental methodology. Agroforestry
Abstracts, 2(4), 127-145.
Jaggi,S., Gupta,V.K. and Sharma,V.K. (2001). Design and analysis of agroforestry
experiments: An overview. Jour. Ind. Soc. of Agroforestry, 3(2), 120-129.
Langton,S. (1990). Avoiding edge effects in agroforestry experiments; the use of
neighbour-balanced designs and guard areas. Agroforestry Systems, 12, 173-185.
464
Agroforestry Experiments
Martin,F.B. (1973). Beehive designs for observing variety competition. Biometrics, 25,
397-402.
Mead,R. (1967). A mathematical model for the estimation of interplant competition.
Biometrics, 23, 189-205
Mead,R. (1986). Notes on 1) Units of observation, 2) Conventional designs for
agroforestry, 3) Bivariate analysis, 4) Nearest neighbour analysis methods, 5)
Statistical approaches to the analysis of agroforestry data. In proceedings of
ICAR-ICRAF workshop, Training course on Agroforestry research, 16-30 Sept.,
Hyderabad, India.
Mead,R. (1979). Competition experiments. Biometrics, 35, 41-54.
Nair,P.K.R. (1993). An introduction to Agroforestry. Kluwer Academic Publishers in
cooperation with ICRAF.
Nelder,J.A. (1962). New kinds of systematic designs for spacing experiments.
Biometrics, 18, 283-307.
Oranga,H.M. (1981). Multivariate statistical analysis in agroforestry research. In:
L.Buck(ed.) Proceeding of Kenya National Seminar on Agroforestry ICRAF,
Nairobi, 93-105.
Oyejola,B.A. and Mead,R. (1981). Statistical assessment of different ways of calculating
Land Equivalent Ratios (LER). Experimental Agriculture, 18, 125-138.
Pinney,A. (1991). Farmer-augmented designs for participatory agroforestry research.
Agroforestry Systems, 15, 259-274.
Pearce,S.C. and Gilliver,B. (1978). The statistical analysis of data from intercropping
experiments. Journal of Agricultural Sciences, 91, 625-632.
Rao,M.R. and Coe,R. (1992). Evaluating the results of agroforestry research.
Agroforestry Today, 4(1), 4-9.
Rao,M.R., Sharma,M.M. and Ong,C.K. (1990). A study of potential of hedgerow
intercropping in semi-arid India using a two-way systematic design. Agroforestry
Systems, 11, 243-258.
Rao,M.R., Sharma,M.M. and Ong,C.K. (1991). A tree/crop interface design and its use
for evaluating the potential of hedgerow intercropping. Agroforestry Systems, 13,
143-158.
Roger,J.H. and Rao,M.R. (1990). Agroforestry field experiments: Discovering the hard
facts Part I, Statistical considerations. Agroforestry Today, 2(1), 4-7.
Veevers,A. and Boffey,T.B. (1975). On the existence of leveled beehive designs.
Biometrics, 31, 963-967.
465