100
P. Goethals . A. Dedecker . W. Gabriels . N. De Pauw
In Fig. 6.5. the effect on the Asellidae population of the removal of the 6 weirs
in the Zwalm river basin is simulated by classification trees. If the two upper maps
are compared, one can quickly notice the generalisation the model makes:
according to the model the Asselidae only colonize the broader river sites (the
only rule generated by the J48-model using tenfold cross validation on all sixty
instances is:'width more than 3.5 meters: Asellidae present, while absent in the
more narrow streams').
Althought the amount of correct1y classified instances is rather good and the
induced model gives an interesting generalisation to easily and reliably predict
Asellidae in the field, the model is not interesting to predict the effects of the
removal of the weirs. The maps at the bottom illustrate that the Asellidae will keep
on colonizing the river stretches in sites 1 to 6 after the removal of the six weirs,
according to the J48 model. When a simple rule ('the site in front of the weir will
have the same characteristics as behind the weir before weir removal, while the
situation behind the weir is not altered') is used to predict the Asellidae behaviour,
similar predictions are made. According to ecological experts however, the
Asellidae do not effectively colonize the sites behind the weir as could be thought
according to the measurements. Most probably the presence of the Asselidae in the
sampies behind the weirs has to be explained by accidental carry-away processes
from the sites in front of the weirs, where the conditions (slow current) are
convenient for the Asselidae. This also explains that in the sampies of the river
stretches some kilometres in front of the weirs, Asellidae are never present. So
according to ecological experts, the Asselidae will most probably not colonize the
Zwalm river under undisturbed conditions and also not when the weirs are
removed.
6.3.2
Artificial Neural Networks
6.3.2.1
Model Development and Validation
From Fig. 6.6. one can clearly observe that artificial neural networks (ANNs)
make better predictions compared to simple probabilistic guesses. Another trend,
similar to the predictions with classification trees is that the reliability of the
models is the highest for very common (Chironomidae, Tubificidae) and
extremely rare taxa (Aplexa, Ephemera, Armiger). The added value of the artificial
neural network is the lowest under these circumstances.
P. Goethals . A. Dedecker . W. Gabriels . N. De Pauw
In Fig. 6.5. the effect on the Asellidae population of the removal of the 6 weirs
in the Zwalm river basin is simulated by classification trees. If the two upper maps
are compared, one can quickly notice the generalisation the model makes:
according to the model the Asselidae only colonize the broader river sites (the
only rule generated by the J48-model using tenfold cross validation on all sixty
instances is:'width more than 3.5 meters: Asellidae present, while absent in the
more narrow streams').
Althought the amount of correct1y classified instances is rather good and the
induced model gives an interesting generalisation to easily and reliably predict
Asellidae in the field, the model is not interesting to predict the effects of the
removal of the weirs. The maps at the bottom illustrate that the Asellidae will keep
on colonizing the river stretches in sites 1 to 6 after the removal of the six weirs,
according to the J48 model. When a simple rule ('the site in front of the weir will
have the same characteristics as behind the weir before weir removal, while the
situation behind the weir is not altered') is used to predict the Asellidae behaviour,
similar predictions are made. According to ecological experts however, the
Asellidae do not effectively colonize the sites behind the weir as could be thought
according to the measurements. Most probably the presence of the Asselidae in the
sampies behind the weirs has to be explained by accidental carry-away processes
from the sites in front of the weirs, where the conditions (slow current) are
convenient for the Asselidae. This also explains that in the sampies of the river
stretches some kilometres in front of the weirs, Asellidae are never present. So
according to ecological experts, the Asselidae will most probably not colonize the
Zwalm river under undisturbed conditions and also not when the weirs are
removed.
6.3.2
Artificial Neural Networks
6.3.2.1
Model Development and Validation
From Fig. 6.6. one can clearly observe that artificial neural networks (ANNs)
make better predictions compared to simple probabilistic guesses. Another trend,
similar to the predictions with classification trees is that the reliability of the
models is the highest for very common (Chironomidae, Tubificidae) and
extremely rare taxa (Aplexa, Ephemera, Armiger). The added value of the artificial
neural network is the lowest under these circumstances.