278
simulation analysis is used to assess the current processes occurring at these scales, and the potential
sensitivity of biogeochemical pools and fluxes to
these types of perturbations.
In this paper, I review the most common methodologies for each of these three types of analyses
in landscape and regional biogeochemistry, identify
their strengths and weaknesses, and briefly note
some of the new techniques available. My emphasis
is on reviewing the approaches and strategies for
designs in landscape- to regional-scale biogeochemistry, rather than describing the detailed methods. For each of the three major categories (pattern
analysis, spatially explicit analysis, and extrapolating to landscape and regional scales), I will describe
the approaches for field analyses and for modeling
analyses.
Pattern Analysis:
Design for Field Studies
To a large extent, empirical pattern analysis has
provided much of our current knowledge of the
controls over biogeochemical pools and fluxes. The
early work of Dockuchaiev (1883), Shaw (1930),
and Jenny (1941), made across large-scale regional
gradients, had a substantial influence on the foundations of soil science and ecosystem ecology. Similarly, Lieth (1978) made great contributions with
respect to our understanding of the controls over
primary productivity, Meentemeyer (1984) over decomposition, and Matson and Vitousek (1987) over
trace gas flux. The relationships generated from
pattern analysis not only contribute to our knowledge of the biogeochemical processes and their
controls, but also provide important predictive
equations. For instance, the relationship generated
by Sala et al. (1988) concerning primary productivity in grasslands is incorporated into the major biogeochemical simulation model for grasslands, Century (Parton et al. 1987). Many of these studies
utilized large databases from multiple sources,
which creates difficulties in that methods are often
not consistent across studies. New empirical studies
that address landscape- to regional-scale variation
in biogeochemical pools that use consistent methods across such large gradients are often more useful. Here, I discuss several issues relevant to the
design of such studies.
Ingrid C. Burke
Stratified Sampling and Discrete Units
Most landscape biogeochemistry studies have utilized stratified random sampling, which requires an
a priori discrete classification of landscapes or
regions. Jeffers (1988) and Groffman (1991) suggest that this type of stratified sampling approach
is the most appropriate and useful design for assessing the controls over biogeochemical processes, and for scaling up studies from the field to
landscapes, regions, or the globe. In this type of
design, investigators utilize prior information about
the system, and stratify in such a way as to attempt
to capture the major variation in the controls over
biogeochemical processes in field units. Clearly, the
decisions made at this point influence field results
and propagate through to the scaling of field measurements to larger areas (Groffman 1991; O'Neill
1986).
Four basic types of a priori classification are usually made, and the choice both reflects the expertise
of the investigator and results in some bias in the
results. First, investigators may group landscape or
regional units by vegetation type (e.g., Zak et al.
1986; Burke et al. 1989; Giblin et al. 1991; Valentine and Binkley 1992; Kielland 1995), and replicate by sampling several different stands representing each vegetation type. For example, Zak et
al. (1986) stratified a forest landscape in Michigan
by forest community type. In this landscape, previous research had defined four community types,
sugar maple-basswood, sugar maple-red oak, and
red oak-black oak, that were the product of site
conditions, especially fertility. Second, groupings
may be made by soil type (Lajtha and Schlesinger
1988; Groffman and Tiedje 1989; Groffman et al.
1992; Groffman and Hanson 1997), since soils integrate many of the processes important to biogeochemical cycling (Jenny 1980; Amundson and
Jenny 1997). For example, Lajtha and Schlesinger
(1988) stratified a desert landscape based upon its
pedogenic age to evaluate large-scale controls over
phosphorus availability. A third type of classification is particularly common in landscapes that are
strongly driven by fluvial processes, in which twodimensional representations of landscapes called
"catenas" (AandahI1948; Jenny 1980) are divided
into units such as summit, midslope (or backslope),
and toeslope (or footslope) (Schimel et al. 1985;
Schimel1986; Aguilar et al. 1988; Groffman et al.
simulation analysis is used to assess the current processes occurring at these scales, and the potential
sensitivity of biogeochemical pools and fluxes to
these types of perturbations.
In this paper, I review the most common methodologies for each of these three types of analyses
in landscape and regional biogeochemistry, identify
their strengths and weaknesses, and briefly note
some of the new techniques available. My emphasis
is on reviewing the approaches and strategies for
designs in landscape- to regional-scale biogeochemistry, rather than describing the detailed methods. For each of the three major categories (pattern
analysis, spatially explicit analysis, and extrapolating to landscape and regional scales), I will describe
the approaches for field analyses and for modeling
analyses.
Pattern Analysis:
Design for Field Studies
To a large extent, empirical pattern analysis has
provided much of our current knowledge of the
controls over biogeochemical pools and fluxes. The
early work of Dockuchaiev (1883), Shaw (1930),
and Jenny (1941), made across large-scale regional
gradients, had a substantial influence on the foundations of soil science and ecosystem ecology. Similarly, Lieth (1978) made great contributions with
respect to our understanding of the controls over
primary productivity, Meentemeyer (1984) over decomposition, and Matson and Vitousek (1987) over
trace gas flux. The relationships generated from
pattern analysis not only contribute to our knowledge of the biogeochemical processes and their
controls, but also provide important predictive
equations. For instance, the relationship generated
by Sala et al. (1988) concerning primary productivity in grasslands is incorporated into the major biogeochemical simulation model for grasslands, Century (Parton et al. 1987). Many of these studies
utilized large databases from multiple sources,
which creates difficulties in that methods are often
not consistent across studies. New empirical studies
that address landscape- to regional-scale variation
in biogeochemical pools that use consistent methods across such large gradients are often more useful. Here, I discuss several issues relevant to the
design of such studies.
Ingrid C. Burke
Stratified Sampling and Discrete Units
Most landscape biogeochemistry studies have utilized stratified random sampling, which requires an
a priori discrete classification of landscapes or
regions. Jeffers (1988) and Groffman (1991) suggest that this type of stratified sampling approach
is the most appropriate and useful design for assessing the controls over biogeochemical processes, and for scaling up studies from the field to
landscapes, regions, or the globe. In this type of
design, investigators utilize prior information about
the system, and stratify in such a way as to attempt
to capture the major variation in the controls over
biogeochemical processes in field units. Clearly, the
decisions made at this point influence field results
and propagate through to the scaling of field measurements to larger areas (Groffman 1991; O'Neill
1986).
Four basic types of a priori classification are usually made, and the choice both reflects the expertise
of the investigator and results in some bias in the
results. First, investigators may group landscape or
regional units by vegetation type (e.g., Zak et al.
1986; Burke et al. 1989; Giblin et al. 1991; Valentine and Binkley 1992; Kielland 1995), and replicate by sampling several different stands representing each vegetation type. For example, Zak et
al. (1986) stratified a forest landscape in Michigan
by forest community type. In this landscape, previous research had defined four community types,
sugar maple-basswood, sugar maple-red oak, and
red oak-black oak, that were the product of site
conditions, especially fertility. Second, groupings
may be made by soil type (Lajtha and Schlesinger
1988; Groffman and Tiedje 1989; Groffman et al.
1992; Groffman and Hanson 1997), since soils integrate many of the processes important to biogeochemical cycling (Jenny 1980; Amundson and
Jenny 1997). For example, Lajtha and Schlesinger
(1988) stratified a desert landscape based upon its
pedogenic age to evaluate large-scale controls over
phosphorus availability. A third type of classification is particularly common in landscapes that are
strongly driven by fluvial processes, in which twodimensional representations of landscapes called
"catenas" (AandahI1948; Jenny 1980) are divided
into units such as summit, midslope (or backslope),
and toeslope (or footslope) (Schimel et al. 1985;
Schimel1986; Aguilar et al. 1988; Groffman et al.