Chapter 13 . Cross-Sectional Estimation of Chlorophy" a in Lakes 257
13.3.2
Multiple Regression Modelling Approach
For each neural network model, a corresponding multiple linear regression model
was developed. The aim of using exactly the same data values was to compare the
performances of two methods (neural network and multiple linear regression)
under exactly the same conditions. To achieve this, the available data was divided
into two equal batches, one of which is used to train the neural network while the
entire data is used to calculate the regression coefficients. Exactly the same data
batch was used to calculate the multiple linear regression models and the
regression coefficients were calculated using the entire data. MatLab of
Mathworks Co. (Demuth and Beale 1998) was used for all calculations.
13.4
Results
The Figures 13.2, 13.3 and 13.4 give the measured data against predictions
utilizing the Neural Network Model for KDR, Eymir and Mogan Lakes
respectively. Similarly Figures 13.5, 13.6 and 13.7 presents the measured data
against predictions utilizing the Multiple Regression Models for these water
bodies. Same data set of each water body was used for both methods.
The multiple regression models are given below:
ForKDR:
chl_a
keban
= 1 0-3.2506.aJk + 0.0067.EC - 0.0016.00 + 0.983N03 -0.0485.pH -0.1331.1'04 + 5.3673.Se<:chi - 0.0323.SS + 0.007. Temp + 0.348
(13.7)
For Mogan lake:
chI a
= 1 0-9.57.NH3 + 1.01.N03 + 0.2831.TP- 0.6682.E.C. + O.oo88.pH -0 265.Secchi - 0.3215.8S+ 0.D248.Temp-0.4849. Turb -0.0919
-
mogan
(13.8)
For Eymir Lake:
chI a = 1 O-O·3455.NH3 + o 4414.N03 - 0.0807.TP + 0.1388.E.C. + 0.0014.pH - 0.1691.Secchi - 2.667.SS + O.OOO9.Temp - 0.0737. TurbO.lX)25 eynur
(13.9)
where chl-a is the chlorophyll-a concentration, mg/I; alk is the alkalinity, mg/I
as CaC0 3 ; EC is the electricai conductivitY,Jlmho; DO is the dissolved oxygen
concentration, mg/i; N0 3 is the nitrate concentration, mg/i; P0 4 is the phosphate
concentration, mg/I; Secchi is the Secchi depth, m ; SS is the suspended so lids
13.3.2
Multiple Regression Modelling Approach
For each neural network model, a corresponding multiple linear regression model
was developed. The aim of using exactly the same data values was to compare the
performances of two methods (neural network and multiple linear regression)
under exactly the same conditions. To achieve this, the available data was divided
into two equal batches, one of which is used to train the neural network while the
entire data is used to calculate the regression coefficients. Exactly the same data
batch was used to calculate the multiple linear regression models and the
regression coefficients were calculated using the entire data. MatLab of
Mathworks Co. (Demuth and Beale 1998) was used for all calculations.
13.4
Results
The Figures 13.2, 13.3 and 13.4 give the measured data against predictions
utilizing the Neural Network Model for KDR, Eymir and Mogan Lakes
respectively. Similarly Figures 13.5, 13.6 and 13.7 presents the measured data
against predictions utilizing the Multiple Regression Models for these water
bodies. Same data set of each water body was used for both methods.
The multiple regression models are given below:
ForKDR:
chl_a
keban
= 1 0-3.2506.aJk + 0.0067.EC - 0.0016.00 + 0.983N03 -0.0485.pH -0.1331.1'04 + 5.3673.Se<:chi - 0.0323.SS + 0.007. Temp + 0.348
(13.7)
For Mogan lake:
chI a
= 1 0-9.57.NH3 + 1.01.N03 + 0.2831.TP- 0.6682.E.C. + O.oo88.pH -0 265.Secchi - 0.3215.8S+ 0.D248.Temp-0.4849. Turb -0.0919
-
mogan
(13.8)
For Eymir Lake:
chI a = 1 O-O·3455.NH3 + o 4414.N03 - 0.0807.TP + 0.1388.E.C. + 0.0014.pH - 0.1691.Secchi - 2.667.SS + O.OOO9.Temp - 0.0737. TurbO.lX)25 eynur
(13.9)
where chl-a is the chlorophyll-a concentration, mg/I; alk is the alkalinity, mg/I
as CaC0 3 ; EC is the electricai conductivitY,Jlmho; DO is the dissolved oxygen
concentration, mg/i; N0 3 is the nitrate concentration, mg/i; P0 4 is the phosphate
concentration, mg/I; Secchi is the Secchi depth, m ; SS is the suspended so lids