40
D. Morrall
The probabilities of mutation and crossover are selected by the user. The
children replace the least fit individuals in the population. The process continues
either until a specified number of generations have been completed or until an
individual in the population meets the success criteria. Because there is a random
component in the GA, no two runs are the same. The trajectory of population
fitness over time will vary between runs (Figure 3.3).
3.4
Applications of Genetic Aigorithms to Ecological
Modelling
In the field of ecology as in other fields, genetic algorithms have been used for
parameter optimization, equation discovery, and pattern searching. Reynolds and
Ford (1999) developed a modelling approach (Pareto Optimal Model Assessment
Cycle) that allows for simultaneous evaluation of multiple output criteria for a
model with a given parameter set. A genetic algorithm is used to generate optimal
parameter sets (Pareto Optimal Sets) and evaluate fitness in terms of their ability
to predict the model output. Reynolds used this approach to parameterize and
evaluate the ecological theory, model structure, and assessment data of WHORL,
a canopy competition model. Through this dynamic process they were able to
reveal deficiencies in the model structure and criteria to make improvements.
Ludvigan et al (1997) used a GA to search for optimal bacterial phospholipid fatty
acid (PLFA) combinations to biogeochemical parameters. Random combinations
of PLFAs were chosen and evolved using a GA with a partial least squares fitness
function. These combinations of engineering and statistical techniques with
evolutionary algorithms provide a robust approach to data evaluation.
Similar in approach to parameter optimization is the use of GAs to look for data
patterns such as subsurface zones of bacterial activity (Mahinthakumar al. 1999)
and fish distributions (D' Angelo et al. 1995). These studies attempt to find the
optimum combination of variables to predict the distributions of organisms in
relation to their habitat. With the surge in use of geographic information systems
(GIS) to present environmental data on a spatial template, these types of analysis
are common and are expected to become more prevalent. Jeffers (1999) and
StockweIl (1999) provide examples of some species distribution models that use
abundance or presence/absence data to predict distributions of organisms as a
function of spatial habitat characteristics. Fielding (1999) suggests that these
types of exploratory analysis are some of the most promising and least
controversial uses of machine leaming applications.
From a practical standpoint, GAs have several limitations in the way that they
have been used historieally in eeology. Many of the eurrent GA applications ean
be accomplished using other more traditional techniques. Often the tradition al
teehniques don't perform as weIl as aGA (e.g., applieations of linear statistics to
non-linear problems), however, they always beg the question "why didn't you use
teehnique X?". Also, GAs are not very transparent to non-users. While it is easy
to explain the basics of how a GA operates, it still seems like magie to the
D. Morrall
The probabilities of mutation and crossover are selected by the user. The
children replace the least fit individuals in the population. The process continues
either until a specified number of generations have been completed or until an
individual in the population meets the success criteria. Because there is a random
component in the GA, no two runs are the same. The trajectory of population
fitness over time will vary between runs (Figure 3.3).
3.4
Applications of Genetic Aigorithms to Ecological
Modelling
In the field of ecology as in other fields, genetic algorithms have been used for
parameter optimization, equation discovery, and pattern searching. Reynolds and
Ford (1999) developed a modelling approach (Pareto Optimal Model Assessment
Cycle) that allows for simultaneous evaluation of multiple output criteria for a
model with a given parameter set. A genetic algorithm is used to generate optimal
parameter sets (Pareto Optimal Sets) and evaluate fitness in terms of their ability
to predict the model output. Reynolds used this approach to parameterize and
evaluate the ecological theory, model structure, and assessment data of WHORL,
a canopy competition model. Through this dynamic process they were able to
reveal deficiencies in the model structure and criteria to make improvements.
Ludvigan et al (1997) used a GA to search for optimal bacterial phospholipid fatty
acid (PLFA) combinations to biogeochemical parameters. Random combinations
of PLFAs were chosen and evolved using a GA with a partial least squares fitness
function. These combinations of engineering and statistical techniques with
evolutionary algorithms provide a robust approach to data evaluation.
Similar in approach to parameter optimization is the use of GAs to look for data
patterns such as subsurface zones of bacterial activity (Mahinthakumar al. 1999)
and fish distributions (D' Angelo et al. 1995). These studies attempt to find the
optimum combination of variables to predict the distributions of organisms in
relation to their habitat. With the surge in use of geographic information systems
(GIS) to present environmental data on a spatial template, these types of analysis
are common and are expected to become more prevalent. Jeffers (1999) and
StockweIl (1999) provide examples of some species distribution models that use
abundance or presence/absence data to predict distributions of organisms as a
function of spatial habitat characteristics. Fielding (1999) suggests that these
types of exploratory analysis are some of the most promising and least
controversial uses of machine leaming applications.
From a practical standpoint, GAs have several limitations in the way that they
have been used historieally in eeology. Many of the eurrent GA applications ean
be accomplished using other more traditional techniques. Often the tradition al
teehniques don't perform as weIl as aGA (e.g., applieations of linear statistics to
non-linear problems), however, they always beg the question "why didn't you use
teehnique X?". Also, GAs are not very transparent to non-users. While it is easy
to explain the basics of how a GA operates, it still seems like magie to the