Agriculture Reference
In-Depth Information
Variates recorded
Whereas in the early generation stages there are many
thousands of test lines, there are very few characters
recorded on each line. In intermediate selection the
number of traits on which selection is based is increased,
often considerably.
Data will be collected prior to planting, throughout
the growing season, at harvest and post-harvest. A major
part of plant breeding is managing the vast data sets
which can arise and to interpret this information to
best advantage in selection.
always be evaluated in comparison to control entries
in replicated yield trials. Randomized Complete Block
(RCB) designs are commonly used for the first rounds
of intermediate generation selection. These designs pro-
vide reasonable error estimates and are fairly robust.
One major advantage of RCB designs is that they can be
used for any number of entries. As the number of surviv-
ing test entries is reduced then more detailed incomplete
block designs such as lattice squares, rectangular lattices
and partially balanced incomplete block designs may
be used.
Lattice squares are amongst the most efficient designs
that can be used for field testing in a plant breeding
programme. A lattice design is similar to a RCB in that
each entry appears once in all replicate blocks. How-
ever, within each replicate block, plots are arranged
into sub-blocks. Analysis of data from lattice designs
allows the actual mean performance of each test entry to
be adjusted due to two dimensional (row and column)
environmental variation. The major problem with lat-
tice squares is that the number of test entries must be
an exact square (i.e. 9, 16, 25, 36, 49, 64, 81 and 100,
etc.). A second restriction is that the most efficient use
of the designs requires high replication. For example a
16 entry design (4
Data analysis and interpretation
It is useful to analyze data as they are collected through-
out the year, so as not have a backlog of analysis which
is needed for decision making at the end of the sea-
son. It is common in plant breeding to have a relatively
quick turn over. For example in winter wheat breed-
ing, evaluation plots are harvested, yields recorded,
samples taken for quality assessment, assessment car-
ried out and decisions made within a few months so
that selection procedures can effectively use all possible
information while still being able to plant selected lines
at the appropriate seasonal time.
If selection is to be successfully applied for any char-
acter it is important that there are indeed significant
differences between test entries. Obviously if an analy-
sis of variance shows no significant difference between
test lines for yield, then there is no genetic variation for
the character, and hence there will be no response to
selection.
It can often be useful to estimate narrow or broad-
sense heritabilities from yield trials. Broad-sense heri-
tabilities can be easily obtained by simply estimating
the genetic component of variance. In cases where test
lines are highly heterozygous then error variances can
be estimated from homozygous control entries in trials.
Narrow-sense heritabilities can be estimated by regres-
sion if a sufficient number of the parental lines are
included in the evaluation trial.
Heritabilities can also be estimated in relation to
response to selection. To achieve this with any accu-
racy it is necessary to retain a certain proportion of
the unselected population and to include these
random
selections along with the deliberate selections in the
following seasons' trial. The practice of retaining a
random sample is highly recommended as it allows
×
+
=
4 lattice square) requires 4
1
5
−
replicates. Larger lattice squares can be used with
n
1
replicates, where
n
2
is the number of entries in the whole
trial.
Rectangular lattice designs allow greater flexibil-
ity in the number of entries and replicates used,
although
each
replicate
must
be
a
rectangle
(i.e.
×
10 plots
5 plots, where sub-blocks would be either
5 to 10 plots). Rectangular lattice designs are not
as efficient as lattice squares in reducing error vari-
ance as sub-block adjustments are made in only one
direction.
One advantage of all lattice designs is that they are
resolvable (i.e. data collected from them can be analyzed
as a RCB design).
Spilt-plot designs are often used in the latter stages
of intermediate selection. The main use is often to eval-
uate a number of lines under differing environmental
conditions at a single location. For example, several
test entries may be assessed under differing nitrogen
levels where genotypes would be main-plots and vary-
ing nitrogen levels are sub-plots. Similarly, spilt-plot
designs can be used for differential chemical treatments
or harvesting dates.
