18. Landscape and Regional Biogeochemistry: Approaches
283
error or confidence intervals. Until ecologists have
agreed upon appropriate fonnulae for estimating
confidence intervals of such estimates, the variance
for each of the elements of the extrapolation should
be presented clearly.
Modeling
The major drawbacks to using the above extrapolation method are the limitations associated with
classification errors. Multiple controls over biogeochemical pools and fluxes co-vary, as mentioned
above, and the classification strategy does not allow
those controls to be separated (Matson et al. 1989).
An alternate approach is to utilize models that allow
for continuous variation in the control factors to
occur.
Either simple statistical models or simulation
models may be utilized. In either case, there are
three steps to the analysis (Costanza et al. 1990;
Burke et al. 1991; Ollinger et al. 1993; Coleman et
al. 1994). The first step is to obtain a georeferenced
database of the major control factors over the biogeochemical pool or flux of interest, for the region
or landscape of interest. Key decisions must be
made regarding (1) which control factors will be
incorporated and (2) what spatial and temporal resolution will be used. The decisions are interactive,
and both are influenced as much by logistical constraints as by scientific construct of the system. For
instance, georeferenced data on soils are available
at specific resolutions, and one is generally not able
to find scales more resolved. Land-use, vegetation,
and net primary productivity data are often generated from satellite imagery (see Chapter 3). In the
second step, the data are integrated into a geographic infonnation system (GIS) (this may be
industry-standard software or generated by the scientist), and a single scale of resolution for applying
the model is selected when the data are "overlaid,"
or spatially registered with one another. In many
cases, the choice will be to aggregate fine-scaled
data to a coarser scale. Considerable attention must
be paid to the algorithms used in scaling up data.
A well-recognized problem is that if linear algorithms are used to scale infonnation up to larger
spatial units, significant error will result if those
data are introduced into a model with nonlinearities
(Cale et al. 1983; Burke et al. 1990; Rastetter et al.
1992; Pierce and Running 1995).
Finally, the model is applied to the layers of input
data. The model may be very simple, such as a
regression model, or a complex simulation model,
as long as the control or input variables of choice
are present in the GIS, and have the appropriate
temporal resolution. Output variables of interest are
stored in the GIS for mapping or analysis. At present, considerable effort is usually required to link
simulation models with GIS, although several automated linkages are being developed (Coleman et
al. 1994).
The process of extrapolating to landscape or regional scales represents the integration of many different types of scientific activities. For instance,
modeling changes in regional C storage involves
first, the integration of large number of field studies
into a coherent body of knowledge, represented by
the simulation model (Fig. 18.1). Simulation models must be rigorously tested in the field using separate data sets from those which generated them
(Lauenroth et al. 1998). In some cases, pattern analysis studies have provided important infonnation
that is included in simulation models (Parton et al.
1987), or geographic data have provided important
infonnation for validating simulation models
(Burke et al. 1989). Finally, large-scale simulations
can be tested against large-scale measurements,
such as eddy accumulation, as a means of increasing our confidence in such extrapolations.
Summary
During the past several decades, the number of
tools and approaches for ecosystem ecologists has
dramatically increased the opportunities for study
of processes at landscape to regional scales. This
activity is crucial for the discipline to apply our
understanding to the scales most relevant to
atmosphere-biosphere interactions, human impacts
on the environment, and global change. Coupled
with the difficulties of studying ecosystem processes at large scales is a current strong limitation
in our ability to express error and uncertainty of our
measurements and models. This combination of the
high degree of societal need for our results, the relatively young state of our discipline at these scales,
and our limitations in expressing uncertainty pro-
283
error or confidence intervals. Until ecologists have
agreed upon appropriate fonnulae for estimating
confidence intervals of such estimates, the variance
for each of the elements of the extrapolation should
be presented clearly.
Modeling
The major drawbacks to using the above extrapolation method are the limitations associated with
classification errors. Multiple controls over biogeochemical pools and fluxes co-vary, as mentioned
above, and the classification strategy does not allow
those controls to be separated (Matson et al. 1989).
An alternate approach is to utilize models that allow
for continuous variation in the control factors to
occur.
Either simple statistical models or simulation
models may be utilized. In either case, there are
three steps to the analysis (Costanza et al. 1990;
Burke et al. 1991; Ollinger et al. 1993; Coleman et
al. 1994). The first step is to obtain a georeferenced
database of the major control factors over the biogeochemical pool or flux of interest, for the region
or landscape of interest. Key decisions must be
made regarding (1) which control factors will be
incorporated and (2) what spatial and temporal resolution will be used. The decisions are interactive,
and both are influenced as much by logistical constraints as by scientific construct of the system. For
instance, georeferenced data on soils are available
at specific resolutions, and one is generally not able
to find scales more resolved. Land-use, vegetation,
and net primary productivity data are often generated from satellite imagery (see Chapter 3). In the
second step, the data are integrated into a geographic infonnation system (GIS) (this may be
industry-standard software or generated by the scientist), and a single scale of resolution for applying
the model is selected when the data are "overlaid,"
or spatially registered with one another. In many
cases, the choice will be to aggregate fine-scaled
data to a coarser scale. Considerable attention must
be paid to the algorithms used in scaling up data.
A well-recognized problem is that if linear algorithms are used to scale infonnation up to larger
spatial units, significant error will result if those
data are introduced into a model with nonlinearities
(Cale et al. 1983; Burke et al. 1990; Rastetter et al.
1992; Pierce and Running 1995).
Finally, the model is applied to the layers of input
data. The model may be very simple, such as a
regression model, or a complex simulation model,
as long as the control or input variables of choice
are present in the GIS, and have the appropriate
temporal resolution. Output variables of interest are
stored in the GIS for mapping or analysis. At present, considerable effort is usually required to link
simulation models with GIS, although several automated linkages are being developed (Coleman et
al. 1994).
The process of extrapolating to landscape or regional scales represents the integration of many different types of scientific activities. For instance,
modeling changes in regional C storage involves
first, the integration of large number of field studies
into a coherent body of knowledge, represented by
the simulation model (Fig. 18.1). Simulation models must be rigorously tested in the field using separate data sets from those which generated them
(Lauenroth et al. 1998). In some cases, pattern analysis studies have provided important infonnation
that is included in simulation models (Parton et al.
1987), or geographic data have provided important
infonnation for validating simulation models
(Burke et al. 1989). Finally, large-scale simulations
can be tested against large-scale measurements,
such as eddy accumulation, as a means of increasing our confidence in such extrapolations.
Summary
During the past several decades, the number of
tools and approaches for ecosystem ecologists has
dramatically increased the opportunities for study
of processes at landscape to regional scales. This
activity is crucial for the discipline to apply our
understanding to the scales most relevant to
atmosphere-biosphere interactions, human impacts
on the environment, and global change. Coupled
with the difficulties of studying ecosystem processes at large scales is a current strong limitation
in our ability to express error and uncertainty of our
measurements and models. This combination of the
high degree of societal need for our results, the relatively young state of our discipline at these scales,
and our limitations in expressing uncertainty pro-