18. Landscape and Regional Biogeochemistry: Approaches
279
1993). More complex versions of landform classification (Swanson et al. 1988; Coppinger et al.
1991) may be used where elevation and aspect are
as important as slope position. Finally, land-use
management classifications are often used (e.g.,
Mosier et al. 1991; Reiners et al. 1994), sometimes
in combination with one of the other grouping
systems.
Once the classification system is chosen, investigators generally select replicate units for study,
and conduct a standard field sampling program
within the replicate units. Data are appropriately
analyzed using analysis of variance type statistics.
Special care needs to be taken with interpretation,
since significance of the "class" or unit only implies
that the investigator's a priori classification represents an axis of strong variability in biogeochemical processes. Since vegetation, soil, landform, and
land use may all co-vary, it is not possible using
this type of sampling regime to isolate the proximal
controls over biogeochemical processes (Groffman
et al. 1988). Two-way analysis of variance, utilizing
overlapping classification schemes and sampling,
can assist with this problem.
Use of an a priori discrete classification scheme
in sampling landscape or regional scale patterns in
biogeochemical pools and processes is the most
common strategy, and it has substantial advantages
(Matson et al. 1989). Ecologists are generally quite
familiar with the appropriate variations on experimental design and statistical analysis, and the peerreview process is relatively simple as a result. Extrapolating from the sampled area to landscapes
and/or regions is relatively simple (see section below), as long as the units chosen reflect most of the
variation at the new scale.
Sampling Continuous Variation
An alternative approach to using discrete classes in
characterizing landscape- to regional-scale patterns
in biogeochemistry is to sample continuous variation in control factors and thus, response variables.
Many of the classic papers that assess large-scale
controls over biogeochemical processes are metadata analyses, utilizing data from a large number of
studies that cross relatively large gradients of temperature, precipitation, vegetation, and soils (e.g.,
Lieth 1978; Meentemeyer 1984; Meentemeyer et
al. 1982, 1985; Matson and Vitousek 1987). The
continuous variation in multiple control factors allows interactions among variables to be statistically
evaluated across a range of values. This type of
design has some strong advantages over the discrete approaches outlined above, in that regression
equations can provide predictive models for use in
other systems (e.g., Pamelo et al. 1998).
Two general types of field sampling designs are
commonly used to capitalize on continuous variation in the controls over biogeochemical processes.
In the first, one or more transects are arrayed along
the major gradients, and sampling is conducted at
regular or random intervals (e.g., Meentemeyer and
Berg 1986; Yonker et al. 1988; Vitousek et al. 1994;
Schimel et al. 1991; Knapp et al. 1993 ; Naiman et
al. 1994; Pouyat et al. 1995; Johanson et al. 1995).
Data may then be analyzed using regression or correlation analysis, analyzing the relationship between the independent variable(s) and the dependent biogeochemical factor of interest.
The second type of design for sampling continuous variation is to utilize a grid sampling scheme
that permits continuous variation in all independent
and dependent variables (Robertson 1987). This
type of design is particularly powerful for testing
for the spatial autocorrelation present in biogeochemical factors; the analysis is summarized in
Robertson (1987). This design and analysis protocol permit the investigator to test for important spatial scales of variation and spatial dependence that
may not be evident a priori (e.g., Robertson 1982,
1987; Robertson et al. 1988, 1997; Halvorson et al.
1994), and can not be assessed using the discrete
approach outlined above. Alternatively, the continuous grid design may be analyzed using nonspatial
multivariate techniques to evaluate the multidimensional relationships among the control factors and
biogeochemical pools and processes. Multivariate
techniques such as principal components analysis
or factor analysis may be used to identify the shared
variance in the data set, for interpretation of the
major controls over distribution and flux (Burke
1989; Naiman et al. 1994). Simple multiple regression or correlation techniques may also be used
(Groffman and Turner 1995) to characterize relationships, as in the transect design.
The continuous approaches offer several distinct
advantages over the discrete approaches. First, the
design is less closely the result of investigator bias
(not requiring stratification), and more combina-
279
1993). More complex versions of landform classification (Swanson et al. 1988; Coppinger et al.
1991) may be used where elevation and aspect are
as important as slope position. Finally, land-use
management classifications are often used (e.g.,
Mosier et al. 1991; Reiners et al. 1994), sometimes
in combination with one of the other grouping
systems.
Once the classification system is chosen, investigators generally select replicate units for study,
and conduct a standard field sampling program
within the replicate units. Data are appropriately
analyzed using analysis of variance type statistics.
Special care needs to be taken with interpretation,
since significance of the "class" or unit only implies
that the investigator's a priori classification represents an axis of strong variability in biogeochemical processes. Since vegetation, soil, landform, and
land use may all co-vary, it is not possible using
this type of sampling regime to isolate the proximal
controls over biogeochemical processes (Groffman
et al. 1988). Two-way analysis of variance, utilizing
overlapping classification schemes and sampling,
can assist with this problem.
Use of an a priori discrete classification scheme
in sampling landscape or regional scale patterns in
biogeochemical pools and processes is the most
common strategy, and it has substantial advantages
(Matson et al. 1989). Ecologists are generally quite
familiar with the appropriate variations on experimental design and statistical analysis, and the peerreview process is relatively simple as a result. Extrapolating from the sampled area to landscapes
and/or regions is relatively simple (see section below), as long as the units chosen reflect most of the
variation at the new scale.
Sampling Continuous Variation
An alternative approach to using discrete classes in
characterizing landscape- to regional-scale patterns
in biogeochemistry is to sample continuous variation in control factors and thus, response variables.
Many of the classic papers that assess large-scale
controls over biogeochemical processes are metadata analyses, utilizing data from a large number of
studies that cross relatively large gradients of temperature, precipitation, vegetation, and soils (e.g.,
Lieth 1978; Meentemeyer 1984; Meentemeyer et
al. 1982, 1985; Matson and Vitousek 1987). The
continuous variation in multiple control factors allows interactions among variables to be statistically
evaluated across a range of values. This type of
design has some strong advantages over the discrete approaches outlined above, in that regression
equations can provide predictive models for use in
other systems (e.g., Pamelo et al. 1998).
Two general types of field sampling designs are
commonly used to capitalize on continuous variation in the controls over biogeochemical processes.
In the first, one or more transects are arrayed along
the major gradients, and sampling is conducted at
regular or random intervals (e.g., Meentemeyer and
Berg 1986; Yonker et al. 1988; Vitousek et al. 1994;
Schimel et al. 1991; Knapp et al. 1993 ; Naiman et
al. 1994; Pouyat et al. 1995; Johanson et al. 1995).
Data may then be analyzed using regression or correlation analysis, analyzing the relationship between the independent variable(s) and the dependent biogeochemical factor of interest.
The second type of design for sampling continuous variation is to utilize a grid sampling scheme
that permits continuous variation in all independent
and dependent variables (Robertson 1987). This
type of design is particularly powerful for testing
for the spatial autocorrelation present in biogeochemical factors; the analysis is summarized in
Robertson (1987). This design and analysis protocol permit the investigator to test for important spatial scales of variation and spatial dependence that
may not be evident a priori (e.g., Robertson 1982,
1987; Robertson et al. 1988, 1997; Halvorson et al.
1994), and can not be assessed using the discrete
approach outlined above. Alternatively, the continuous grid design may be analyzed using nonspatial
multivariate techniques to evaluate the multidimensional relationships among the control factors and
biogeochemical pools and processes. Multivariate
techniques such as principal components analysis
or factor analysis may be used to identify the shared
variance in the data set, for interpretation of the
major controls over distribution and flux (Burke
1989; Naiman et al. 1994). Simple multiple regression or correlation techniques may also be used
(Groffman and Turner 1995) to characterize relationships, as in the transect design.
The continuous approaches offer several distinct
advantages over the discrete approaches. First, the
design is less closely the result of investigator bias
(not requiring stratification), and more combina-