80
F. Recknagel
availability and predation pressure by fish. If we focus lake modelling on the
euphotic zone as the scene of primary and secondary production in lakes, we can
imply that plankton communities are almost homogenously distributed and almost
instantaneously responding to exogenous disturbances. As the predietion and
explanation of instantaneous algal abundance and succession appear to be the
biggest challenge to freshwater ecologists, the AA simulation of the spatial
distribution of individuals as suggested for terrestrial ecosystems (e.g. Schmitz
and Booth 1997) seems no longer relevant. Neither adequate knowledge nor data
would be available to realistically reflect individual or spatial aspects of algae and
zooplankton. By contrast the AA simulation of aquatic ecosystems needs to focus
on the temporal distribution ofplankton populations (respective functional groups)
by means of state variable-based AA embodying evolutionary computation.
5.4.1
Aigal Species Simulation by Adaptive Agents
Adaptive agents simulation of algal species dynamies is currently designed and
developed according to Fig. 5.2. Four agents are considered initially to represent
blue-green algae species typically competing in eutrophie freshwaters in summer:
Microcystis, Oscillatoria, Anabaena and Phormidium. (see Fig. 5.2a). These four
agents interact by competition and are determined by environmental driving forces
such as solar radiation, water temperature, and nutrient loadings.
5.4.1.1
Embodiment 01 Evolutionary Computation in Agents
Each single agent is embodied by artificial Neural Networks (ANN), evolving
differential equations (EOE) or evolving roles (ER) in order to maximise (adapt)
their performance (abundance) in relation to current environmental conditions
(nutrient loadings, light, temperature and abundance of competitors).
Case studies on ANN (Recknagel 1997; Recknagel et al. 1997), EOE
(Whigham and Recknagel 2001) and ER (Bobbin and Recknagel 2001; Recknagel
et al. 2002) have been conducted for the predietion of algal abundance and
succession in lakes and reservoirs. Fig. 5.3. shows simulation results for
Microcystis (a) and Oscillatoria (b) in Lake Kasumigaura predicted by ANN and
ER. The underlying ER used for the same-day predictions in Fig. 5.3. are
documented in Table 5.1.
Examples in Fig. 5.4. are based on 7-days-ahead predictions for chlorophyll-a
and Microcystis in Lake Kasumigaura performed by EOE and ER (see Tab. 5.2.)
and ANN (Recknagel et al. 2002), which were trained and extracted from the Lake
Kasumigaura data base (Takamura et al. 1992). The underlying OE was adopted
from the deterministie lake model SALMO (Recknagel and Benndorf 1982).
Ouring the AA simulation of algal dynamics in a specific lake, each agent
adapts steadily to occurring environmental conditions by producing the best
adapted model or "offspring" agent based on its evaluation and selection of mates,
recombination strategy and mutation strategy (see Fig. 5.2b). This will be
F. Recknagel
availability and predation pressure by fish. If we focus lake modelling on the
euphotic zone as the scene of primary and secondary production in lakes, we can
imply that plankton communities are almost homogenously distributed and almost
instantaneously responding to exogenous disturbances. As the predietion and
explanation of instantaneous algal abundance and succession appear to be the
biggest challenge to freshwater ecologists, the AA simulation of the spatial
distribution of individuals as suggested for terrestrial ecosystems (e.g. Schmitz
and Booth 1997) seems no longer relevant. Neither adequate knowledge nor data
would be available to realistically reflect individual or spatial aspects of algae and
zooplankton. By contrast the AA simulation of aquatic ecosystems needs to focus
on the temporal distribution ofplankton populations (respective functional groups)
by means of state variable-based AA embodying evolutionary computation.
5.4.1
Aigal Species Simulation by Adaptive Agents
Adaptive agents simulation of algal species dynamies is currently designed and
developed according to Fig. 5.2. Four agents are considered initially to represent
blue-green algae species typically competing in eutrophie freshwaters in summer:
Microcystis, Oscillatoria, Anabaena and Phormidium. (see Fig. 5.2a). These four
agents interact by competition and are determined by environmental driving forces
such as solar radiation, water temperature, and nutrient loadings.
5.4.1.1
Embodiment 01 Evolutionary Computation in Agents
Each single agent is embodied by artificial Neural Networks (ANN), evolving
differential equations (EOE) or evolving roles (ER) in order to maximise (adapt)
their performance (abundance) in relation to current environmental conditions
(nutrient loadings, light, temperature and abundance of competitors).
Case studies on ANN (Recknagel 1997; Recknagel et al. 1997), EOE
(Whigham and Recknagel 2001) and ER (Bobbin and Recknagel 2001; Recknagel
et al. 2002) have been conducted for the predietion of algal abundance and
succession in lakes and reservoirs. Fig. 5.3. shows simulation results for
Microcystis (a) and Oscillatoria (b) in Lake Kasumigaura predicted by ANN and
ER. The underlying ER used for the same-day predictions in Fig. 5.3. are
documented in Table 5.1.
Examples in Fig. 5.4. are based on 7-days-ahead predictions for chlorophyll-a
and Microcystis in Lake Kasumigaura performed by EOE and ER (see Tab. 5.2.)
and ANN (Recknagel et al. 2002), which were trained and extracted from the Lake
Kasumigaura data base (Takamura et al. 1992). The underlying OE was adopted
from the deterministie lake model SALMO (Recknagel and Benndorf 1982).
Ouring the AA simulation of algal dynamics in a specific lake, each agent
adapts steadily to occurring environmental conditions by producing the best
adapted model or "offspring" agent based on its evaluation and selection of mates,
recombination strategy and mutation strategy (see Fig. 5.2b). This will be