Chapter 13 . Cross-Sectional Estimation of Chlorophyll a in Lakes 251
13.1.3
The Use of Artificial Neural Networks in Environmental Modelling
There are numerous examples of the use of neural networks in environmental
modelling. Moreau et al. (1999) embedded neural networks in Lotka -Volterra
predator-prey models. Brion and Lingireddy (1997) used neural networks in
identification of the sources of microbial contamination. Zhang and Stanley
(1997) adopted neural networks for water demand forecasting. Robertson and
Morison (1999) attempted to estimate the age of fish automatically with a neural
network algorithm that proved successful at least for some fish species. The use of
neural network algorithms in modelling and analysis of eutrophication in lakes is
also quite promising because of the complex nature of the problem.
Several studies have been carried out on the use of neural networks in
eutrophication modelling and Lake Management in recent years. Scardi (1996)
used neural networks as to estimate phytoplankton production, Recknagel et al.
(1997), Recknagel (1997) and Yabunaka et al. (1997) predicted chlorophyll-a
concentration and algal species abundance as a function of sampled water quality
parameters. Some zooplankton, such as Rotifers and Diaphanosoma sp., were
added as variables to simulate the predator grazing in these studies. Karul et al.
(1998a) developed an input-output model where measured water quality
parameters were used to estimate chlorophyll-a concentrations. Keiner and Brown
(1998) used neural networks to estimate chlorophyll-a concentrations at the ocean
surface as an alternative to linear regression methods. Results of all of the abovecited works achieved satisfactory levels of precision. However, there is
insufficient information to compare the effectiveness of neural network
approaches against the use of multiple regression methods. One notable exception
(Karul et al. 1999b) was based on only a single water body.
The main objective of this study was to develop neural network models for
different water bodies to simulate eutrophication process. It was thought that the
neural network based models that are adequately trained with several
environmental factors could be a better approach with more precise predictions
than multiple regression methods. A further objective of this study was to compare
the performance of the neural network models with that of multiple linear
regression models.
13.2
Oata and Lakes
Data collected from three very different Turkish water bodies, the Keban Dam
Reservoir (KDR), Mogan and Eymir Lakes, have been utilized in this study. Table
13.1 summarizes the basic properties of these water bodies in a comparative way.
13.1.3
The Use of Artificial Neural Networks in Environmental Modelling
There are numerous examples of the use of neural networks in environmental
modelling. Moreau et al. (1999) embedded neural networks in Lotka -Volterra
predator-prey models. Brion and Lingireddy (1997) used neural networks in
identification of the sources of microbial contamination. Zhang and Stanley
(1997) adopted neural networks for water demand forecasting. Robertson and
Morison (1999) attempted to estimate the age of fish automatically with a neural
network algorithm that proved successful at least for some fish species. The use of
neural network algorithms in modelling and analysis of eutrophication in lakes is
also quite promising because of the complex nature of the problem.
Several studies have been carried out on the use of neural networks in
eutrophication modelling and Lake Management in recent years. Scardi (1996)
used neural networks as to estimate phytoplankton production, Recknagel et al.
(1997), Recknagel (1997) and Yabunaka et al. (1997) predicted chlorophyll-a
concentration and algal species abundance as a function of sampled water quality
parameters. Some zooplankton, such as Rotifers and Diaphanosoma sp., were
added as variables to simulate the predator grazing in these studies. Karul et al.
(1998a) developed an input-output model where measured water quality
parameters were used to estimate chlorophyll-a concentrations. Keiner and Brown
(1998) used neural networks to estimate chlorophyll-a concentrations at the ocean
surface as an alternative to linear regression methods. Results of all of the abovecited works achieved satisfactory levels of precision. However, there is
insufficient information to compare the effectiveness of neural network
approaches against the use of multiple regression methods. One notable exception
(Karul et al. 1999b) was based on only a single water body.
The main objective of this study was to develop neural network models for
different water bodies to simulate eutrophication process. It was thought that the
neural network based models that are adequately trained with several
environmental factors could be a better approach with more precise predictions
than multiple regression methods. A further objective of this study was to compare
the performance of the neural network models with that of multiple linear
regression models.
13.2
Oata and Lakes
Data collected from three very different Turkish water bodies, the Keban Dam
Reservoir (KDR), Mogan and Eymir Lakes, have been utilized in this study. Table
13.1 summarizes the basic properties of these water bodies in a comparative way.