Chapter 10 . Aigal Species Succession in Rivers
10.3.3
Neural Network Validation and Knowledge Discovery on Aigal
Succession
201
Model validation was based on visual comparisons between observed and
simulated output values. With the best performing RNN, two types of sensitivity
analyses as described in Jeong et al. (2001a) were implemented: 'Most Influencing
Parameter (MIP)' and 'Sensitivity on Wide-ranged Disturbance (SWD)'.
The network was disturbed by ±1 to 2 SD for the sensitivity analyses.
According to Zar (1984), ±1 SD represents commonly occurring variation, and ±2
SD covers about 95% of total data variation. The sensitivity analysis with ±1 SD
can explain general conditions, while disturbance of ±2 SD may suggest specific
and infrequent interactions between algal species and input variables. The results
of sensitivity analyses were interpreted compared with known ecologieal
information. All RNN models were developed by means of the neural network
shell NeuroSolutions 3.0 (NeuroDimension, 1999).
10.4
Results and Discussion
10.4.1
Limnological Aspects and Plankton Dynamics in the Lower Nakdong
River
Time series data from the lower Nakdong River indicate hypertrophie conditions
and distinct annual and seasonal variability (Table 10.2). Due to the increased
flow regulation, the ecosystem has been modified to become a river-reservoir
hybrid (100 et al. 1997). Construction of an estuarine barrage in conjugation with a
water intake has increased water retention time and acce1erated eutrophication in
the lower 50 km of the river. Depending on the total amount of rainfall during the
summer, the river exhibits distinctive seasonal characteristics. Rainfall patterns
during the summer monsoon and typhoon events drive changes in physicalchemical parameters in the lower Nakdong River (Park 1998; Lee et al. 1999).
Rotifera dominated the zooplankton community of the river from 1994 to 1998,
while Cladocera and Copopoda were much less abundant. Zooplankton
populations did not display significant inter-annual variation. However, there were
"dear water phases" in early spring and autumn, as earlier documented and
attributed to macrozooplankton grazing of phytoplankton (Kim et al. 1998; Kim et
al. 2001).
10.3.3
Neural Network Validation and Knowledge Discovery on Aigal
Succession
201
Model validation was based on visual comparisons between observed and
simulated output values. With the best performing RNN, two types of sensitivity
analyses as described in Jeong et al. (2001a) were implemented: 'Most Influencing
Parameter (MIP)' and 'Sensitivity on Wide-ranged Disturbance (SWD)'.
The network was disturbed by ±1 to 2 SD for the sensitivity analyses.
According to Zar (1984), ±1 SD represents commonly occurring variation, and ±2
SD covers about 95% of total data variation. The sensitivity analysis with ±1 SD
can explain general conditions, while disturbance of ±2 SD may suggest specific
and infrequent interactions between algal species and input variables. The results
of sensitivity analyses were interpreted compared with known ecologieal
information. All RNN models were developed by means of the neural network
shell NeuroSolutions 3.0 (NeuroDimension, 1999).
10.4
Results and Discussion
10.4.1
Limnological Aspects and Plankton Dynamics in the Lower Nakdong
River
Time series data from the lower Nakdong River indicate hypertrophie conditions
and distinct annual and seasonal variability (Table 10.2). Due to the increased
flow regulation, the ecosystem has been modified to become a river-reservoir
hybrid (100 et al. 1997). Construction of an estuarine barrage in conjugation with a
water intake has increased water retention time and acce1erated eutrophication in
the lower 50 km of the river. Depending on the total amount of rainfall during the
summer, the river exhibits distinctive seasonal characteristics. Rainfall patterns
during the summer monsoon and typhoon events drive changes in physicalchemical parameters in the lower Nakdong River (Park 1998; Lee et al. 1999).
Rotifera dominated the zooplankton community of the river from 1994 to 1998,
while Cladocera and Copopoda were much less abundant. Zooplankton
populations did not display significant inter-annual variation. However, there were
"dear water phases" in early spring and autumn, as earlier documented and
attributed to macrozooplankton grazing of phytoplankton (Kim et al. 1998; Kim et
al. 2001).