Adjusting Neural Network to a Particular Problem: Neural Network-Based Empirical Biological Model for Chlorophyll Concentration in the Upper Ocean

Table 2

Comparison of error statistics: bias, RMSE, mean absolute error (MAE), correlation coefficient or cross-correlation (CC), and scatter index (SI) on an independent validation set for three NN ensembles: (i) ensemble of NNs with one output (chl-a), (ii) ensemble of LN-NNs with one output (ln(chl-a)), and (iii) ensemble on LN-NNs with two outputs, ln(chl-a) and ln(). All NNs have 23 inputs and 30 hidden neurons.

		Bias	RMSE	MAE	CC	SI
NN, output: chl-a		-8.3e-3	0.159	0.096	0.608	0.914
LN-NN, output: ln(chl-a)		2.0e-3	0.097	0.052	0.788	0.559
LN-NN, outputs: ln(chl-a) and ln()	chl-a	-4.0e-3	0.090	0.048	0.801	0.520
		3.e-4	0.018	0.009	0.807	0.226