Chapter 13 . Cross-Sectional Estimation of Chlorophyll a in Lakes 261
Performances o/multiple regression models:
Similar to artificial neural network models, assessment of the performances of the
developed multiple regression models were made possible with the help of a group
of regression plots (Figures 13.5, 13.6 and 13.7). The linear regression coefficients
were 0.55,0.88 and 0.69 for Keban Dam Reservoir, Mogan Lake and Eymir Lake
respectively. Again, R-values of regression plots were relatively higher for Mogan
and Eymir Lakes as compared to R-value for KDR. However, R-value of multiple
regression plot was significantly lower than R-value of artificial neural network
plot for each water body.
Comparisons 0/ the Performances 0/ Artificial Neural Networks and multiple
regression models:
Examination and comparison of the figures (with the calculated R-values) related
to the regression plots and for neural network model results gives an obvious
impression on the superiority of neural network model for KDR, Mogan Lake and
Eymir Lake. So it was concluded that the neural network model seemed to predict
chlorophyll-a with a better performance than that of the selected multiple
regression models for the examined water bodies of entirely different character.
As a concluding remark it can be stated that, there is a potential in artificial
neural network approach to be used as a modelling tool to estimate major
parameters of eutrophication (i.e. chlorophyll-a) because of its inherent property
of being able to be trained for very complex and non-linear systems. Neural
network can be trained to recognize the environment to predict the system's
response to the conditions of the environment even in highly variable water bodies
with respect to location and time.
13.5.2
Recommendations
Since this study had the main intentions of getting initial information related to i)
the performances of artificial neural networks in estimating chlorophyll-a
concentrations in different water bodies and ii) comparing the performances of
artificial neural networks with that of simple regression methods, sophisticated
statistical evaluation methods were not planned to be employed as comparison
tools at the beginning. However, it is recommended to apply such tools in future to
strengthen the conclusions derived from this particular study. It is further
suggested that the performance of artificial neural networks should be tested
utilizing the relevant data obtained from several other water bodies.
Acknowledgments
Middle East Technical University through research fund AFP-97-03-11-04
supported this study. Data used in the studies were provided by State Hydraulic
Performances o/multiple regression models:
Similar to artificial neural network models, assessment of the performances of the
developed multiple regression models were made possible with the help of a group
of regression plots (Figures 13.5, 13.6 and 13.7). The linear regression coefficients
were 0.55,0.88 and 0.69 for Keban Dam Reservoir, Mogan Lake and Eymir Lake
respectively. Again, R-values of regression plots were relatively higher for Mogan
and Eymir Lakes as compared to R-value for KDR. However, R-value of multiple
regression plot was significantly lower than R-value of artificial neural network
plot for each water body.
Comparisons 0/ the Performances 0/ Artificial Neural Networks and multiple
regression models:
Examination and comparison of the figures (with the calculated R-values) related
to the regression plots and for neural network model results gives an obvious
impression on the superiority of neural network model for KDR, Mogan Lake and
Eymir Lake. So it was concluded that the neural network model seemed to predict
chlorophyll-a with a better performance than that of the selected multiple
regression models for the examined water bodies of entirely different character.
As a concluding remark it can be stated that, there is a potential in artificial
neural network approach to be used as a modelling tool to estimate major
parameters of eutrophication (i.e. chlorophyll-a) because of its inherent property
of being able to be trained for very complex and non-linear systems. Neural
network can be trained to recognize the environment to predict the system's
response to the conditions of the environment even in highly variable water bodies
with respect to location and time.
13.5.2
Recommendations
Since this study had the main intentions of getting initial information related to i)
the performances of artificial neural networks in estimating chlorophyll-a
concentrations in different water bodies and ii) comparing the performances of
artificial neural networks with that of simple regression methods, sophisticated
statistical evaluation methods were not planned to be employed as comparison
tools at the beginning. However, it is recommended to apply such tools in future to
strengthen the conclusions derived from this particular study. It is further
suggested that the performance of artificial neural networks should be tested
utilizing the relevant data obtained from several other water bodies.
Acknowledgments
Middle East Technical University through research fund AFP-97-03-11-04
supported this study. Data used in the studies were provided by State Hydraulic