180
H. Hoang . F. Recknagel . J. Marshall . S. Choy
approach was used to identify input variables that exert some influence on outputs
and to predict the ecological consequences of altering input variables by
simulating various scenarios. The latter approach utilised a broader range of input
variables and data of the Queensland stream system including sites that were
affected by anthropogenic impacts.
Sensitivity analyses were conducted for each single ANN model in order to
refine the selection of input variables and strengthen the models' validity.
However the graphical representations of sensitivity results also illustrated the
nature of relationships between environmental variables and the occurrence of
macroinvertebrate taxa. Selected results of the sensitivity analysis from 'clean
water' and 'dirty water' models of the Queensland stream system are documented
in this chapter and findings are discussed in the context of literature knowledge on
stream macroinvertebrates.
9.2
Study Sites
The Queensland river and stream network spreads over the territory of the federal
state of Queensland (Australia). The climate conditions of Queensland range from
high rainfall areas (1600 mmJannum) in the tropical Northeast to low rainfall areas
(200 mmJannum) in the Southeast. Study sites are representative for the
catchments of all major and minor rivers.
9.3
Materials and Methods
9.3.1
Oata
A comprehensive database of the Queensland stream system was used for the
development of the neural network models. The database was divided into 897
datasets of reference sites and 1159 datasets of test sites.
Each site-specific dataset contained 39 physical variables, 17 potentially
impacted environmental variables and colonisation patterns of 40
macroinvertebrates taxa at family level. Different combinations of data were used
for the development of specific models.
H. Hoang . F. Recknagel . J. Marshall . S. Choy
approach was used to identify input variables that exert some influence on outputs
and to predict the ecological consequences of altering input variables by
simulating various scenarios. The latter approach utilised a broader range of input
variables and data of the Queensland stream system including sites that were
affected by anthropogenic impacts.
Sensitivity analyses were conducted for each single ANN model in order to
refine the selection of input variables and strengthen the models' validity.
However the graphical representations of sensitivity results also illustrated the
nature of relationships between environmental variables and the occurrence of
macroinvertebrate taxa. Selected results of the sensitivity analysis from 'clean
water' and 'dirty water' models of the Queensland stream system are documented
in this chapter and findings are discussed in the context of literature knowledge on
stream macroinvertebrates.
9.2
Study Sites
The Queensland river and stream network spreads over the territory of the federal
state of Queensland (Australia). The climate conditions of Queensland range from
high rainfall areas (1600 mmJannum) in the tropical Northeast to low rainfall areas
(200 mmJannum) in the Southeast. Study sites are representative for the
catchments of all major and minor rivers.
9.3
Materials and Methods
9.3.1
Oata
A comprehensive database of the Queensland stream system was used for the
development of the neural network models. The database was divided into 897
datasets of reference sites and 1159 datasets of test sites.
Each site-specific dataset contained 39 physical variables, 17 potentially
impacted environmental variables and colonisation patterns of 40
macroinvertebrates taxa at family level. Different combinations of data were used
for the development of specific models.