294
Valerie T. Eviner, F. Stuart Chapin III, and Charles E. Vaughn
Experimental Setup
The appropriate design for nutrient addition experiments is similar to that of most field experiments.
For example, many experiments use a randomized
block design, in which replicate blocks are set out
in "representative patches" of the ecosystem, with
buffer strips between treatment plots. Each treatment is then randomly assigned within blocks. The
size of the buffer strips depends on the mobility of
the nutrient and the lateral root extension of the
dominant plants. The most difficult task in implementing this design is in deciding what is a representative patch, and what is the domain over which
the results can be generalized. The results of the
experiment can only be rigorously generalized to
the experimental area. For example, if the experiment is set out on a hillside, and patches of bare
ground are avoided, the results can be generalized
only to vegetated components of the ecosystem on
that hillside. This would be an appropriate design
for looking at the effect of nutrients on the competitive interactions among plants. In a study of this
sort, inclusion of bare patches would only increase
the variance without adding insight about competitive processes (Hobbie and Chapin 1998). If the
experiment is intended to be generalized to a large
region, either the replicate blocks must be dispersed
over the entire region or the experiment must
be repeated over the entire region. For example,
to generalize about whether nutrient limitation
changed through postglacial succession at Glacier
Bay, Alaska, plots were distributed throughout a
100-krn length of the bay, with 10 replicate blocks
per successional stage but only one replicate block
in each location (Chapin et al. 1994). In another
series of studies conducted by several research
groups, similar nutrient addition treatments were
applied to tussock tundra throughout its range in
Alaska, Britain, and Scandinavia (Tamm 1954;
Goodman and Perkins 1968; Shaver and Chapin
1980; Shaver and Chapin 1995), demonstrating that
growth of the dominant species (Eriophorum vaginatum) in this vegetation type was consistently
nutrient-limited, but that the identity of the limiting
nutrient varied regionally.
The extent of replication necessary in a nutrientaddition study depends on the magnitude and variability of nutrient limitation and the size of the nutrient addition. For experiments in which expected
effects are large, a sample size of 4 to 6 may be
adequate. However, if effects may be more subtle,
as in isotope addition experiments or experiments
simulating atmospheric deposition, or if the experiment deliberately incorporates substantial landscape heterogeneity (to increase the generalizability
of results), larger sample sizes (e.g., 8 to 10) are
essential. The actual experimental design chosen is
usually a compromise between what is desirable
and what is feasible.
Experimental Design
There are many factors to consider in designing
a nutrient addition experiment, and the choices
are specific to the question and system. There are
many controls on the effectiveness of a given nutrient addition, including climate, soils, the potential of the vegetation to respond to nutrients, land
use, date of germination relative to date of fertilization, limitation of other nutrients, and decreases in biomass due to diseases and herbivores
(FAO 1987). These factors could confound the
interpretation of nutrient-addition experiments by
affecting the availability of nutrients added and
the measurements of biomass response. These
confounding factors can be minimized by carefully deciding the form of nutrient to add, the
amount to add, the frequency and timing of additions, and when to measure the response
variables.
The question to be addressed should guide which
of these complicating issues to focus on. For example, when determining which nutrients are limiting at a given site, it is critical to add the different
nutrients in forms with similar availabilities. Most
N additions (ammonium [NHt], nitrate [NOll,
urea) are immediately soluble and available for
plant uptake. It would not be valid to compare the
biomass response to these soluble N fertilizers with
response to rock phosphate or elemental S, which
have much slower nutrient supply rates. In addition,
factors such as adsorption, trace gas loss, and leaching can prevent any of the added nutrients from
being available to plants. These factors should be
accounted for, and addition rates of different nutrients should be adjusted to ensure similar supply
rates. This is especially crucial when comparing nutrient limitation across sites. Site differences in hydrology, soil adsorption, pH, and vegetation can
Valerie T. Eviner, F. Stuart Chapin III, and Charles E. Vaughn
Experimental Setup
The appropriate design for nutrient addition experiments is similar to that of most field experiments.
For example, many experiments use a randomized
block design, in which replicate blocks are set out
in "representative patches" of the ecosystem, with
buffer strips between treatment plots. Each treatment is then randomly assigned within blocks. The
size of the buffer strips depends on the mobility of
the nutrient and the lateral root extension of the
dominant plants. The most difficult task in implementing this design is in deciding what is a representative patch, and what is the domain over which
the results can be generalized. The results of the
experiment can only be rigorously generalized to
the experimental area. For example, if the experiment is set out on a hillside, and patches of bare
ground are avoided, the results can be generalized
only to vegetated components of the ecosystem on
that hillside. This would be an appropriate design
for looking at the effect of nutrients on the competitive interactions among plants. In a study of this
sort, inclusion of bare patches would only increase
the variance without adding insight about competitive processes (Hobbie and Chapin 1998). If the
experiment is intended to be generalized to a large
region, either the replicate blocks must be dispersed
over the entire region or the experiment must
be repeated over the entire region. For example,
to generalize about whether nutrient limitation
changed through postglacial succession at Glacier
Bay, Alaska, plots were distributed throughout a
100-krn length of the bay, with 10 replicate blocks
per successional stage but only one replicate block
in each location (Chapin et al. 1994). In another
series of studies conducted by several research
groups, similar nutrient addition treatments were
applied to tussock tundra throughout its range in
Alaska, Britain, and Scandinavia (Tamm 1954;
Goodman and Perkins 1968; Shaver and Chapin
1980; Shaver and Chapin 1995), demonstrating that
growth of the dominant species (Eriophorum vaginatum) in this vegetation type was consistently
nutrient-limited, but that the identity of the limiting
nutrient varied regionally.
The extent of replication necessary in a nutrientaddition study depends on the magnitude and variability of nutrient limitation and the size of the nutrient addition. For experiments in which expected
effects are large, a sample size of 4 to 6 may be
adequate. However, if effects may be more subtle,
as in isotope addition experiments or experiments
simulating atmospheric deposition, or if the experiment deliberately incorporates substantial landscape heterogeneity (to increase the generalizability
of results), larger sample sizes (e.g., 8 to 10) are
essential. The actual experimental design chosen is
usually a compromise between what is desirable
and what is feasible.
Experimental Design
There are many factors to consider in designing
a nutrient addition experiment, and the choices
are specific to the question and system. There are
many controls on the effectiveness of a given nutrient addition, including climate, soils, the potential of the vegetation to respond to nutrients, land
use, date of germination relative to date of fertilization, limitation of other nutrients, and decreases in biomass due to diseases and herbivores
(FAO 1987). These factors could confound the
interpretation of nutrient-addition experiments by
affecting the availability of nutrients added and
the measurements of biomass response. These
confounding factors can be minimized by carefully deciding the form of nutrient to add, the
amount to add, the frequency and timing of additions, and when to measure the response
variables.
The question to be addressed should guide which
of these complicating issues to focus on. For example, when determining which nutrients are limiting at a given site, it is critical to add the different
nutrients in forms with similar availabilities. Most
N additions (ammonium [NHt], nitrate [NOll,
urea) are immediately soluble and available for
plant uptake. It would not be valid to compare the
biomass response to these soluble N fertilizers with
response to rock phosphate or elemental S, which
have much slower nutrient supply rates. In addition,
factors such as adsorption, trace gas loss, and leaching can prevent any of the added nutrients from
being available to plants. These factors should be
accounted for, and addition rates of different nutrients should be adjusted to ensure similar supply
rates. This is especially crucial when comparing nutrient limitation across sites. Site differences in hydrology, soil adsorption, pH, and vegetation can