Fuzzy network models (FNMs) for ecosystem evaluation are not a
substitute for statistical and process models. Rather, FNMs are most
valuable when used as logic frameworks for integrating the outputs from
other models. Consider a hypothetical ecosystem evaluation in which 100
statistical models were developed and applied to various dimensions of the
analysis and another 20 simulations of other system components were run.
A logic framework for integrating all these results might be useful. Because
FNMs are formal specifications for interpreting information, they are
cognitive maps of the problem specification (Stillings et al. 1987). They help
identify questions to be answered, the relevant intermediate states and
processes, the information required, and how the results are related to
each other. It is important to note that these logic networks are not just
specifications, but are themselves models that can be fed data and produce
interpretable output. Furthermore, in systems like NetWeaver, the specification provides an intuitive, graphical explanation for the derivation of
results so the model is not a black box.
8.2.1.1.2 Bayesian Belief Networks
Another class of semantic models are Bayesian belief networks (BBNs)
(Ellison 1996). Bayesian belief networks are based on probability theory,
whereas FNMs are based on set theory. The practical implication of this
difference is that BBNs are best suited to applications where the problem
is relatively narrow and well defined and most conditional probabilities are
known, while FNMs are best suited to applications where the problem is
broad and abstract and a significant proportion of the conditional probabilities are unknown.
8.2.1.2 Data Visualization
Visualization of the relationships and interactions among variables can aid
model formulation and design. When the relationships among variables
are clearly understood, model design and behavior will be enhanced,
and more realistic estimations and predictions will result. Most current statistical packages contain sophisticated graphics packages to allow twodimensional (2-D) projection of a three-dimensional (3-D) data space. True
3-D viewing is possible with specialized projection systems and eyewear
(polarized lenses, alternating liquid-crystal-display lenses, or virtual reality
goggles). It is possible to visualize the interactions of five variables in
a 3-D representation with length, width, height, color, and animation. For
example, consider a representation of tree growth across a region with latitude being length, longitude being width, average monthly temperature
being height, monthly growth rate being color, and time lapse as the animation. A good source for information on this topic is the Digital Visualization Analysis Laboratory of NASA (http://dval-www.larc.nasa.gov).
8. Evolving Approaches and Technologies
137
substitute for statistical and process models. Rather, FNMs are most
valuable when used as logic frameworks for integrating the outputs from
other models. Consider a hypothetical ecosystem evaluation in which 100
statistical models were developed and applied to various dimensions of the
analysis and another 20 simulations of other system components were run.
A logic framework for integrating all these results might be useful. Because
FNMs are formal specifications for interpreting information, they are
cognitive maps of the problem specification (Stillings et al. 1987). They help
identify questions to be answered, the relevant intermediate states and
processes, the information required, and how the results are related to
each other. It is important to note that these logic networks are not just
specifications, but are themselves models that can be fed data and produce
interpretable output. Furthermore, in systems like NetWeaver, the specification provides an intuitive, graphical explanation for the derivation of
results so the model is not a black box.
8.2.1.1.2 Bayesian Belief Networks
Another class of semantic models are Bayesian belief networks (BBNs)
(Ellison 1996). Bayesian belief networks are based on probability theory,
whereas FNMs are based on set theory. The practical implication of this
difference is that BBNs are best suited to applications where the problem
is relatively narrow and well defined and most conditional probabilities are
known, while FNMs are best suited to applications where the problem is
broad and abstract and a significant proportion of the conditional probabilities are unknown.
8.2.1.2 Data Visualization
Visualization of the relationships and interactions among variables can aid
model formulation and design. When the relationships among variables
are clearly understood, model design and behavior will be enhanced,
and more realistic estimations and predictions will result. Most current statistical packages contain sophisticated graphics packages to allow twodimensional (2-D) projection of a three-dimensional (3-D) data space. True
3-D viewing is possible with specialized projection systems and eyewear
(polarized lenses, alternating liquid-crystal-display lenses, or virtual reality
goggles). It is possible to visualize the interactions of five variables in
a 3-D representation with length, width, height, color, and animation. For
example, consider a representation of tree growth across a region with latitude being length, longitude being width, average monthly temperature
being height, monthly growth rate being color, and time lapse as the animation. A good source for information on this topic is the Digital Visualization Analysis Laboratory of NASA (http://dval-www.larc.nasa.gov).
8. Evolving Approaches and Technologies
137