Mechanistic formulations in estuarine ecosystem
models have largely focused on lower trophic levels and
water quality (e.g., NPZ models), especially the relationships between nutrient loading and light and how these
factors impact phytoplankton growth, survival of submerged grasses, and development of hypoxia/anoxia
(e.g., Kremer and Nixon, 1978; Cerco and Noel, 2004).
In contrast, models focused on higher trophic levels, such
as Ecopath with Ecosim (EwE, see www.ecopath.org),
tend to contain fewer mechanistic functions. Both of these
approaches make attempts at “ecosystem” conditions in
somewhat simplified terms; LTL models represent the
higher trophic levels as a simple closure term, and HTL
models represent the lower trophic levels as a forcing
term. In fisheries management applications, modeling
efforts have typically employed multiple species
approaches such as EwE, MSVPA (multispecies virtual
population analysis) models, or multispecies production
models, while more mechanistically formulated examples
typically take the form of individual-based or bioenergetic
models (Latour et al., 2003; Travers et al., 2007). An intermediate step toward development of E2E models involves
coupling of single-species models of shellfish and fish to
larger, compartmental LTL models (e.g., Cerco and Noel,
2007).
Network analysis provides another means for predicting
the exchange of energy and materials amongst various ecosystem compartments and through entire food webs using
mass-balance constraints (Dame and Christian, 2006). As
a precursor to the now ubiquitous EwE models, network
analyses such as those of Baird and Ulanowicz (1989) and
Ulanowicz and Wulff (1991) have been used to generate
metrics of energy flow that could be used for estuarine comparison. Applying network analyses in the Chesapeake and
Baltic, Ulanowicz and Wulff (1991) were able to quantify
characteristics related to the transfer and flow of energy
through the two ecosystems to make so-called anatomical
comparisons in the structure and composition of ecosystem
components. While these network analyses can be very
effective at revealing ecosystem structure, they are data
intensive and may not provide the type of mechanistic
understanding or forecasting capability frequently desired
for management applications.
Nutrient and energy budget and mass-balance
approaches are another modeling approach that has been
especially useful in estuarine comparisons and as tools
for management (Boynton and Nixon, 2012). These types
of models are based on the principles of conservation of
mass and energy, where the goal is to quantify all input
and output terms (e.g., carbon and/or nitrogen) for
a particular marine system. Considerable insight can be
gained by comparing these budgets for different marine
systems even when there are unquantified terms (see
examples in Kremer et al., 2000). While these massbalance models are not typically dynamic, they provide
a numerical framework for understanding major processes
and inputs to a given ecosystem.
The modeling process
Models are developed through a process that proceeds
from conception through formulation, testing, and application (Figure 5). The type and complexity of the model
used are dependent upon the research question or application. The modeler must first determine the boundary of
their system to be modeled, both in terms of space (i.e.,
the model domain) and what to include (i.e., forcing functions vs. state variables). The modeler must then decide
how much biological and biogeochemical complexity to
include, balancing the need for a realistic model with data
available for formulation and model testing (i.e., calibration and verification). Similarly, the modeler must decide
on the appropriate spatial resolution, from representing
the estuary as a single box, through one- and
two-dimensional arrays of boxes, to a highly resolved
3D mesh. These choices determine the balance between
model precision (i.e., ability to reproduce observations
with minimal error), realism (i.e., degree to which the
model is an accurate representation of the real system),
and generality (i.e., degree to which the model is readily
transferable to new systems) (Levins, 1966). Recent cases
of “participatory modeling” have provided examples of
how the numerical modeler may communicate and interact with non-modeling end users to make these choices
and have provided early successes in how this approach
may improve the applicability of model output to realworld problems (Voinov and Bousquet, 2010).
Once these choices are made, the model formulations
are developed and parameterized. The completed model
must then be tested against (i.e., calibrated to) a set of
observations to maximize the fit between model and data
as much as possible. At this point, modeling becomes an
iterative process, in which one will need to go back in
the process to adjust parameter values within reasonable
ranges to improve the fit. If the fit does not improve, one
may need to go further back to reconsider some of the formulations that were chosen or even further back to
reevaluate the structure and assumptions of the model.
Once calibration is successful, the model is then ideally
verified (also termed “validation” or “confirmation”)
against an independent dataset – perhaps a portion of the
observations held back from calibration or data from
a different system or time period. The goal of verification
is to reproduce the independent dataset without needing to
make further changes to parameter values or model structure. Once one has confidence in the quality of the model,
its behavior can be further analyzed via sensitivity analysis and skill assessment, or the model can be used to
address the original research question via heuristic exploration, hypothesis testing, scenario analysis, and/or management application.
Summary
Ecological and ecosystem models have emerged as powerful research, synthesis, and management tools over the
218
ECOLOGICAL MODELING
models have largely focused on lower trophic levels and
water quality (e.g., NPZ models), especially the relationships between nutrient loading and light and how these
factors impact phytoplankton growth, survival of submerged grasses, and development of hypoxia/anoxia
(e.g., Kremer and Nixon, 1978; Cerco and Noel, 2004).
In contrast, models focused on higher trophic levels, such
as Ecopath with Ecosim (EwE, see www.ecopath.org),
tend to contain fewer mechanistic functions. Both of these
approaches make attempts at “ecosystem” conditions in
somewhat simplified terms; LTL models represent the
higher trophic levels as a simple closure term, and HTL
models represent the lower trophic levels as a forcing
term. In fisheries management applications, modeling
efforts have typically employed multiple species
approaches such as EwE, MSVPA (multispecies virtual
population analysis) models, or multispecies production
models, while more mechanistically formulated examples
typically take the form of individual-based or bioenergetic
models (Latour et al., 2003; Travers et al., 2007). An intermediate step toward development of E2E models involves
coupling of single-species models of shellfish and fish to
larger, compartmental LTL models (e.g., Cerco and Noel,
2007).
Network analysis provides another means for predicting
the exchange of energy and materials amongst various ecosystem compartments and through entire food webs using
mass-balance constraints (Dame and Christian, 2006). As
a precursor to the now ubiquitous EwE models, network
analyses such as those of Baird and Ulanowicz (1989) and
Ulanowicz and Wulff (1991) have been used to generate
metrics of energy flow that could be used for estuarine comparison. Applying network analyses in the Chesapeake and
Baltic, Ulanowicz and Wulff (1991) were able to quantify
characteristics related to the transfer and flow of energy
through the two ecosystems to make so-called anatomical
comparisons in the structure and composition of ecosystem
components. While these network analyses can be very
effective at revealing ecosystem structure, they are data
intensive and may not provide the type of mechanistic
understanding or forecasting capability frequently desired
for management applications.
Nutrient and energy budget and mass-balance
approaches are another modeling approach that has been
especially useful in estuarine comparisons and as tools
for management (Boynton and Nixon, 2012). These types
of models are based on the principles of conservation of
mass and energy, where the goal is to quantify all input
and output terms (e.g., carbon and/or nitrogen) for
a particular marine system. Considerable insight can be
gained by comparing these budgets for different marine
systems even when there are unquantified terms (see
examples in Kremer et al., 2000). While these massbalance models are not typically dynamic, they provide
a numerical framework for understanding major processes
and inputs to a given ecosystem.
The modeling process
Models are developed through a process that proceeds
from conception through formulation, testing, and application (Figure 5). The type and complexity of the model
used are dependent upon the research question or application. The modeler must first determine the boundary of
their system to be modeled, both in terms of space (i.e.,
the model domain) and what to include (i.e., forcing functions vs. state variables). The modeler must then decide
how much biological and biogeochemical complexity to
include, balancing the need for a realistic model with data
available for formulation and model testing (i.e., calibration and verification). Similarly, the modeler must decide
on the appropriate spatial resolution, from representing
the estuary as a single box, through one- and
two-dimensional arrays of boxes, to a highly resolved
3D mesh. These choices determine the balance between
model precision (i.e., ability to reproduce observations
with minimal error), realism (i.e., degree to which the
model is an accurate representation of the real system),
and generality (i.e., degree to which the model is readily
transferable to new systems) (Levins, 1966). Recent cases
of “participatory modeling” have provided examples of
how the numerical modeler may communicate and interact with non-modeling end users to make these choices
and have provided early successes in how this approach
may improve the applicability of model output to realworld problems (Voinov and Bousquet, 2010).
Once these choices are made, the model formulations
are developed and parameterized. The completed model
must then be tested against (i.e., calibrated to) a set of
observations to maximize the fit between model and data
as much as possible. At this point, modeling becomes an
iterative process, in which one will need to go back in
the process to adjust parameter values within reasonable
ranges to improve the fit. If the fit does not improve, one
may need to go further back to reconsider some of the formulations that were chosen or even further back to
reevaluate the structure and assumptions of the model.
Once calibration is successful, the model is then ideally
verified (also termed “validation” or “confirmation”)
against an independent dataset – perhaps a portion of the
observations held back from calibration or data from
a different system or time period. The goal of verification
is to reproduce the independent dataset without needing to
make further changes to parameter values or model structure. Once one has confidence in the quality of the model,
its behavior can be further analyzed via sensitivity analysis and skill assessment, or the model can be used to
address the original research question via heuristic exploration, hypothesis testing, scenario analysis, and/or management application.
Summary
Ecological and ecosystem models have emerged as powerful research, synthesis, and management tools over the
218
ECOLOGICAL MODELING