Chapter 13
A Comparison between Neural Network Based
and Multiple Regression Models for Chlorophyll-a
Estimation
C. Karul . S. Soyupak
13.1
Introduction
13.1.1
Eutrophication in Water Bodies and Relevant Models
Eutrophication and associated algal blooms are serious problems in many lakes
and reservoirs. Deterioration of water quality for human consumption, limitation
of recreational use, depletion of dissolved oxygen levels below tolerable levels for
certain fish species and severe ecosystem degradation are amongst the adverse
effects of eutrophication (Ryding and Rast 1989).
Various types of simulation models have been developed to predict the
magnitude and timing of algal blooms. One major class of models for predicting
water quality parameters (i.e. algal concentrations) can be named "as empirical
water body models". They were primarily developed as extensions of the
phosphorus model. Models developed by Dillon and Rigler (1974); Rast and Lee
(1978); and Bartsch and Gakstatter (1978) are based on a fit of a log-log plots of
Chlorophyll-a and P. They are the typical examples of such models. Later Smith
and Shapiro (1981) have presented a modified correlation that takes into account
the potential nitrogen limitation. Specifically, some regression models predict the
algal concentrations as a function of flushing corrected average annual phosphorus
inflow concentrations as summarized by Ryding and Rast (1989). Some empirical
statistical models developed for 233 Florida lakes predict logarithm of
chlorophyll-a concentrations as functions of "logarithm of total phosphorus
concentrations" and "logarithm of total nitrogen concentrations"(Canfield et al.
1983; 1984). The authors has stressed the applicability of simple models with very
few parameters for predicting chlorophyll-a concentrations and they further
A Comparison between Neural Network Based
and Multiple Regression Models for Chlorophyll-a
Estimation
C. Karul . S. Soyupak
13.1
Introduction
13.1.1
Eutrophication in Water Bodies and Relevant Models
Eutrophication and associated algal blooms are serious problems in many lakes
and reservoirs. Deterioration of water quality for human consumption, limitation
of recreational use, depletion of dissolved oxygen levels below tolerable levels for
certain fish species and severe ecosystem degradation are amongst the adverse
effects of eutrophication (Ryding and Rast 1989).
Various types of simulation models have been developed to predict the
magnitude and timing of algal blooms. One major class of models for predicting
water quality parameters (i.e. algal concentrations) can be named "as empirical
water body models". They were primarily developed as extensions of the
phosphorus model. Models developed by Dillon and Rigler (1974); Rast and Lee
(1978); and Bartsch and Gakstatter (1978) are based on a fit of a log-log plots of
Chlorophyll-a and P. They are the typical examples of such models. Later Smith
and Shapiro (1981) have presented a modified correlation that takes into account
the potential nitrogen limitation. Specifically, some regression models predict the
algal concentrations as a function of flushing corrected average annual phosphorus
inflow concentrations as summarized by Ryding and Rast (1989). Some empirical
statistical models developed for 233 Florida lakes predict logarithm of
chlorophyll-a concentrations as functions of "logarithm of total phosphorus
concentrations" and "logarithm of total nitrogen concentrations"(Canfield et al.
1983; 1984). The authors has stressed the applicability of simple models with very
few parameters for predicting chlorophyll-a concentrations and they further